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.pcbi.1002867 | Passive Dendrites Enable Single Neurons to Compute Linearly Non-separable Functions | Local supra-linear summation of excitatory inputs occurring in pyramidal cell dendrites, the so-called dendritic spikes, results in independent spiking dendritic sub-units, which turn pyramidal neurons into two-layer neural networks capable of computing linearly non-separable functions, such as the exclusive OR. Other neuron classes, such as interneurons, may possess only a few independent dendritic sub-units, or only passive dendrites where input summation is purely sub-linear, and where dendritic sub-units are only saturating. To determine if such neurons can also compute linearly non-separable functions, we enumerate, for a given parameter range, the Boolean functions implementable by a binary neuron model with a linear sub-unit and either a single spiking or a saturating dendritic sub-unit. We then analytically generalize these numerical results to an arbitrary number of non-linear sub-units. First, we show that a single non-linear dendritic sub-unit, in addition to the somatic non-linearity, is sufficient to compute linearly non-separable functions. Second, we analytically prove that, with a sufficient number of saturating dendritic sub-units, a neuron can compute all functions computable with purely excitatory inputs. Third, we show that these linearly non-separable functions can be implemented with at least two strategies: one where a dendritic sub-unit is sufficient to trigger a somatic spike; another where somatic spiking requires the cooperation of multiple dendritic sub-units. We formally prove that implementing the latter architecture is possible with both types of dendritic sub-units whereas the former is only possible with spiking dendrites. Finally, we show how linearly non-separable functions can be computed by a generic two-compartment biophysical model and a realistic neuron model of the cerebellar stellate cell interneuron. Taken together our results demonstrate that passive dendrites are sufficient to enable neurons to compute linearly non-separable functions.
| Classical views on single neuron computation treat dendrites as mere collectors of inputs, that is forwarded to the soma for linear summation and causes a spike output if it is sufficiently large. Such a single neuron model can only compute linearly separable input-output functions, representing a small fraction of all possible functions. Recent experimental findings show that in certain pyramidal cells excitatory inputs can be supra-linearly integrated within a dendritic branch, turning this branch into a spiking dendritic sub-unit. Neurons containing many of these dendritic sub-units can compute both linearly separable and linearly non-separable functions. Nevertheless, other neuron types have dendrites which do not spike because the required voltage gated channels are absent. However, these dendrites sub-linearly sum excitatory inputs turning branches into saturating sub-units. We wanted to test if this last type of non-linear summation is sufficient for a single neuron to compute linearly non-separable functions. Using a combination of Boolean algebra and biophysical modeling, we show that a neuron with a single non-linear dendritic sub-unit whether spiking or saturating is able to compute linearly non-separable functions. Thus, in principle, any neuron with a dendritic tree, even passive, can compute linearly non-separable functions.
| Seminal neuron models, like the McCulloch & Pitts unit [1] or point neurons (see [2] for an overview), assume that synaptic integration is linear. Despite being pervasive mental models of single neuron computation, and frequently used in network models, the linearity assumption has long been known to be false. Measurements using evoked excitatory post-synaptic potentials (EPSPs) have shown that the summation of excitatory inputs can be supra-linear or sub-linear [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], and can summate in quasi-independent regions of dendrite [13].
Supra-linear summation, the dendritic spikes, has been described for a variety of active dendritic mechanisms. For this type of local summation the measured EPSP peak is first above then below the expected arithmetic sum of EPSPs as shown on Figure 1A. Synapse-driven membrane potential depolarization can open [3], [4], [5], [6], or NMDA receptor [6], [7], [8], [9] channels sufficiently to amplify the initial depolarization, and evoke a dendritic spike.
Contrary to the supra-linear summation of dendritic spikes, a saturating sub-linear summation can arise from passive properties of the dendrite [10], [11], [12]. For this type of local summation the measured EPSP peak is always below the expected arithmetic sum of all EPSPs as shown on Figure 1B. Rall's theoretical work [14], [15], subsequently confirmed experimentally [12], showed that passive sub-linear summation of overlapping inputs is a straightforward consequence of the classic model for conductance-driven current injection into the membrane (where , , and are respectively the time varying current, the synaptic conductance, and the membrane voltage, and where is the equilibrium voltage of the channel).
Dendritic spikes inevitably alter the potential range of single neuron computation. Prior theoretical studies found that dendrites could be divided up into multiple, independent sub-units of integration [16], [17], [18], [19], [20] with sigmoidal or Heaviside activation functions (as shown on Figure 1C). They argued that these dendritic spikes turn synaptic integration into a two stage process: first, synaptic inputs are summed in independent sub-units; second, the output of these sub-units is linearly summed at the soma. Such a two-stage architecture makes the neuron computationally equivalent to a two-layer artificial neural network, greatly expanding a neuron's computational capacities. It has been shown that spiking dendritic sub-units can enhance the feature storage capacity [18], the generalization capacity [17], [21], the computation of binocular disparity [22], the direction selectivity [23], [24], the creation of multiple place fields [25] or the computation of object-feature binding problems [26]. These enhancements may be explained by the ability of a neuron with a sufficient number of spiking dendritic sub-units to compute linearly non-separable functions whereas seminal neuron models like McCulloch & Pitts cannot [27].
These prior studies made two assumptions that may not generalize to all neurons. Firstly, they supposed that the number of independent dendritic sub-units is potentially large; however, for different dendritic morphologies this number may be greatly reduced due to electrotonic coupling or compactness [28], [29], [30]. Secondly, dendritic spikes may not be present in all neuron types, because they lack the specific voltage-gated channels or because the active channel types act to balance each other [10], [11], [31], [32]. Consequently these neuron types could only support a saturating form of non-linear integration. The cerebellar stellate cell is an interesting example because it contradicts both assumptions: it is electrically compact, resulting in a modest number of independent dendritic non-linear sub-units, perhaps on the order of 10 sub-units, as has also been estimated for retinal ganglion cells [33]; and its dendrites are passive, with linear integration of inputs in the peri-somatic region and strictly sub-linear integration in the distal dendritic region [10].
If non-linear computation by dendrites were possible for small numbers of sub-units and for passive dendrites, then this would show that enabling linearly non-separable computation by single neurons is, in principle, a general property of dendrites. We thus set out to answer three key questions: (1) whether a single non-linear dendritic sub-unit is sufficient to enable a neuron to compute linearly non-separable functions, so that multiple sub-units are not a necessary requirement; (2) whether saturating dendritic sub-units, and not just a spiking dendritic sub-units, are sufficient to enable a neuron to compute linearly non-separable functions; and (3) if so, whether the saturating and spiking non-linearities increase computational capacity in the same way.
To answer these questions, we have used a binary neuron model that accounts for non-linear dendritic integration, using either spiking (Figure 1C) or saturating (Figure 1D) activation functions. Using a binary model (Figure 1E) allowed us to study the quantitative increase and qualitative changes in computational capacity using Boolean algebra [34]. A Boolean function is defined by a set of input variables, each taking the value 0 or 1, and a target output value of 0 or 1 for each -dimensional vector that can be made by all possible combinations of values of the input variables (see Material and Methods Boolean Algebra for formal definition). Figure 1F illustrates three well-known examples of Boolean functions, each a function of input variables: AND, NAND, and XOR. The set of Boolean functions computable by a binary neuron model provides a lower bound on the realm of potentially computable algebraic functions by a neuron. Thus, specifying capacity in terms of Boolean functions lets us list the boundaries on a neuron's accessible set of all computable functions.
Using this model, we proceeded on two fronts: first, we used numerical analysis to test if and how much an additional non-linear dendritic sub-unit enables a neuron to compute linearly non-separable functions; second, we used formal analytical proofs to show that the numerical results generalise to an arbitrary number of non-linear sub-units. We found numerically that adding a single non-linear dendritic sub-unit, either a spiking or a saturating unit, allows the neuron to compute some positive linearly non-separable Boolean functions. Analytically, we showed that provided a sufficient number of either spiking or saturating dendritic sub-units a neuron is capable of computing all positive linearly non-separable Boolean function.
Second, our numerical analysis showed that a neuron could compute a function using two distinct implementation strategies: a local strategy where each dendritic sub-unit can trigger a somatic spike, implying that the maximal responses of a dendritic sub-unit always correspond to a somatic spike; and a global strategy where a somatic spike requires the activation of multiple dendritic sub-units, implying that the maximal response of a dendritic sub-unit may not correspond to a somatic spike. This last result may explain why neurons in layer 2/3 of the visual cortex can stay silent when a calcium response from a dendritic sub-unit is maximal [35]. Analytically, we prove that a neuron with spiking dendritic sub-units can use both strategies to compute a function, whereas a neuron with saturating dendritic sub-units can use only a global strategy to compute a function.
Finally, we show how examples of linearly non-separable functions can be implemented in a reduced, generic biophysical model with either a saturating or a spiking dendrites. Moreover, we show that with electrically compact and passive dendrites, a realistic biophysical model of the cerebellar stellate cell can compute a linearly non-separable function. In conclusion, our study thus extends prior work [16], [19], [20] to show that even a compact neuron with passive dendrites can compute linearly non-separable functions.
We present in this section the binary neuron models we used to address our questions with a numerical and a formal analysis. We considered two types of dendritic non-linearities modeled by two families of non-linear activation functions , (Figure 1B–C and definition in Materials and Methods spiking and saturating dendritic activation functions). The first family, , modeled dendritic spikes as observed in [6], [13], [4]. The second family, , modeled dendritic saturation as observed in [12], [10], [11]. Both are parameterized by two non-negative parameters for threshold and maximum output .
For our numerical analysis, based on large parameter searches, we added the output of either of these activation functions to a strictly linear sub-unit integrating the same inputs:(1)where is a binary input vector of length , and are non-negative integer-valued weight vectors, and the somatic activation function gives if the result of synaptic integration is above and otherwise. (Note that if is a linear function () then the previous equation can be rewritten as a single linear weighted sum corresponding to the seminal linear neuron model known as the McCulloch & Pitts unit [1]). Poirazi et al [18] have already established that a single non-linear dendritic sub-unit on its own is not sufficient to increase a neuron model's computation capacity. We thus added a non-linear dendritic sub-unit to the somatic non-linearity precisely to assess the impact of adding either a spiking dendritic non-linearity () or a saturating dendritic non-linearity () on single neuron computation. As a corollary, this model includes neuron classes that have a peri-somatic and a distal dendritic region of integration, like cerebellar stellate cell interneurons [10] and layer 2/3 pyramidal neurons [36].
For our formal analysis, which are the three Propositions presented in the Results, we used the generic two-stage neuron model with dendritic sub-units, analogous to [18]:(2)For both our numerical and formal analysis, the neuron input-output mapping is defined as a Boolean function, and each parameter set produces a unique Boolean function . Here we focused on the effect of non-linear EPSP summation, and thus used only non-negative weight vectors. Consequently, an increase in input could only increase (or not change) the output , never decrease it; therefore we were studying the neuron's ability to compute positive Boolean functions (see Material and Methods Boolean Algebra for formal definition and Lemma 1 for proof).
In terms of neuron physiology, these binary models are quite general. One can interpret the binary input vector across multiple scales, from the pattern of active and inactive individual pre-synaptic neurons up to the set of active and inactive pre-synaptic cell assemblies afferent to the neuron. In this perspective, the weight represents the peak EPSP magnitude produced when a pre-synaptic neuron or a pre-synaptic cell assembly is active. Similarly the Boolean output of 1 could represent a single spike, a burst of spikes, or a change in rate – whatever it is that is read out by the downstream neurons. For instance, in our biophysical model, a single binary variable corresponds to the synchronous activity of a 100 pre-synaptic neurons, and will show that binary models can lead to informative results in situations where the actual number of pre-synaptic input neurons is in the range consistent with existing data.
We first sought answers to two questions: (1) whether a single non-linear dendritic sub-unit is sufficient to enable a neuron to compute linearly non-separable functions, so that multiple sub-units are not a necessary requirement; (2) whether saturating dendritic sub-units, instead of spiking dendritic sub-units, are sufficient to enable a neuron to compute linearly non-separable functions. For a given size of the input vector , we numerically enumerated the sets of positive Boolean functions computable by the different models (, , or ), by searching through their free parameters: the dendritic sub-unit activation function parameters , the weight vectors , and the neuron output threshold . For each set of parameter values, we computed the corresponding Boolean function for that neuron model. This numerical analysis enabled us to determine the computational capacity for each neuron model (, , and ) as the number of Boolean functions these models can compute.
To determine the computational capacity, we counted only the computable representative Boolean functions (see Materials and Methods Boolean Algebra for formal definition). Moreover, we controlled the parameter searches using two analytically known sizes of Boolean function sets: first, the size of the set of all representative positive Boolean functions [34], [37], known for a number of binary variables up to 6; second, within this set of functions, the number of linearly separable representative Boolean functions [37]. This last number corresponded to the exact computational capacity for the purely linear model (). Therefore by comparing to these two known sizes we could see if the model including a dendritic non-linearity ( or ) enabled computation of linearly non-separable functions and, if so, what proportion of those functions could be accessed. The relationship between these sets of functions and the set that can be accessed by a model including a non-linear dendritic sub-unit is illustrated schematically in Figure 2B.
Having established that saturating dendritic sub-units also enhance the computational capacity of a neuron, we then sought to answer the third question: do the saturating and spiking non-linearities increase computational capacity in the same way? Our aim was to understand how the implementation of linearly non-separable Boolean functions depends on the type of dendritic non-linearity. For concreteness, we numerically assessed the implementation of all three linearly non-separable Boolean functions defined for inputs; the functions are presented in Table 2. Interestingly, these three functions map onto known computational problems in neuroscience.
To help identify these functions we named them in reference to [26]: the Feature Binding Problem (FBP), the dual Feature Binding Problem (dFBP), and the partial Feature Binding Problem (pFBP). In the FBP, the function gives an output only when disjoint sets of inputs are active ( or ) and not when any other mixture is active, and is hence analogous to the problem of responding only to combinations of sensory features that uniquely identify objects [26]; conversely, the dFBP is the dual of the FBP (for a definition of duality see [34]); finally the pFBP is another form of the feature binding problem where the neuron fires for objects with overlapping features. We return to the general implications of identifying these binding problem functions in the Discussion.
Propositions 1 and 2 demonstrate that a Boolean function can be implemented using at least two different methods. These methods are either based on the disjunctive (DNF) or the conjunctive (CNF) normal form expressions. The DNF expression decomposes a Boolean function into a disjunction of terms: using this method each dendritic sub-unit computes a term, and an input vector elicits a somatic spike when this input vector activates at least one dendritic sub-unit. The CNF expression decomposes the Boolean function into a conjunction of clauses: using this method each dendritic sub-unit computes a clause, and an input vector elicits a somatic spike when this input vector activates all dendritic sub-units at the same time.
The DNF and CNF expressions provide rigorous methods for implementing Boolean functions where dendritic sub-units respectively correspond to terms or clauses. They are also respective examples of two distinct classes of implementation strategies: local strategies in which the activation of a single dendritic sub-unit can make the neuron spike for at least one input vector (e.g. the DNF-based method); and global strategies in the which a single dendritic sub-unit is insufficient to make the neuron spike for any input vector (e.g. the CNF-based method). Local and global correspond to two distinct categories of implementation method, the DNF-based or CNF-based methods are two extreme examples of these categories. Notably, when there is more than one dendritic sub-unit, we can distinguish multiple forms of local strategy according to how many of the N sub-units are able individually to make the neuron spike at least once (see Figure S4 in Text S1).
Our numerical analyses provide further examples of how this local vs global distinction can be observed even with only one non-linear dendritic sub-unit.
The preceding work used the framework of binary neuron models to establish the qualitative and quantitative gain in computational capacity obtained through the addition of a single dendritic non-linearity. This framework enabled us to study in detail how a linearly non-separable function can be implemented. Furthermore, this Boolean framework also enabled us to formally prove the results (Proposition 1,2,3) extracted from our large parameter searches (Figure 2). To demonstrate that this approach also gave us meaningful insights into biological single neuron computations, we implemented linearly non-separable Boolean functions in reduced and in realistic biophysical models; specifically, we showed that both the saturating and spiking forms of nonlinear dendritic summation enabled a neuron to implement linearly non-separable function in a more realistic framework.
Our study addressed the implication of non-linear dendritic summation of inputs for single neuron computation (for reviews see [42], [43], [44]). Previous theoretical studies which addressed this question assumed that two conditions may be necessary: (1) dendritic spikes [16] and (2) the existence of multiple non-linear sub-units [18]. These two conditions are true only for neurons that possess the right balance of voltage-gated channels in their dendrites to produce dendritic spikes (e.g. see [8]), and spatially extended dendritic trees to ensure sufficient electrotonic distance between dendritic sections (e.g. see [13]). In this paper we investigated whether these two conditions, dendritic spikes and a large number of independent sub-units, were necessary for neurons to compute linearly non-separable Boolean functions.
We have shown that a single non-linear dendritic sub-unit, spiking or not, is sufficient to increase a neuron's computational capacity. Using large parameter searches, we showed that a spiking or a saturating dendritic sub-unit enables a neuron to compute linearly non-separable functions. Moreover, we analytically proved that with a sufficient number of saturating dendritic sub-units a neuron could compute all positive Boolean functions (Proposition 2). To our knowledge, this is the first demonstration that passive dendrites can expand the number of Boolean function computable by a single neuron.
We have shown that a neuron with multiple dendritic sub-units can implement a Boolean function using either a DNF/local strategy, where each dendritic sub-unit can independently trigger the post-synaptic spike response, or a CNF/global strategy, where dendritic sub-units must cooperate to trigger a post-synaptic spike response. In the latter implementation strategy dendritic tuning does not imply the neuron tuning: a given stimulus can elicit the maximal response for a dendritic branch and yet elicit no response for the neuron, as observed in [35]. Moreover, we showed that a neuron with saturating dendritic sub-units cannot implement Boolean functions using a DNF/local strategy.
Finally, we used reduced and realistic biophysical neuron models to provide a proof-of-concept. We showed that a realistic model of cerebellar stellate cell interneuron can implement a linearly non-separable Boolean function.
Our study has established three sufficient conditions for how nonlinear summation in dendrites increases the computational capacity of a single neuron. First, we showed that a single non-linear dendritic sub-unit is sufficient for a neuron to compute a significant amount of linearly non-separable functions. This result apparently differs from previous studies which found that a single non-linear sub-unit was insufficient to increase the capacity of neurons to compute arithmetic functions [18]. This previous study considered only dendritic non-linearities, but we have shown here that a dendritic non-linearity associated with the somatic non-linearity enables the computation of linearly non-separable functions. We have presented in the present study a situation where a spiking dendritic sub-unit did not trigger a somatic spike, as observed in [45], [46], [3], and still this sub-unit enabled the computation of a linearly non-separable function. Therefore a single dendritic branch is sufficient to boost the computational capacity of a neuron even if its non-linear region cannot trigger on its own a somatic spike. The existence of linear and non-linear integrating regions in our reduced models is supported by the recent demonstration that cerebellar stellate cells or layer 2/3 cortical pyramidal neurons can display both types of integration, even on single dendritic branches: linear EPSP summation in proximal/peri-somatic regions and nonlinear summation in distal regions [10], [36]. While this proximal-distal arrangement could amplify the distal EPSPs and compensate for the electrotonic distance between the inputs and the soma, as the authors argued in [47], [48], we showed how this arrangement can allow neurons to compute linearly non-separable functions.
Second, we showed that a saturating non-linear sub-unit is sufficient for a neuron to compute linearly non-separable functions. To the best of our knowledge, only a single prior study proposed that saturation can enhance single neuron computation, within the context of the coincidence detection in the auditory system [49]. Our study extends this result and show that a neuron with passive dendrites can compute linearly non-separable functions.
Third, we found that scattered synaptic contact is sufficient to implement linearly non-separable functions. This is in contrast to a prior study [26] which suggested that, to implement feature object binding problems, synaptic inputs carrying information about an object should cluster on a single dendritic sub-unit. We have shown here implementation strategies solving FBPs where inputs coding for separate features of the same object are distributed over the whole dendritic tree (see Figure 3 and Supplementary Figures S1 and S2 in Text S1).
The positive linearly non-separable functions that a neuron can compute using a nonlinear dendritic integration can be described as feature binding problems (see Table 2) as defined in [26]: the task of signaling that separate elementary features belong to the same object, and are separate from features defining other objects. For example, a feature can be a sensory input corresponding to a position or an orientation of a bar and an object can be any conjunction of these two features, such as an oriented bar at a given location. The binding problem to be solved by the neuron is then to fire for a bar with a specific orientation present at a specific location, indicating its preferred object, and not to fire for any other combination of such features.
Binocular disparity [22] is a type of FBP problem: in this case input variables code for features coming from the left eye, and code for features coming from the right eye; the neuron then maximally responds if inputs only come from a single eye. Second, the generation of multiple place fields in single dentate gyrus cells [25] is a FBP problem; in this case and are independent sets of input features defining separate spatial locations, and the neuron maximally responds for either of these sets. Finally, binaural coincidence detection [49] is a dFBP problem; in this case code for inputs coming from the left ear and codes for inputs from the right ear; the neuron maximally responds if inputs come from both ears simultaneously. These example applications illustrate the ubiquity of FBP and dFBP-like problems solved by the brain, and consequently why implementations of these Boolean functions by single neurons may be important for understanding neural computation.
The existence of non-linear summation in separate regions of the same neuron's dendrite suggests that a dendritic branch is a finer-grained computational unit than the whole neuron [50]. Our results show that, even when single neuron computation is enhanced by dendritic nonlinearities, it does not necessarily follow that there is a unit of computation smaller than the neuron. We found that a DNF/local strategy implementing such strict independence of dendritic sub-unit computation was possible, but only for the dendritic-spike nonlinearity. We found that an increase in computational capacity could be equally achieved by a global coordination of dendritic sub-units whether the dendritic nonlinearity was of saturating or spiking form. For such a global strategy, there is no sense in which each dendritic sub-unit separately computes a response to its inputs, and thus do not form an independent computational part of the whole function. A specific consequence of the CNF/global strategy is that separate dendritic regions have a different tuning from the whole neuron. This is consistent with Jia et al's [35] observation that a cortical layer 2/3 neuron can maximally respond to a given direction of moving gratings, even though individual dendritic branches are tuned to different orientations or to no orientation. Our results thus show how such a lack of evidence for independent dendritic integration does not imply a lack of dendritic computation. Therefore, the CNF/global implementation strategy suggests that non-linear dendrites may not replace neurons as a basic computational unit but rather expand neurons' computational capacities.
A activation function takes as input a local weighted linear sum and output , this output depends on the type of activation function: spiking or saturating, and on two parameters , the threshold of the activation function, and , the height. The two type of activation functions are defined as follows:
Spiking activation function.
Saturating activation function.The difference between a spiking and a saturating activation functions is that whereas if below . To formally characterize this difference we define here sub-linearity and supra-linearity of an activation function on a given interval . These definitions are analogous to the one given in [20]:
Supra-linearity and sub-linearity.
Note that these definitions also work when using -tuples instead of couples on the interval (useful in Lemma 3). Note that whatever , is both supra and sub-linear on whereas is strictly sub-linear on the same interval.
is not supra-linear on because for all , by definition of . Moreover, is sub-linear on because and for at least one such that and . All in all, is strictly sub-linear on .
Similarly to , is sub-linear on because and for at least one such that and . Moreover, is supra-linear because and for at least one such that and but . All in all, is both sub-linear and supra-linear.
Note that Maass [51] determined the upper limit on the computational capacity of networks made of piece-wise linear threshold functions. However, these activation functions are defined on whereas a saturating activation function is defined on , moreover, the ‘simplest’ studied examples of these type of activation functions are both sub-linear and supra-linear whereas a saturating activation function is strictly sub-linear.
The input output mapping of a binary neuron model is a Boolean function. Let us recall some definitions of this extensively studied mathematical object [34], [52]:
Boolean function. A Boolean function of variables is a function on into , where is a positive integer.
We first introduce a definition what will be useful for our numerical analysis:
Representative Boolean function. is representative of a set of functions if can generate all functions by permuting the labels of the input variables.
For example, for , input vectors can be ordered as follows: ; given these input vectors, the output vectors and are two instances of a function set, as swapping the input labels such that and turns one into the other. This set can be represented by either or . We define the computational capacity of a neuron as the number of representative functions it can compute for a given .
Because of their importance here we recall the definition of positive Boolean functions and linearly separable Boolean functions:
Positive functions. Let be a Boolean function on . is positive such that (meaning that : )
Linearly separable functions. Let be a Boolean function on . is linearly separable if and only if there exist a in and a in such that for all If there exist no such and , we said that is linearly non-separable
In order to describe Boolean functions, it is useful to decompose them into positive terms and positive clauses:
Terms and Clauses.
These terms and clauses can then define the Disjunctive or Conjunctive Normal Form (DNF or CNF) expression of a Boolean function , particularly:
Disjunctive Normal Form (DNF). A complete positive DNF is a disjunction of prime positive terms :
Conjunctive Normal Form (CNF). A complete positive CNF is a conjunction of prime positive clauses :It has been shown that all positive Boolean functions can be expressed as a positive complete DNF ([34] Theorem 1.24); similarly all positive Boolean functions can be expressed as a positive complete CNF. These complete positive DNF or CNF are the shortest possible DNF or CNF descriptions of positive Boolean functions.
First, we generated the list of positive Boolean functions of variables from the list of positive Boolean functions of variables based on [53]. This method generates multiple times the same function, so we removed identical functions from the total list of positive functions.
Second, we extracted from this list of functions the set of representative Boolean functions. We sequentially enumerated the list of monotone Boolean functions; for each monotone function we permuted the input variables label to generate all its children. If none of these children were present in the list of representative functions - initially empty - we recorded the current monotone function in this list. Finally we checked whether the size of this list corresponded to the number of representative positive Boolean functions, which is equal to the number of NP-equivalence classes of unate functions of or fewer variables [37]. With our procedure we also found this number for , which is 490,013,148.
In the three different conditions (, , and ), we systematically enumerated all the integer-valued sets of parameters for different parameter ranges up to the limits given in Table 1 for . This Table displays the parameter values for which the number of computable Boolean functions stops growing. For instance for all positive linearly separable functions can be implemented in a linear model () with integer-valued weights between 0 and 5, and a threshold between 0 and 9. For each parameter set we computed the associated Boolean functions; if this function was in the previously generated list of positive representative function we removed it from the list and recorded the set of parameters in a hdf5 data file. We then went to the next set of parameters and repeated the operation.
All these operations were programmed using python 2.7.1. We used numpy version 1.5 (www.numpy.org) for matrix operation, and h5py 1.3.1 (A. Collette, HDF5 for Python, 2008; http://h5py.alfven.org) to record the parameter sets in an hdf5 file.
This method provides lower bounds on the computational capacity of a neuron with a non-linear dendritic sub-unit. Therefore in and condition the actual computational capacity is superior to the one presented in Figure 2, because we may have missed parameter values for .
For the neuron to correctly produce output of to both input vectors and and for the strategy to be local means that:for response to , because the dendritic sub-unit triggers a somatic spike andfor response to , because in this case the dendritic sub-unit also triggers a somatic spike. In each case, one weight is necessarily larger than or equal to the other; let it be for the first equation and for the second. As is strictly sub-linear (for definition see Materials and Methods), it follows that:andAs each weight is thus at least , we can add these two inequalities to obtain:Finally, as adding any positive weight to the left-hand side does not change the sign of the inequality, so the neuron must also output for input , giving a false positive. Therefore the FBP function cannot be computed using a saturating nonlinearity and a local implementation strategy.
Lemma 1. A two stage neuron with non-negative synaptic weights and increasing activation functions necessarily implements positive Boolean functions
Proof. Let be the Boolean function representing the input-output mapping of a two stage neuron, and two binary vectors and such that . We have non-negative local weights , so for a given dendritic unit we have:We can sum inequalities for all , and because are increasing activation functions:We can sum the inequalities corresponding to every dendritic unit. As , the Heaviside activation function of the somatic unit is increasing we obtain:
Lemma 2. A term (resp. a clause) can be implemented by a unit with a supra-linear (resp. sub-linear) activation function
Proof. We need to provide the parameter sets of a activation function implementing a term (resp. a clause) with the constraint that the activation function is supra-linear (resp. sub-linear). Indeed, a supra-linear activation function (like the spiking activation function) with the parameter set if and otherwise and implements the term . A sub-linear activation function (like the saturating activation function) with the parameter set if and otherwise and implements the clause .
Lemma 3. A term (resp. a clause) cannot be implemented by a unit with a strictly sub-linear (resp. supra-linear) activation function
Proof. We prove this lemma for a term, the proof is similar for a clause. Let be the term defined by , with . First, for all input vectors such that with and it follows that , implying that . One can sum all these elements to obtain the following equality . Second, for all input vectors such that for all then implying that . Putting the two pieces together we obtain:This inequality shows that the tuple of points defining a term must have supra-linear. Therefore, by Definition 2, cannot be both strictly sub-linear and implement a term.
A formal treatment of the three propositions is also given in [41].
We built the compartmental biophysical models with NEURON software version 7.1 [54] coupled with Python 2.7.1. (www.python.org) Morphological parameters (e.g. dendrites' diameters) are described in the Figure 4A. The axon is not modeled because of its negligible contribution to the conductance load.
For the reduced models, the majority of parameters are set to their default value within this NEURON version. The active membrane parameters are the standard Hodgkin-Huxley channels (hh in NEURON) also used with their default parameters; the default and non-default parameters defining passive and active properties of this model are given in Table 3. To model AMPA synapses we used the built-in Exp2Syn synapses; for NMDA synapses we used nmdanet.mod from [55]. The range of tested synaptic weights which correspond to the maximum conductance and the synaptic parameters are described in Table 3. We used biophysical modeling to illustrate the mapping between our abstract binary neuron model and a full multi-compartment dynamical neuron model. We designed the scripts for testing the biophysical models input-output mapping such that they can be used with any type of arbitrarily detailed biophysical neuron model; the script is available in the ModelDB (when this manuscript will be accepted).
For the realistic model of cerebellar stellate cells, the morphological and biophysical parameters are taken from [10].
|
10.1371/journal.pgen.1001041 | Requirement of Male-Specific Dosage Compensation in Drosophila Females—Implications of Early X Chromosome Gene Expression | Dosage compensation equates between the sexes the gene dose of sex chromosomes that carry substantially different gene content. In Drosophila, the single male X chromosome is hypertranscribed by approximately two-fold to effect this correction. The key genes are male lethal and appear not to be required in females, or affect their viability. Here, we show these male lethals do in fact have a role in females, and they participate in the very process which will eventually shut down their function—female determination. We find the male dosage compensation complex is required for upregulating transcription of the sex determination master switch, Sex-lethal, an X-linked gene which is specifically activated in females in response to their two X chromosomes. The levels of some X-linked genes are also affected, and some of these genes are used in the process of counting the number of X chromosomes early in development. Our data suggest that before the female state is set, the ground state is male and female X chromosome expression is elevated. Females thus utilize the male dosage compensation process to amplify the signal which determines their fate.
| When substantially different, sex chromosomes present the challenge of not only gene dose inequity between the sexes, in the heterogametic sex where one chromosome (frequently the Y) carries few genes, but also an inequity relative to the autosomes which are diploid. Dosage compensation refers to the process which equates gene dose between the sexes. Recent results, however, indicate that the mammalian X chromosome avoids monosomy and has a level of expression that is two-fold relative to the autosomes. Hyperactive X chromosome expression in Caenorhabditis elegans has also been suggested, and dosage compensation in the hermaphrodite appears to lower expression of the X chromosomes to match autosome levels. We find that, before the female state is set in Drosophila, the X chromosomes may also express their genes at the two-fold male level and that this level of expression is used to female advantage to consolidate their sex determination. Together, the results suggest that elevated X chromosome expression may be the norm, and that the various dosage compensation processes different organisms utilize reflect a mechanism to counteract an initial hyperactive X chromosome state.
| When the sex chromosomes carry substantially different gene numbers, dosage compensation is necessary to equalize gene expression between the two sexes. In the three best studied model systems Drosophila, C. elegans and mammals where males are XY and females XX, this involves targeting X-specific components which modify the chromatin and transcription of X-linked genes. In each of these cases the end result is different; Drosophila upregulates transcription of the male X by about two-fold, C. elegans downregulates transcription of both X chromosomes in the hermaphrodite by approximately half, and mammals generally shut down transcription of one of the two female X chromosomes (reviewed in [1]).
As it is the Drosophila male which requires dosage compensation, mutation of genes strictly dedicated to this process results in male lethality. The first male specific lethal identified, maleless (mle; [2]), is indeed involved in dosage compensation as are the next identified male lethals, msl-1 and msl-2 [3]. msl-3 identified by Uchida et al. [4] and males absent on the first (mof; [5]) complete the proteins collectively known as the male specific lethals (msls; reviewed in [1], [6], [7]). In addition to these proteins, two RNAs on the X chromosome (the roX RNAs), which are not present in females, are also essential for dosage compensation [8]. Although roX1 and roX2 show no sequence similarity and do not have an open reading frame that could encode a significantly sized protein, they function redundantly; either roX is adequate for function, while loss of both RNAs is required for a failure in dosage compensation and male lethality [9]. The MSL proteins and roX RNAs function as a complex, coating the male X chromosome; the X chromosome is hypertranscribed and MOF acetylates histone H4 on lysine 16. Finally, a protein that appears to be part of the dosage compensation complex (DCC) but is required by both sexes is the JIL histone H3 kinase. JIL is also enriched on the male X chromosome but its loss leads to lethality in both sexes [10].
In 1980, Skripsky and Lucchessi [11] reported that females heterozygous for a Sex-lethal (Sxl) null allele, Sxlf1, and homozygous for mle showed morphological characteristics indicative of sex transformations. Sxl is the Drosophila sex determination master switch, which is on in females but off in males. The Sxlf1/+; mle/mle sex transformation result was confirmed and extended by Uenoyama et al. [12] who observed similar effects with two different mle alleles as well as msl-2 and msl-3. This argued that this phenomenon was not unique to mle, but likely a general property of the msls.
These results, a requirement of male specific genes in females, present a paradox. First, homozygous msl− females show no sex transformations and are fully viable [3]. Second, besides controlling differentiation, a key function of Sxl is to turn off the male dosage compensation system to prevent hypertranscription of the two female X chromosomes, which would otherwise lead to female lethality. As a splicing and translation regulator, Sxl alters the splicing and inhibits translation of msl-2 mRNA so preventing assembly of the DCC [13]–[17]. The absence of MSL-2 also destabilizes MSL-1 and MSL-3 assuring inactivation of the dosage compensation machinery.
The initial activation of Sxl is transcriptional, at the Sxl ‘establishment’ promoter, SxlPe [18]. In cycle 12 of embryogenesis, SxlPe responds to activating X linked genes (known members: sisterless-a (sis-a), sisterless-b (sis-b), runt (run) and unpaired (upd)), in conjunction with positive maternal factors such as Daughterless, balancing their dose against the negative effect of genes on the autosomes (deadpan (dpn), the only identified member) and maternal factors such as Groucho (Gro) and Extramacrochetae (hereafter collectively referred to as the X∶A ratio; reviewed in [19]). Protein from SxlPe transcripts alters the splicing of transcripts from the ‘maintenance’ promoter, SxlPm, first transcribed in cycle 14 in both sexes. In the absence of Sxl protein, default splicing includes a translation terminating exon into the transcripts from SxlPm. As male embryos do not activate SxlPe, Sxl protein is absent and a splicing change on SxlPm transcripts is only effected in females. Females thus set in motion a splicing autoregulatory feedback loop which serves to maintain Sxl expression, and sexual identity, through the rest of the life cycle [20].
Returning to the paradox of a female requirement of male specific genes, one explanation is that XX embryos with only one copy of Sxl fail to reliably activate the gene. These XX cells would be male and are presumably eliminated, due to the gene imbalance from inappropriate dosage compensation. However, when one or more of the msls is mutant, these masculinized XX cells might survive since assembly of the male DCC is prevented. The resulting clones grow but are sexually transformed, so accounting for the observed sex transformations.
While plausible, the above explanation does not account for the recessive nature of Sxl null alleles, which have very high viability (Figure 1A). This high viability requires these females survive the removal of those pockets of male tissue with inappropriate dosage compensation, as Sxl hemizygous females show no male differentiation. Figure 1A also shows that the viability of females with only one Sxl+ allele is badly compromised if maternal MSL-3 is removed. Maternal MSL-1 was next most effective followed by MSL-2, while the effect of MLE was negligible relative to wild type. These data demonstrate a synergism of these msls with Sxl for female viability, as one wild type copy of both Sxl and the msl, is present in these animals. Contrary to the expectation that female survival might be improved if partially masculinized tissue did not perform male dosage compensation effectively, it would appear that females have a need for the msls, when the dose of Sxl is halved. Consistent with our findings, some of the Sxlf1/+; msl/msl combinations in Uenoyama et al. [12] also showed reduced female viability.
The above viability results prompted us to analyze whether key activators of Sxl - the numerator genes sis-a and sis-b, would show a similar interaction with the msls. Figure 1B shows that the effect of a sis-a, sis-b double mutant chromosome is more extreme than a Sxl null, when crossed to mothers mutant for each of the msls. The greater effect of the sis genes is not surprising, given they function in a dose sensitive process to activate Sxl and so determine female sex. What is surprising is that the msls interact with the numerators to promote female viability.
To test whether the loss of a single numerator gene could also affect females, we performed crosses with reduced dose of either sis-a or sis-b. Since msl-3 showed the strongest overall interaction, this msl was examined. Figure 2A shows that sis-a as well as sis-b alone affected females, with sis-b having the stronger effect. The sis gene interactions suggest that very early steps in the female sex determination process are compromised. Testing two Sxl alleles, an early (Sxlf9) versus a late (SxlM1,f12) defective allele, indicated that the early defective allele had an effect, almost as strong as sis-a alone, while the late defective allele did not. These data are consistent with the view that early, dose sensitive events in female sex determination are influenced by the msls. The late Sxl transcripts may not turnover or be as dose sensitive as the early transcripts, so a 50% reduction may not be sufficient to sensitize the females.
sis-a and sis-b are zygotic in their role in female sex determination. To determine whether the effect observed with the msls was maternal or zygotic, reciprocal crosses to Figure 1B were performed. Under these conditions, halving the dose of each of the four msls, including msl-2, reduced female viability (Figure 2B). The zygotic effect was generally weaker than the maternal.
A maternal effect of msl-2 is surprising given that the protein is not detected in females [13], [15], [17]. We note that a maternal effect of msl-2 was also described by Uenoyama et al. [12]. msl-2 RNA is deposited into the egg (Flybase microarray data; http://www.flybase.org/), so the strength of the zygotic effect is presumably influenced by the amount of maternal protein/RNA of each of the msls.
As the msls, particularly MSL-3 and MOF, have been shown to bind to both autosomal and X-linked genes where they might perform an unknown role, we wondered whether the entire male DCC, including MOF and the roX RNAs, influenced female viability. With the numerator gene dose compromised, halving mof dose had an effect, as did roX1 which was much stronger in effect than roX2 (Figure 2B). Since the roX RNAs function redundantly, the impact of roX1 and the weaker interaction of roX2 can be explained by the fact that first expression during embryogenesis is later for roX2than for roX1 [9]. Combined, these results indicate that the msls affect an early event and that the entire male DCC is required for promoting female viability.
The foregoing suggests an event early in Sxl expression is altered by the DCC. To directly assess the effect of the DCC on Sxl transcription, in situs were performed with Sxl probes specific for either the early or late transcripts. Embryos from homozygous mutant mle1, msl-1L-60, msl-21, msl-31 or roX1ex6, roX2− double mutant mothers, mated to heterozygous msl males were analyzed. For the roX1, roX2 double mutant embryos, the roX1−, roX2− males have a duplication of roX2+ on their Y chromosome so only the females are roX1−, roX2−.
In wild type embryos, SxlPe is not activated until cycle 12, its expression becomes stronger in cycles 13 and 14 before it rapidly ceases expression early in cycle 14. For all the msls about half the embryos showed weaker than normal expression of SxlPe, as judged by the size and intensity of the in situ dots on their X chromosomes (Figure 3). The fraction was higher in the roX1−, roX2− cross where all the females are expected to be mutant. These data indicate that the entire DCC complex is used to upregulate transcription from SxlPe.
If, as the data suggest, the primary reason for female lethality is the failure to activate Sxl, a constitutive allele (such as SxlM1) which bypasses the X∶A ratio should rescue them. Since msl-3 showed the strongest interaction in the genetic tests, we determined whether the presence of SxlM1 could rescue the lethality of sis-a, b or Sxl dose reduction in embryos from mothers homozygous for msl-31. The rescue (Figure 4) of 72.8% and 98.7% of the females by SxlM1 for sis-a, b or Sxl dose reduction, respectively, demonstrates that female lethality is primarily caused by the inadequate expression of Sxl.
We next examined whether transcripts from the maintenance promoter, SxlPm, were affected. As shown in Figure 5, this promoter was also affected by loss of DCC components. For msl-1, msl-2 and msl-3 about 50% of the embryos, presumably the homozygotes, showed weaker expression. For mle and the roX1−, roX2− double mutants almost all the females (2-dots/cell embryos) showed weaker than normal expression. As noted for SxlPe, most of the females are mutant for roX1−, roX2− (excepting the few non-disjunction embryos that also receive the Y with a duplicated roX2+ gene), however, only 50% of the embryos are homozygous for mle1. The mle1 data suggest the maternal contribution of MLE is important for SxlPm expression, an effect that appears distinct from the loss of the DCC since for SxlPe only half the embryos were affected. This may be an outcome of SxlPm relying more heavily on maternal MLE compared to the other msls. Alternatively, because MLE also affects the stability of roX1 RNA, which has a larger role in sex determination than roX2 (Figure 2B), the effect of mutating MLE may be amplified as it not only eliminates the maternal MLE but also reduces the levels of roX1 RNA, acting as a double mutation. Despite this unexplained effect on SxlPm by maternal MLE, the data together indicate that both Sxl promoters are susceptible to the DCC and suggest that Sxl, which resides on the X, is a dosage compensated gene. Consistent with the idea that transcription elongation and not initiation is altered by the DCC [21], transcription from both Sxl promoters, which are regulated by different factors, is affected.
Although the in situs of Ore R embryos did not show the distinctly different classes we observed with the msl embryos, to control for the possibility that the quality of the in situs was responsible for generating a poor signal in half the embryos from msl mutant mothers, in situs for SxlPm transcripts were simultaneously performed with a distinguishable second probe - the segmentation gene hairy which has a striped pattern of expression. As seen in Figure 6, embryos from msl mutant mothers that are at the same developmental stage as Ore R embryos have comparable hairy stripes but poor SxlPm signal, indicating that the poor signal is not an artifact of the in situs but an effect of the msls on Sxl transcription.
The in situs are qualitative and the nuclear dots detect transcription directly off the chromosomes, indicating only high levels of transcription. For a better measure, we performed quantitative RT-PCR analysis on 2–3 h and 2.5–3.5 h embryos for SxlPe and SxlPm expression, respectively. Embryos were from homozygous mutant mothers for msl-21, msl-31 or roX1ex6, roX2− double mutants, mated to heterozygous msl males. As for the in situs, in the roX1−, roX2− embryos only the females are doubly mutant as males have a duplication of roX2+ on their Y. RNA levels were normalized to tubulin levels and compared to Ore R embryos which were set to 1.
Figure 7 shows that SxlPe is expressed at lower levels than Ore R embryos in all three msl genotypes. The median for msl-21 embryos was slightly above, for msl-31 and roX1−, roX2− embryos the median slightly below half of Ore R. The medians for SxlPm were also close to half, except for the roX1−, roX2− genotype which was closer to 0.7. SxlPm is transcribed in both sexes and all the males have a functional roX2 gene in the roX1−, roX2− embryos. In these embryos, SxlPm gave a value of 0.7, suggesting males are transcribing SxlPm at close to normal levels while the females express SxlPm at close to half. This would suggest that functional roX2 RNA is present mid-way through cycle 14, a little earlier than in situs can detect [9].
A value close to 0.5 for both promoters (excepting SxlPm for roX1−, roX2−) was a little surprising given the in situ results which show about half the embryos have close to normal levels of transcription. It suggests that the DCC may be upregulating the expression of Sxl by a little more than two-fold, not unlike the roX genes [22]. Alternatively, and not mutually exclusive, it may also indicate that at 2–3 h of development most of the DCC is assembled primarily from maternal reserves and the presence of one wild type chromosome in half the embryos (from the heterozygous fathers) makes a small contribution. With respect to SxlPe, the qRT-PCRs score embryos whose average age is slightly younger than the in situs, at cycles 13 and 14 (2.75–3.25 h). Close examination of those in situs shows few embryos in early cycle 13 with uniform, wild type levels of SxlPe expression. However, when the membranes begin to drop between the nuclei later in cycle 13, the class with more uniform expression resembling wild type, is more readily observed (data not shown). By late cycle 13 and cycle 14, the zygotic contribution of the wild type chromosome from the heterozygous fathers must begin, and the two different classes are more readily apparent in cycle 14 embryos (Figure 3). For SxlPm, the data are more consistent with the DCC having slightly greater than a two-fold effect.
Effects of the DCC were also scored for some of the sex determination genes in the 2–3 h collections from homozygous msl-21 or msl-31 mothers. msl-31 embryos show sis-a, like Sxl, with a median expression close to 0.5, while run and dpn gave medians close to 1 as expected for non-dosage compensated genes (sis-b could not be reliably scored as it has an anti-sense transcript, CG32816). upd appeared reduced to ∼0.7 but this was not statistically significant and the data showed greater variability than for the other genes. This may be because upd begins expression later (cyc 13; [23]), and half the embryos are beginning to perform normal dosage compensation. Also, there are 2 DCC high-affinity sites (see Discussion below) relatively close to upd. These are predicted to make upd less sensitive to the loss of MSL-3, since MSL-3 is required for spreading of the DCC from its initial entry sites.
For msl-21 embryos, the upd median did drop to ∼0.5, consistent with the loss of MSL-2 having a greater effect than MSL-3 for genes with close DCC entry sites. run did not show a significant change from wild type. However, unlike for the msl-31 embryos, sis-a was slightly elevated relative to wild type, while dpn mRNA, at a low level of significance, showed a small decrease. As the msl-31 embryos show that sis-a is dependent on the DCC, these latter data suggest that besides dosage compensation, MSL-2 may have an additional role, one that perhaps affects mRNA stability. MSL-2 affects the steady state levels of the roX RNAs [24]; such an activity could explain the greater variability in the values we measured for msl-21 embryos. To test this, in situs of sis-a mRNA were performed to determine if over time, the mRNA levels would show a change consistent with accumulation. Indeed, we found this to be the case (Figure S1), suggesting that in the case of sis-a MSL-2 may serve to destabilize its RNA. During the early cycles, embryos from msl-21 mothers had signal which was generally weaker than wild type, but by cycle 12 when the message has its highest accumulation in wild type [25], the accumulated levels in the msl-21 embryos were even higher. While alternative explanations, e.g. repression of the sis-a promoter by MSL-2 are also plausible, this effect would have to occur at some but not all stages of sis-a transcription and be independent of the DCC, as loss of MSL-3 shows the predicted 2-fold drop in sis-a mRNA levels.
Despite the suggestion of an additional role beyond dosage compensation for MSL-2, the qRT-PCR data show that the 2 Sxl promoters are expressed at approximately half their normal levels by the loss of the DCC. Expression of other X-linked genes also appears to be similarly affected, very clearly evident in the msl-31 embryos. This indicates the DCC functions relatively early, and may also affect the handful of genes known to be expressed during these early stages of embryonic development [26], [27].
The data argue for a role of the male DCC in females, a function not ascribed to it, and the complex has not been detected in female embryos [28]–[30]. Our data suggest that prior to the full activation of Sxl there is a brief window of male dosage compensation in females, after which Sxl protein is predicted to shut down MSL-2 expression, and destabilize the entire DCC. Not all anti-MSL antibodies have been reported to detect the complex at this early stage, even in males (see [30]). Given this limitation, we used an anti-MSL-1 antibody from the Lucchesi lab which has high sensitivity and enhanced the signal with an M3TAP construct [31]. These embryos were co-stained with anti-Sxl antibodies and closely examined around the cellular blastoderm stage. Figure 8 shows that there is indeed a very brief stage, in mid cycle 14, when it is possible to simultaneously detect both Sxl and the DCC in females. The ant-MSL-1 signal in the female nuclei is not as bright and generally covers an area of DNA larger than in males, presumably the two X chromosomes.
The effect of the DCC on Sxl transcription early in embryogenesis explains the contradiction of why genes that are normally off in females are required to promote their viability. In the absence of the msls and a functional male DCC, transcription of some of the genes on the X chromosome as well as Sxl is not elevated by two-fold. This effectively weakens the X∶A ratio and lowers the levels of the Sxl early as well as late transcripts, which when low enough leads to female lethality. With respect to SxlPe, insufficient levels of early protein are produced and splicing of SxlPm transcripts into the female mode is compromised. With respect to SxlPm, a reduction may compromise establishment of the female state as the autoregulatory splicing feedback loop would have to rely on reduced mRNA and protein levels.
In the absence of mutations in feminizing genes, lowering of Sxl expression by the msls is not detrimental, presumably as the very process which would lead to female lethality - male dosage compensation - is no longer functional, while the Sxl positive autoregulatory feedback loop slowly establishes itself into the female state. Without the DCC, however, reduced dose of feminizing genes, particularly the dose sensing X-linked genes or numerators, lowers SxlPe transcript levels further, and has deleterious consequences for females. Sxl dose also has an effect, but unlike the numerators Sxl is not strictly dose sensitive, and not unexpectedly, when its copy number is halved, has a less profound effect on female viability. It should be noted that extremely low levels of Sxl in females, even in the absence of the male DCC is lethal [32]. Sxl protein directly performs a female dosage compensation role, reducing the levels of X-linked genes such as run [33]; the latter is not upregulated by the male DCC [[34]; Figure 7].
Our data indicate that some of the earliest expressed genes on the X, the numerators as well as Sxl, are dosage compensated. Dosage compensation is a chromosome wide phenomenon, and, at the least, the effect of the msls can be detected as early as cycle 13. Previous work timed the DCC in males at cellular blastoderm (Stage 5, [30]) and early gastrulation (Stage 6, [29]). Our data (qRT-PCRs and in situs) suggest dosage compensation sets in earlier, by 2–3 h in development and appears to initially rely on maternal stores and the zygotic expression of the roX RNAs (roX1 primarily). As discussed by McDowell et al. [30] antibody sensitivity sets the limit for the prior studies. The present studies relied on different assays, which may account for the difference. Indeed, we were also unable to detect convincing signal in males, which is normally stronger, much before blastoderm by antibody staining (Figure 8). It is also possible that early in development the DCC is harder to detect directly as there are fewer genes being transcribed, so less of the complex may have assembled onto the X chromosome before cycle 14, when the mid-blastula transition occurs and there is a large transcriptional burst. The zygotic expression of roX1 RNA has been placed at around 2 h of embryogenesis [9], consistent with the effects we observe.
Targeting of the DCC to the X chromosome, rather than the autosomes, is thought to rely on transcription marks, sequence elements (∼150 MREs – MSL recognition elements and ∼130 HAS – high affinity sites), and other unknown elements [35], [36]. The identified sequence element set is still incomplete since the two data sets show an overlap of 69%; it is predicted that the X chromosome may have as many as 240–300 elements (reviewed in [7]). Examination of the published MRE and HAS shows the closest element to Sxl ∼139 Kbp 5′ of the gene. This distance is on the large side, although it should be noted that all elements which target the DCC to genes on the X remain to be identified; as an example, the white gene has its closest known MRE/HAS 93 Kbp away but its mini form in transgenes, which does not include this site, is clearly dosage compensated. Finally, ChIP data (modENCODE, Flybase) show Sxl with strong H4K16 acetylation marks, a modification dependent on the DCC. ChIP data for JIL-1 kinase also suggest the DCC is at Sxl.
None of the other sex determination genes, other than upd (two 3′ elements at ∼5.6 and 6.8 Kbp away) had an element within 10 Kbp (sis-a ∼26 Kbp, sis-b ∼38 Kbp), consistent with the observation that the msls involved in spreading the DCC from its entry sites on the X (MSL-3, MOF and the roX RNAs), are required for their elevated expression. upd, the exception, showed greater sensitivity to the loss of MSL-2 than MSL-3, as might be expected for dosage compensation which is less dependent on spreading. An interesting correlation is that run which is not compensated, had its closest elements ∼343 (5′) and ∼273 Kbp (3′) away, further than the rest of the other known key sex determination genes.
By using the DCC before the female state is established, Sxl capitalizes on the default male state. Transcription from SxlPe is amplified, an effect unique to females as males do not transcribe from SxlPe. Determination of female identity is thus consolidated. As expression of Sxl protein levels is established, Sxl protein subsequently shuts down the DCC and eliminates the very difference in gene dose between the sexes which set in motion, as well as augmented, its own activation. Implicit, is that before the establishment of Sxl expression, each X-chromosome in females is transcribed at 2X levels, as in males, and our qRT-PCR data of some of the key sex determination genes would support this view. The conventional X∶ A ratio would then be 4∶2 rather than 2∶2, and in males 2∶2 rather than 1∶2 (Figure 9).
In that there is a 2-fold difference between the sexes, this scenario is mathematically the same. However, there are practical and functional implications. An X∶ A ratio that is transiently 4∶2 rather than 2∶2 in females, would have some of the X-linked genes which activate SxlPe at twice the level of their putative counteracting autosomal or denominator genes. In a screen which seeks suppression of a female-specific lethal condition due to a decrease in numerator elements, it would require the equivalent of two autosomal genes to be mutated to reestablish an X∶ A ratio favorable for female survival. Obtaining two mutations in genes functioning in the same process at once is unlikely, which may have skewed the outcome of screens which sought to identify zygotic autosomal genes. It may not be a coincidence that the only autosomal acting component identified is dpn [37], [38]. As both Dpn and Run bind the co-repressor Gro [39], [40] but have opposing effects on SxlPe, it has been speculated they may antagonize each other [39], [41]. Screens may have repeatedly identified dpn as it would be counteracting a gene expressed at its chromosomal equivalent, since run is not upregulated by the male DCC.
On a more general level, our data suggest an upregulation of transcription of the Drosophila X, and may reflect a universal requirement of elevated X chromosome expression to avoid monosomy. Recent microarray analysis of mouse ES cells indicates that mammalian dosage compensation is more complex than previously thought: there is higher expression of the X chromosome relative to the autosomes giving them equivalence, i.e. chromosome per chromosome the X is overexpressed by about two-fold relative to each autosome [42]–[44]. As differentiation proceeds, females lose expression of one of their X chromosomes, silencing it through inactivation. Put in other words, the mammalian X chromosome is not monosomic in expression but rather is hyperactive, and the process of dosage compensation appears to shut down elevated X chromosome transcription in females. (Hyperactive X chromosome expression in C. elegans has also been suggested [42], so dosage compensation in the hermaphrodite would then serve to lower the X chromosomes to match autosome levels).
In this regard, Drosophila would not be very different from mammals except that rather than inactivating one of the female X's, Sxl inactivates the mechanism which upregulates X chromosome specific expression. In all cases, dosage compensation avoids tetrasomy of the X. What the components are which specifically upregulate the mammalian or C. elegans X chromosome—the Drosophila male DCC counterpart—remain to be determined.
Flies were reared under uncrowded conditions on standard cornmeal medium. All crosses were done at 25°C; Ore R was the wild type control. Progeny were counted out to 8 days from the first day of eclosion. Description of genes can be found in Flybase (http://www.flybase.org/).
These were done as in Erickson and Cline [25]. The Sxl early (407 nt) and late (1039 nt) transcript specific probes were generated by the primers, respectively: 5′ GTTCCACTCGTGACAAGTCC 3′and 5′ GTTTCTAAGCAGATCCCG 3′; 5′ GCGAAACGTGCACACTGC 3′ and 5′ GGGCGATGCTTGCATGTTGC 3′ (T7 promoter sequence removed). For hairy, the primers 5′ CCAGAACCTGCTGCTCAT TCG 3′ and 5′ GGGAAAGCGGCTA ACCTCGTTC 3′; for sis-a the primers 5′ CAAAATGCACTACGCCGACG 3′ and 5′ GCATCGTGTCCAACATGACG 3′ were used. All in situs were repeated at least once. Each batch was done simultaneously with an Ore R control, and had sufficient embryos so that several representatives of each cycle could be examined. M3TAP embryos [31] were stained for Sxl (mouse) and MSL-1 (rabbit) as previously described [45]. To enhance the MSL signal, the M3TAP was first bound (blocked) by the same anti-rabbit fluorescent secondary used for the anti-MSL-1 primary before addition of the primary antibody.
Embryos were collected on apple juice agar plates for one hour and aged for the appropriate time. They were washed off the plate, dechorionated with 50% chlorox, washed extensively with 1x PBST and frozen at −80°C. RNA was extracted from the frozen embryos using tri-reagent as per manufacturer's protocol. An additional phenol extraction was performed on the purified RNA, followed by DNAse treatment. A PCR test was performed on the RNA to confirm the lack of DNA, after which 4 ug of the RNA was reverse transcribed (RT) with AMV RT at 50°C for 15 min followed by 1.5 h at 42°C. A small amount (2 ng) of Sxl primer (5′ CGT GTC CAG CTG ATC GTC GG 3′) was added to the oligo-dT mix (100 ng) per RT, as the stage specific 5′ exons of Sxl are distant from the polyA tail. The quantitative PCRs were performed in triplicate on a Bio-Rad iQ5 thermocycler; Ct values that showed a difference of greater than 0.5 from the other two replicates were discarded. For each genotype a minimum of 3 separate RNA samples was analyzed. PCR products were between 200 and 300 bp; primers for SxlPe 5′ CTGTTCGACCATGTCGTCCTA C and CTA CCACCGCTGCCCAGCGAC, SxlPm 5′ GTGGTTATCCCCCATATGGC 3′ and 5′ CTA CCACCGCTGCCCAGCGAC 3′, sis-a 5′ CGTATACGCACCGTATCGCGG 3′ and 5′ GCATCGTGTCCAACATGACG, runt 5′ CGACGAAAACTACTGCGGCG 3′ and CCAGCCAAGCGGGATTCAGC, upd 5′ GAAAGCGGAACAGCAACTGG 3′ and 5′CAGGAACTTGTAGTTGTGCG 3′, dpn 5′ CCGATTATGGAGAAACGTCGC 3′ and 5′ CTGAGCCGCTGACGAACACC. Statistical data analysis was completed using Microsoft Excel and GraphPad Prism.
|
10.1371/journal.ppat.1000887 | Identification of a Mutant PfCRT-Mediated Chloroquine Tolerance Phenotype in Plasmodium falciparum | Mutant forms of the Plasmodium falciparum transporter PfCRT constitute the key determinant of parasite resistance to chloroquine (CQ), the former first-line antimalarial, and are ubiquitous to infections that fail CQ treatment. However, treatment can often be successful in individuals harboring mutant pfcrt alleles, raising questions about the role of host immunity or pharmacokinetics vs. the parasite genetic background in contributing to treatment outcomes. To examine whether the parasite genetic background dictates the degree of mutant pfcrt-mediated CQ resistance, we replaced the wild type pfcrt allele in three CQ-sensitive strains with mutant pfcrt of the 7G8 allelic type prevalent in South America, the Oceanic region and India. Recombinant clones exhibited strain-dependent CQ responses that ranged from high-level resistance to an incremental shift that did not meet CQ resistance criteria. Nonetheless, even in the most susceptible clones, 7G8 mutant pfcrt enabled parasites to tolerate CQ pressure and recrudesce in vitro after treatment with high concentrations of CQ. 7G8 mutant pfcrt was found to significantly impact parasite responses to other antimalarials used in artemisinin-based combination therapies, in a strain-dependent manner. We also report clinical isolates from French Guiana that harbor mutant pfcrt, identical or related to the 7G8 haplotype, and manifest a CQ tolerance phenotype. One isolate, H209, harbored a novel PfCRT C350R mutation and demonstrated reduced quinine and artemisinin susceptibility. Our data: 1) suggest that high-level CQR is a complex biological process dependent on the presence of mutant pfcrt; 2) implicate a role for variant pfcrt alleles in modulating parasite susceptibility to other clinically important antimalarials; and 3) uncover the existence of a phenotype of CQ tolerance in some strains harboring mutant pfcrt.
| Plasmodium falciparum resistance to the antimalarial drug chloroquine has been found to result primarily from point mutations in PfCRT, which provide a highly sensitive marker of in vivo treatment failure and in vitro resistance. Debate has nonetheless continued about the singular role of mutant PfCRT and the contribution of the parasite genetic background. To address this, we have generated recombinant P. falciparum lines expressing a mutant pfcrt allele, or the reference wild type allele, in three distinct chloroquine-sensitive strains. Their analysis reveals a spectrum of responses ranging from high-level resistance to a previously unrecognized tolerance phenotype. The latter is characterized by virtually unchanged chloroquine IC50 values, significantly elevated IC90 values, and the ability to recrudesce after exposure to drug concentrations that are lethal to chloroquine-sensitive parasites. This tolerance phenotype was also observed in an isolate from French Guiana, confirming its presence in malaria-endemic regions. Mutant PfCRT significantly affected parasite responses to other antimalarials, including ones used in artemisinin-based combination therapies, in a strain-dependent manner. Our data suggest that successful CQ treatment of drug-resistant parasites is dependent on both host immunity and the strain-dependent extent to which mutant pfcrt imparts resistance.
| The massive use of chloroquine (CQ) in the 20th century heralded substantial gains in the global fight against malaria. These advances were later lost as CQ resistance (CQR) arose and spread throughout malaria-endemic areas [1], [2]. Today, CQ and the alternative first-line antimalarial sulfadoxine-pyrimethamine have officially been mostly replaced by artemisinin-based combination therapies (ACTs) [3]. Nevertheless, CQ continues to be widely used in parts of sub-Saharan Africa at the household level, presumably because of its ability to provide temporary relief from symptoms for patients unable to afford ACTs or other expensive drugs [4], [5]. Recent findings also suggest the possibility of reintroducing CQ-based combination therapies into African regions where an extended hiatus from CQ use has led to the dominance of CQ-sensitive Plasmodium falciparum parasites that have outcompeted the less-fit CQ-resistant strains [6]. At the cellular level, CQ is thought to act by accumulating to low millimolar concentrations in the acidic digestive vacuole of asexual intra-erythrocytic Plasmodium parasites, wherein it interferes with the detoxification of iron-bound heme moieties produced as a result of hemoglobin degradation [7].
Clinical and epidemiological studies reveal that CQR emerged on very few occasions despite its abundant use, leading researchers to initially posit a multigenic basis of resistance [8]. This theory was challenged by the finding that CQR was inherited as a single locus in a genetic cross between the CQ-resistant Dd2 (Indochina) and the CQ-sensitive HB3 (Honduras) clones [9], [10]. The causal determinant in this locus was ultimately identified as the P. falciparum chloroquine resistance transporter (pfcrt), whose 49 kDa protein product PfCRT resides on the DV membrane [11], [12]. Comparison of the Dd2 and HB3 sequence revealed eight point mutations that all mapped to sites within or near several of the 10 putative transmembrane domains [11].
Quantitative trait loci analysis of the HB3×Dd2 cross has revealed that mutant pfcrt from the Dd2 parent accounts for >95% of the CQ response variation among the progeny [13]. Further evidence supporting pfcrt as the primary determinant of CQR has come from studies of culture-adapted field isolates, which show extensive linkage disequilibrium surrounding the pfcrt locus in CQ-resistant isolates [14]. These data suggest that strong selective sweeps drove mutant pfcrt through P. falciparum populations across the globe, a notion also supported by more recent studies of nucleotide diversity in geographically distinct strains [15], [16]. The PfCRT K76T mutation, ubiquitous to CQ-resistant strains, has proven to be a highly sensitive marker of CQR in vitro and is associated with a significantly increased risk of CQ treatment failure in vivo [17]–[19].
While these studies have demonstrated the primary importance of pfcrt in CQR, other evidence suggests that additional genes might contribute to the CQR phenotype. Most notably, a strain-dependent association has been demonstrated between mutant pfcrt and point mutations in pfmdr1. This may reflect parasite physiologic adaptations to counteract the fitness cost of mutant PfCRT, or an independent role for pfmdr1 in CQR [1], [8], [19]–[22]. Nevertheless, even with identical pfcrt and pfmdr1 alleles, large variations in response to CQ can exist, suggestive of a secondary effect of additional parasite factors [13], [23]–[25].
Clinically, resistance to CQ is graded by the World Health Organization ETF-LTF-ACPR system (corresponding to early treatment failure, late treatment failure, or adequate clinical and parasitological response), based on the time to manifest clinical or parasitological evidence of treatment failure [26]. Studies aimed at dissecting the roles of pfcrt and pfmdr1 mutations in modulating the different grades of in vivo resistance have shown an increased risk of early treatment failure with PfCRT K76T, which in some reports is augmented with PfMDR1 N86Y [27], [28]. However, the PfCRT K76T molecular marker cannot reliably predict CQ treatment failure, revealing moderate specificity of this marker. Discordance between in vitro parasite responses and in vivo patient outcomes following CQ treatment can be as high as 20% [17], [29]. This discordance can be partially attributed to host and environmental factors, including patient immunity, individual pharmacokinetic differences, polyclonal infections, and limitations in obtaining repeated measurements of drug susceptibilities with patient isolates [30]. An additional explanation could be the variable presence of additional parasite determinants.
We have previously adopted allelic exchange strategies to show that different mutant pfcrt alleles could confer verapamil (VP)-reversible CQR in a single, defined genetic background, the CQ-sensitive strain GC03 [31]. A separate transfection-based study found that pfcrt-mediated CQR in two geographically distinct strains, Dd2 (from Indochina) and 7G8 (from Brazil), was entirely dependent on the presence of the K76T mutation [32]. These strains were chosen as they encode a PfCRT haplotype frequently observed in Africa and Asia (Dd2) or in Papua New Guinea, South America and India (7G8). Both alleles have been documented in multiple clinical trials to be highly specific for CQ treatment failures, with repeated evidence of significant selection for mutant pfcrt of either allelic type in early or late treatment failures. Frequencies of mutant alleles in those cases often attained 100% [17], [33]–[37]. Trials were conducted in Africa, Southeast Asia, South America or the Oceanic region.
Here, we have assessed the effect of mutant pfcrt on the CQ response of three CQ-sensitive strains. We also describe two isolates from French Guiana that provide clinical validation of our genetic investigations. Our data reveal the existence of a mutant PfCRT-mediated CQ tolerance phenotype in some strains of P. falciparum.
To define the impact of mutant pfcrt on CQ response in diverse genetic backgrounds, we developed an allelic exchange strategy based on a single round of homologous recombination and single-site crossover integration (Figure 1A), and applied this to the CQ-sensitive P. falciparum strains 3D7 (isolated in the Netherlands), D10 (Papua New Guinea), and GC03 (a progeny of the HB3×Dd2 genetic cross). This strategy differed from an earlier approach that required two rounds of allelic exchange to generate the desired recombinants [31]. Briefly, we constructed selectable transfection plasmids that contained a 2.9 kb pfcrt insert consisting of 0.5 kb of the endogenous 5′ untranslated region (UTR), exon 1, intron 1, and the remaining exons 2–13 (Figure 1A). This truncated 5′ UTR fragment (termed Δ5′) was previously observed by luciferase assays to give insignificant levels of activity (A. Sidhu, unpublished data). Single-site crossover between the pfcrt insert and the homologous pfcrt sequence upstream of codons 72–76 was predicted to replace the endogenous pfcrt gene with a recombinant allele harboring all the single nucleotide polymorphisms from the 7G8 or Dd2 pfcrt allele. Expression of this recombinant allele was driven by the endogenous full-length (3.0 kb) 5′ UTR and a previously characterized, functional 0.7 kb 3′ UTR (termed Py3′) from the pfcrt ortholog in Plasmodium yoelii [31]. In addition to these pBSD-7G8 and pBSD-Dd2 constructs, we also generated the control pBSD-GC03 plasmid that encoded the wild type (WT) pfcrt sequence in order to obtain recombinant control parasites.
3D7, D10 and GC03 parasites were transfected with the pBSD-7G8, pBSD-Dd2, or pBSD-GC03 plasmids and screened monthly by PCR for homologous recombination at the pfcrt locus. With the 7G8 and GC03 alleles, integration into the pfcrt locus was first detected within 60 days of electroporation, and subsequently cloned by limiting dilution. In contrast, the Dd2 allele failed to show PCR evidence of homologous recombination even after 200 days of continuous culture in 3 separate transfection experiments, suggesting that this allele was detrimental to the growth of 3D7 and D10 parasites (data not shown). Repeated efforts failed to transfect 7G8 and Dd2 pfcrt alleles into the CQ-sensitive strains MAD1 and Santa Lucia (from Madagascar and Santa Lucia, a kind gift of Drs Milijaona Randrianarivelojosia and Dennis Kyle respectively), as well as HB3. Recombinant parasites either never appeared following plasmid electroporation and drug selection, or the plasmids never integrated into the pfcrt locus.
Successful transfection of the 3D7, D10 and GC03 strains produced the recombinant mutant clones 3D77G8-1, 3D77G8-2, D107G8-1, D107G8-2, GC037G8-1 and GC037G8-2 (all generated from the plasmid containing the 7G8 pfcrt sequence) or the recombinant control clones 3D7c, D10c and GC03c clones (generated with the control plasmid harboring the WT pfcrt sequence; Table 1). Southern hybridization of EcoRI/BglII-digested genomic DNA samples with a pfcrt probe confirmed the expected recombinant locus, as evidenced by the loss of a 3.9 kb band present in the WT lines and the acquisition of 4.4 kb and 6.7 kb bands consistent with recombination in pfcrt (results shown for the 3D7 and D10 clones in Figure 1B). The 7.2 kb bands present in 3D7C, D107G8-1, and D107G8-2 were indicative of integration of tandem plasmid copies into the pfcrt locus.
We confirmed these recombination events using PCR analyses with a 5′ UTR-specific primer (p1) and an exon 5-specific primer (p2), which revealed a change in size from the 1.8 kb WT-specific band to a shorter 1.5 kb band in the recombinant controls and mutants reflecting the loss of introns 2–4 (Figure 1C). The recombinant controls and mutants also showed the acquisition of PCR bands specific for the full-length functional copy of the pfcrt locus (2.2 kb, p1+p3) and the downstream truncated copy (1.1 kb, p4+p5) (Figure 1C). Sequencing of these PCR products (data not shown) confirmed that the integration event placed the K76T mutation in the functional locus, and that the WT allele was displaced to the downstream non-functional locus. Reverse-transcriptase (RT)-PCR assays on synchronized ring stage RNA with primers specific to exons 2 and 5 (p6+p2) produced a single band corresponding to cDNA, with no evidence of genomic DNA contamination (data not shown). Sequence analysis of those products detected transcripts only from the functional recombinant locus (under the control of the 3.0 kb full-length 5′ UTR) and not from the downstream truncated locus (data not shown). No endogenous WT locus was detected in any recombinant clone.
To precisely assess the transcriptional status of the functional vs. the truncated pfcrt copies, we performed quantitative real-time RT-PCR utilizing primers specific for transcripts containing the Py3′ vs. Pf3′ UTRs respectively. Quantification of pfcrt steady state transcript levels was made by extrapolation from a standard curve generated from genomic DNA of D10C, which has a single copy of the pBSD-GC03 plasmid integrated into the pfcrt locus (Figure 1B). Results showed that transcription from the functional pfcrt allele with Py3′ accounted for 93–95% of the total pfcrt transcript within each line (Figure 1D). Western blot analysis showed that PfCRT protein levels in the recombinant 3D7 and D10 lines were 55–66% and 45–63% those observed in the parental controls respectively (Figures 1E and 1F). This finding of reduced pfcrt transcript and protein expression levels following allelic exchange is consistent with earlier pfcrt transfection studies [31], [32], [38]. Importantly, those studies have shown that reduced pfcrt expression in recombinant lines causes a concomitant reduction in CQ IC50 values, which thus become lower than the IC50 values observed in parasites harboring non-recombinant pfcrt. In our drug assays, the IC50 value refers to the drug concentration that inhibits incorporation of [3H]-hypoxanthine, a marker of in vitro parasite growth, by 50%.
Once the desired integration events were confirmed, we assessed the effect of mutant pfcrt on the CQ response in the recombinant lines. In the 3D7 background, mutant pfcrt was found to confer a 2.7-fold increase in CQ IC50 values (mean±SEM CQ IC50 values of 84±14 nM and 79±11 nM for 3D77G8-1 and 3D77G8-2 respectively) compared to the 3D7 recombinant control (29±2 nM, P<0.001; Figure 2A, Table S1). These values were 2.4-fold lower than the IC50 values for WT 7G8 (190±14 nM). For the D10 mutants, there was no significant increase in CQ IC50 values for D107G8-1 and D107G8-2 compared to D10C (63±11 nM, 71±16 nM, and 45±3 nM, respectively, P>0.05).
When tested against the primary in vivo metabolite monodesethyl-chloroquine (mdCQ), a significant decrease in susceptibility was found in both genetic backgrounds. The 3D7 mutant clones demonstrated a 10-fold increase in mdCQ IC50 values compared to 3D7C (P<0.001, Figure 2B). In comparison, the IC50 values for the D107G8-1 mutant were 5-fold higher than D10C (P<0.01, Figure 2B, D107G8-2 was not tested). Nevertheless, the mdCQ IC50 values in both backgrounds were approximately 2–fold lower than those observed in WT 7G8, suggesting that mutant pfcrt was insufficient to confer high-level mdCQ resistance to 3D7 and D10 parasites.
These findings of a relatively moderate, strain-dependent decrease in CQ susceptibility in the 3D7 and D10 pfcrt mutants contrasted with our earlier observation that the introduction of 7G8 mutant pfcrt in the GC03 background resulted in CQ IC50 values >100 nM [31]. To directly compare the effects of mutant pfcrt between strains, and to assess for any potentially confounding differences in our transfection strategies, we generated recombinant control (GC03C) and mutant clones expressing the 7G8 allele (GC037G8-1 and GC037G8-2) using our single-round transfection strategy. These clones were confirmed by PCR, sequencing, and Southern hybridization, and were found to have similar levels of pfcrt RNA and protein expression compared to the 3D7 and D10 clones (data not shown). In the GC03 background, introduction of the 7G8 mutant pfcrt allele increased the CQ IC50 values 4.7-fold (P<0.001), from 27±3 nM for GC03C to ∼130±8 nM for both recombinant clones (Figure 2A, Table S1), and increased the mdCQ IC50 values by 9-fold (P<0.01; Figure 2B). These determinations included four independent assays that directly compared GC037G8-1 and GC037G8-2 with the C67G8 line. The latter was produced using our earlier pfcrt modification strategy involving consecutive rounds of allelic exchange [31]. C67G8 also expresses the 7G8 pfcrt allele in the GC03 background, yet differs from the clones produced in the current study in that C67G8 contains both the human dihydrofolate reductase and the bsd selectable markers, and lacks the 0.5 kb 5′UTR present in the downstream pfcrt loci in the GC037G8-1 and GC037G8-2 clones (see Figure 1A). Drug assays with these lines produced CQ IC50 values of 131±7, 129±8 and 130±7 nM for GC037G8-1, GC037G8 and C67G8 respectively (Table S1). These results are comparable to our published data with C67G8 (127±17 nM; [31]) and are consistent with both allelic exchange strategies producing the same CQ responses. Our data from all three strains also provide clear evidence that the degree of CQR conferred by mutant pfcrt is strain-dependent.
We also found that the genetic background influenced the degree of VP chemosensitization, a hallmark of P. falciparum CQR [39]. In 3D7 and D10, expression of mutant pfcrt conferred a VP reversibility of 24±1% and 28±1% (calculated as the mean±SEM of percent reversibility for all CQ and mdCQ values), compared to 44±2% for GC03 (Figure S1). Notably, significant VP reversibility occurred in the D10 mutants despite the lack of a significant increase in CQ IC50 values (Figure 2D, Table S1). By comparison, VP reversibility for 7G8 CQ and mdCQ responses was 46±3% (Figure 2B). This is lower than the degree of VP reversibility that results from expression of the Dd2 pfcrt allele [31], [40].
Analysis of the dose response curves generated during these studies revealed a more complex picture than was evident from the IC50 values alone. For all three genetic backgrounds, introduction of the 7G8 mutant allele into the CQ-sensitive strains caused a pronounced change in the slope of the dose-response profiles, with evidence of continued growth at high CQ concentrations (Figures 2C–E). This was particularly pronounced for the recombinant D107G8-1 and D107G8-2 lines, whose CQ IC50 values were similar to those of D10 and D10C, yet whose IC90 values (i.e. the drug concentrations that inhibited [3H]-hypoxanthine uptake into cultured parasites by 90%) were greatly elevated. Indeed, analysis of the CQ IC90/IC50 ratios for the lines in each genetic background revealed significant increases in the mean ratios of the mutant lines (Figure S2). For the 3D7 and D10 backgrounds in particular, the relatively modest increase in CQ IC50 values appeared to be compensated by an increased ability of these parasites to withstand high CQ concentrations.
We posited that these elevated IC90 values imparted by mutant pfcrt subtly reflected a CQ tolerance phenotype. To test this, we assayed our lines for the ability to survive treatment with 50 nM CQ, a concentration that was lethal after three generations of exposure for all three WT strains, and 80 nM CQ, which substantially exceeded each of their CQ IC90 values (Table S1).
Parental, control, and mutant lines were assayed for in vitro recrudescence (defined as 50% of cultures testing positive for growth) after a six-day exposure to CQ. The parental and recombinant control lines from the 3D7, D10, and GC03 backgrounds showed no signs of growth at 30 days post-exposure to 50 nM CQ (Figures 3A and 3B). In contrast, 3D77G8-1 recrudesced at 9 and 13 days post-treatment with 50 nM and 80 nM CQ respectively (Figure 2A). We also tested 3D77G8-1 that had been pretreated with 50 nM CQ for 3 generations approximately 30 days earlier (3D77G8-1/preCQ), and observed similar rates of recrudescence. All untreated lines were positive at day 7, as was WT 7G8 that showed no inhibition of growth with 80 nM CQ treatment.
Although the introduction of mutant pfcrt resulted in no significant increase in CQ IC50 values in the D10 background, both D107G8-1 and pretreated D107G8-1/preCQ recrudesced at days 13 and 17 with treatment with 50 nM and 80 nM CQ, respectively (Figure 3B). In the GC03 background, GC037G8-1 showed no inhibition of growth at 7 days with both 50 nM and 80 nM CQ treatments, reflecting the high-level CQR phenotype imparted by mutant pfcrt in this strain.
Given the evidence that mutant pfcrt was insufficient to confer CQR in all genetic backgrounds, we asked whether there were CQ-sensitive parasites harboring mutant pfcrt in the field. After an extensive search, this led to the identification of two clinical isolates from French Guiana that express the PfCRT K76T marker for CQR but are sensitive to CQ. These isolates, G224 and H209, were harvested in 2003 and 2004, respectively, and were genotyped at the pfcrt and pfmdr1 loci. The PfCRT haplotype of G224 was found to be identical to that of 7G8, whereas H209 possessed a C350R mutation that has not been previously described (Table 1). Both G224 and H209 possessed a single copy of pfmdr1 with the same haplotype that differed from 7G8 only at position 1034. Western blot analyses revealed equivalent levels of PfCRT expression compared to 7G8 (data not shown).
Drug susceptibility assays using CQ and mdCQ showed that these strains had low IC50 values for CQ (mean IC50 values of 52±8 nM and 35±7 nM for G224 and H209, respectively) and mdCQ (mean IC50 values of 349±46 nM and 70±9 nM) (Figures 4A and 4B, Table S1). Further, both G224 and H209 demonstrated VP reversibility of their CQ and mdCQ response (averaging 37% and 35%, respectively; Table S1, Figure S1). Analysis of the CQ inhibition curves revealed that the IC90 values were skewed towards the IC90 of 7G8 (Figure 4C), reminiscent of the effect seen in our 3D7 and D10 mutant pfcrt lines (Figures 2C and 2D). This was particularly pronounced for G224, whose IC90 for CQ was 123±27 nM. When tested for in vitro recrudescence after a 6-day exposure to CQ, G224 recrudesced at days 11 and 17 when treated with 50 nM and 80 nM CQ respectively (Figure 3D). Interestingly, H209 showed recrudescence at days 21 and 25 for 50 nM and 80 nM CQ respectively, despite having a very low CQ IC90 value of 44±7 nM.
To test whether the host strain also influenced the effect of mutant pfcrt on parasite response to other drugs, particularly those currently used in ACTs, we tested our lines against quinine (QN), artemisinin (ART), monodesethyl-amodiaquine (mdADQ, the potent in vivo metabolite of amodiaquine), lumefantrine (LMF), and piperaquine (PIP). The responses of the French Guiana isolates G224 and H209 were also assessed.
In the 3D7, D10 and GC03 backgrounds, we observed no effect of mutant pfcrt on QN response (Figure 5A, Table S1). Interestingly, the highest QN IC50 values were observed with H209, which showed a moderately high level of resistance (405±40 nM). When tested against ART, introduction of mutant pfcrt showed a significant 2–fold decrease in IC50 values in the D10 and GC03 backgrounds, when compared to recombinant clones expressing WT pfcrt (P<0.05 and P<0.01 respectively; Figure 5B). 3D77G8-1 also yielded a 33% lower ART IC50 compared to the 3D7C control, however this did not attain statistical significance (P = 0.06). Again, the highest ART IC50 values were observed with H209 (Table S1). For mdADQ, 3D77G8-1 had a 1.5-fold increase in IC50 value compared to 3D7C (P<0.05), and GC037G8-1 showed an even more pronounced (2.6-fold) increase compared to GC03C (P<0.01; Figure 5C). There was no effect of mutant pfcrt on mdADQ response in the D10 background. With this drug, G224 and H209 were both moderately resistant, as was 7G8.
Introduction of mutant pfcrt was also found to confer significantly increased sensitivity to LMF in all three strains, equating to a 23%, 44%, and 35% decrease in IC50 values for the 3D7, D10, and GC03 backgrounds respectively (Figure 5D). H209 was also found to be less susceptible to LMF than was G224, mirroring their responses to QN and ART. Finally, we found that mutant pfcrt had a significant effect on PIP response only in the D10 background, in which D107G8-1 was 1.7-fold less sensitive than D10C (P<0.05). G224 and H209 were found to be 2.7- and 1.7-fold more sensitive to PIP when compared to 7G8.
Here, we provide evidence that the genetic background of P. falciparum determines whether expression of mutant pfcrt allele confers a full CQR phenotype, as defined by CQ IC50 values that exceed the in vitro CQR threshold [30], or instead mediates increased tolerance to CQ, as evidenced by dose-response shifts manifesting primarily at the IC90 level. All our recombinant clones expressing mutant pfcrt recrudesced in vitro after being exposed for three generations to concentrations of CQ that were uniformly lethal to CQ-sensitive parasites; however the rate of recrudescence varied with the genetic background. The GC03 mutant lines, which had the highest CQ IC50 values, showed no growth inhibition. In contrast, the pfcrt-mutant lines generated in the 3D7 and D10 backgrounds, as well as the clinical isolates G224 and H209, required 1–3 weeks for the detection of recrudescent parasites.
Based on our findings, we propose that IC50 values, which typically constitute the sole measurement of CQ response in vitro, adequately identify high-level CQR but are insufficient to detect strains that have low-level resistance or manifest tolerance to CQ. Instead, our data suggest that accurate determinations of IC90 values provide a more predictive measure of whether parasites can recrudesce in the presence of CQ concentrations that are lethal to drug-sensitive parasites, a trait that we here refer to as CQ tolerance. Tolerance is also apparent in decreased parasite susceptibility to the primary drug metabolite mdCQ. We posit that pfcrt-mediated CQ tolerance might be an important component of late treatment failures in patients. These are classified as cases where symptoms occur during a follow-up period of 4–28 days post CQ treatment, or asymptomatic infection appearing 7–28 days post-treatment (see the WHO 2006 publication on malaria treatment: http://whqlibdoc.who.int/publications/2006/9241546948_eng_full.pdf). In contrast, early treatment failures might result more often from infections with parasites in which mutant pfcrt exerts a higher degree of CQR. Early treatment failures are classified as the development of clinical or parasitological symptoms during the first three days following CQ treatment. We note that clinically, care must be taken when evaluating early failures, as these can also include patients that respond relatively slowly to treatment yet progress to full cure. Moreover, the joint effects of low-level CQ resistance reported here and acquired protective immunity might help explain why CQ treatment can successfully cure some infections harboring mutant pfcrt parasites in semi-immune individuals [1], [18]. The importance of immunity in shaping the host's ability to resolve drug-resistant infections harboring mutant pfcrt was first demonstrated in work from Mali that found that successful CQ treatment of pfcrt mutant parasites was strongly dependent on age, a known surrogate for protective immunity in endemic areas [18]. These data complement other observations in the malaria literature indicating that the immune response can allow a relatively ineffective drug to clear an infection, and even at times clear infections without therapy [41], [42]. Our data extend these reports by suggesting that successful CQ treatment of drug-resistant parasites is dependent both on the level of host immunity and the strain-dependent extent to which mutant pfcrt imparts CQR.
A review of our CQ IC50 data reveals a relatively weak effect of mutant pfcrt, which attained the widely used in vitro CQR threshold of 80–100 nM only for GC03 (Figure 2, Table S1). This threshold, however, was based on studies with field isolates [30] and does not readily extrapolate to our pfcrt-modified parasite lines. Our data (Figure 1) show that these lines underexpress pfcrt, a consequence of allelic exchange into this locus that was earlier shown to cause artificially low CQ IC50 values whose level of reduction was concordant with the degree of reduced expression [31], [32], [38]. In our current study, the importance is not the absolute levels of CQR that we measured, but rather the finding that the genetic background of CQ-sensitive strains dictates a spectrum of mutant pfcrt-mediated changes in CQ response that ranges from tolerance to high-level resistance.
We note that our data were obtained with the 7G8 pfcrt allele, which is known to have appeared independently in South America and the Oceanic region in or near Papua New Guinea and has recently spread throughout India [43]. The 7G8 haplotype (C72S/K76T/A220S/N326D/I356L) shares only two mutations (K76T/A220S) with the Dd2 haplotype (I74E/N75E/K76T/A220S/Q271E/N326S/I356T/R371I) that is common to Africa and SE Asia [11], [14], [44]. Our earlier allelic exchange studies on recombinant lines generated in the GC03 strain found that the 7G8 pfcrt haplotype confers a lower degree of resistance than that imparted by the Dd2 allele (averaging 15% and 45% less or CQ and mdCQ respectively). This was consistent with the intrinsic differences observed between the parental 7G8 and Dd2 strains [31]. It is possible that in the D10 and 3D7 strains, higher degrees of resistance might have been observed with the Dd2 allele, however we were unable to test this. We note that D10 originates from Papua New Guinea where the 7G8 allele is highly prevalent, and our lack of success with introducing the Dd2 pfcrt allele into either this strain or 3D7 suggests a physiologic context that precludes expression and normal viability. Other evidence of a fitness cost imparted by the Dd2 allele comes from studies in Malawi showing that this allele is progressively lost from the parasite population in the absence of sustained CQ pressure [45], [46].
Field studies have sometimes reported discordance in the association of K76T and in vitro CQR, suggesting the contribution of other genetic loci [24], [47]–[49]. However, the interpretation of these results has been confounded by potential inaccuracies stemming from measuring one-time drug responses from frequently polyclonal fresh patient isolates. Our study provides, to the best of our knowledge, the first report of culture-adapted, monoclonal isolates that harbor mutant pfcrt and that, based on multiple drug susceptibility assays, show low CQ IC50 values that fail to meet the standard criteria for CQR. These findings, obtained with the G224 and H209 isolates from French Guiana, therefore provide indisputable evidence that mutant pfcrt is insufficient to confer CQR to all genetic backgrounds. Nevertheless, both isolates exhibited tolerance to high CQ concentrations and recrudesced under CQ pressure. Microsatellite typing revealed a close genetic similarity between G224 and 7G8 (Table 1), with the exception of the residue at PfMDR1 position 1034 that could potentially affect CQ response [22], [50].
Of particular interest, H209 was highly sensitive to CQ and yet demonstrated delayed recrudescence (Figure 4). This might in part be attributable to the PfCRT C350R charge substitution in transmembrane domain 9, a region postulated to function in substrate binding and translocation [51]. Studies are underway to introduce the H209 pfcrt allele, encoding the C350R mutation, into GC03 parasites to compare these to the GC037G8 parasites whose expressed pfcrt allele differs only at codon 350 (Table 1). We note that an adjacent charge substitution at residue 352 (Q352K/R) was previously selected by QN pressure in a CQ-resistant line, with a concomitant reversion to CQ-sensitivity [52]. The H209 line also showed elevated IC50 values for QN, as well as ART, when compared to G224 and 7G8 (Figure 4). Of note, QN-doxycycline, and more recently artemether-LMF, have been implemented as first line antimalarials in French Guiana since the cessation of CQ use for the treatment of P. falciparum malaria in the mid 1990s [53]. Indeed, a recent report from French Guiana documented the existence of several field isolates with elevated artemether IC50 values (>30 nM in 7 of 289 isolates), suggesting decreased susceptibility to this agent [54]. G224 was tested at that time and found to have an artemether IC50 value of ∼1 nM. H209, which yielded artemisinin IC50 values two-fold higher than G224 (Table S1), was isolated one year later. Our subsequent studies reveal comparable IC50 values between these two lines with the more potent clinical derivatives artemether, artesunate and artemether (values provided in Table S1).
ACTs are rapidly assuming the role of first line antimalarials around the world [55]. Our studies with isogenic pfcrt-modified lines confirm previous reports that mutations in PfCRT can significantly affect parasite susceptibility to many of the antimalarials that constitute these ACTs [19], [56], and provide evidence that for certain drugs this effect is strain-dependent (Figure 5). In the case of the fast-acting ART, all three strains displayed enhanced susceptibility upon introduction of mutant pfcrt. With the amodiaquine metabolite mdADQ, elevated IC50 values were noted in two of the three recipient strains, supporting earlier epidemiological evidence that mutant PfCRT might contribute to a multigenic basis of amodiaquine resistance ([57]–[59]; see below). The opposite effect was observed with the bisquinoline PIP, which is highly effective against CQ-resistant strains of P. falciparum [60], and for which we observed a strain-dependent increase in susceptibility. For LMF, significantly enhanced susceptibility was observed in all three genetic backgrounds, supporting recent field studies [59], [61]. The generally enhanced potency of LMF and artemisinin derivatives against mutant pfcrt parasites bodes well for the widely used LMF-artemether co-formulation. The enhanced susceptibility conferred by the mutant pfcrt 7G8 allele to the ACT partner drugs LMF and PIP, but not amodiaquine, has potentially important implications in regional antimalarial drug policy.
Our pfcrt and CQ data speak to a requirement for additional parasite factors that, at least in some strains, either augment the level of PfCRT-mediated CQR or on the contrary, create an intracellular physiologic environment in which PfCRT is unable to exert its full capacity to dictate CQR [62], [63]. pfmdr1 would appear to be one gene that contributes to this strain-dependent effect. Transfection-based studies have shown that in CQ-resistant strains that harbor mutant pfcrt, mutations in pfmdr1 can contribute to elevated CQ IC50 values, but only in a subset of strains. Mutant pfdmr1 alone shows no effect on CQ response in sensitive parasites harboring wild-type pfcrt [19], [20]. Evidence from CQ treatment trials in African, Southeast Asia and the Oceanic region show that mutant pfmdr1 is associated with an increased risk of CQ treatment failure, however this risk is usually substantially higher in the presence of mutant pfcrt [17], [28], [36], [64], [65]. Of note, while mutant pfcrt is virtually ubiquitous to CQ treatment failures, mutant pfmdr1 is often absent ([65] and references therein). Functional assays have yet to be developed to test whether pfmdr1 can directly reduce drug toxicity, or instead is associated with CQR because of its non-random association with mutant pfcrt, which potentially could relate to improved parasite fitness [66].
We note that in our study, pfmdr1 cannot account for differences in the extent to which mutant pfcrt affects CQ response, as both the resistant 3D7 and the tolerant D10 mutants (3D77G8 and D107G8 respectively) share the same wild-type pmfdr1 haplotype (Table 1). The highly resistant GC03 mutants (GC037G8) differ in having the pfmdr1 N1042D mutation that in allelic exchange studies had no impact on CQ response (although it did affect a number of other antimalarials including QN, mefloquine and ART; [21]). Clear evidence that mutant forms of PfCRT and PfMDR1 can combine in a region-specific manner to create higher levels of drug resistance comes from the recent study by Sa et al. [67], showing that the 7G8 South American haplotypes of these two determinants produce high-level resistance to mdADQ. This study also found that the Asian/African Dd2 haplotype of PfCRT was associated with high level CQR with minimal apparent contribution from variant PfMDR1 haplotypes.
Why has no gene other than pfmdr1 been found associated with CQR? In the case of the HB3×Dd2 genetic cross where mutant pfcrt was clearly the primary determinant, evidence that modulatory factors must exist was provided by the 2.7-fold spread in CQ IC50 values observed among the CQ-resistant progeny [13]. Such factors may be present within the 36 kb CQR-associated linkage group harboring pfcrt [10], [68], or potentially might already be present in the HB3 parent, thereby rendering this competent for CQR and masking the inheritance of a secondary determinant [8]. To test the latter hypothesis, we attempted to introduce mutant pfcrt into the HB3 strain, but were unable to obtain integrants in three independent transfection experiments (data not shown). Independent genomic approaches analyzing linkage disequilibrium in CQ-resistant isolates have also failed to identify any gene besides pfcrt [14]–[16], [69], as elaborated upon below.
The genetic identity of these secondary determinants associated with CQR may reflect the geographic distribution of distinct PfCRT haplotypes around the globe [19]. Indeed, the PfCRT 7G8 haplotype found in South America and the Pacific is typically associated with PfMDR1 N1042D/D1246Y (±S1034C), whereas the PfCRT Dd2 haplotype common to Asia and Africa is often associated in CQ-resistant isolates with PfMDR1 N86Y [43], [50]. Identifying additional genetic determinants has been complicated by the complexity of performing genome-wide association studies with large numbers of culture-adapted parasite lines from different geographic regions and comparing these to parasite drug responses [50], [70]. Major advances have recently been achieved in a seminal study by Mu et al. [69], who performed genome-wide association studies with a 3,000 single nucleotide diversity array probed with DNA from189 culture-adapted P. falciparum lines from Africa, Asia, Papua New Guinea and South America, and compared their genetic diversity with CQ response. When accounting for local population structures, the authors found associations between CQ response and changes in pfcrt, pfmdr1, and surprisingly a putative tyrosine kinase (PF11_0079). These associations could only readily be discerned in African populations where a sufficient number of CQ-sensitive strains could be identified; as opposed to South American, Asian and Papua New Guinean strains where mutant pfcrt remained at a high prevalence. Of the genes listed above, pfcrt stood out as being one of handful of genes in the parasite genome that were apparently under very substantial selection pressure in all three populations studied - Asia, Africa and South America. No other genes were convincingly associated with CQR, even though a number of genes potentially involved in drug transport (including the putative drug/metabolite transporter PF14_0260, and the ABC transporters PF13_0271 and PFA0590w) were found to be under lesser selection pressure in local populations. We note that evidence of selection was also observed in genes adjacent to pfcrt, although these may simply represent genetic hitchhiking and insufficient time for genetic recombination to have disrupted these associations.
Our conclusion from these studies is that mutant pfcrt has been the dominant genetic force that has driven CQR across the globe, with some degree of participation from mutant pfmdr1, and that even the phenotype of CQ tolerance observed herein in D10 parasites expressing mutant pfcrt would appear sufficient to confer substantial levels of viability during a course of CQ treatment. This level of protection against drug onslaught, while appearing modest in vitro, appears to have sufficed for selection and rapid mobility through parasite populations subjected to CQ treatment. Experiments to define secondary determinants that can augment CQR would require, as an example, deeper sequence coverage of the set of 189 genotypically and phenotypically characterized isolates mentioned above [69], followed by quantitative trait loci analysis that computationally subtracted the dominant effect of pfcrt to identify potential residual associations in local parasite population structures.
Other hypothesis-driven approaches to identify secondary parasite factors could involve investigations into the function of mutant PfCRT and the cellular basis of CQ mode of action. Recent studies based on heterologous expression of codon-harmonized, surface-expressed PfCRT in Xenopus laevis oocytes have recently provided compelling evidence that mutant PfCRT can transport CQ [71], a finding consistent with earlier evidence from Pichia pastoris and Dictyostelium discoideum [72], [73]. The Xenopus study also identified peptides that could interfere with transport of radiolabeled CQ through mutant PfCRT, raising the possibility that PfCRT is involved in transport of certain peptide sequences out of the DV and into the cytoplasm ([74] and references therein). Secondary factors could potentially alter the kinetics of peptide production (resulting from hemoglobin proteolysis in the DV) or their translocation into the parasite cytosol and subsequent conversion into amino acids that can be incorporated into newly synthesized proteins.
Other potential factors could relate to the tri-peptide glutathione (GSH) and redox regulation. Interestingly, an earlier study by Ginsburg and colleagues reported that altering the intracellular levels of GSH caused a corresponding shift in CQ susceptibility in P. falciparum [75]. Work from these authors led to the hypothesis that GSH could degrade iron-bound heme (a toxic byproduct of hemoglobin degradation) that might be released into the parasite cytosol as a result of CQ action [76]. Further support for a relationship between GSH and levels of CQR was recently obtained following the genetic disruption of the P. falciparum gene PfMRP (PFA0590w), whose ABC transporter product has been localized to the parasite surface. These knockout parasites, generated in the CQ-resistant W2 strain, accumulated more radioactive GSH and CQ and became less resistant to CQ as well as several other antimalarials [77]. Indirect additional evidence of a potential link between CQR and GSH comes from the recent report that PfCRT homologs in Arabidopsis thaliana can mediate GSH transport when assayed in Xenopus oocytes [78]. Collectively, these data suggest that GSH homeostasis is related to CQR, and possibly to PfCRT, in a strain-dependent manner. A multifactorial, and potentially region-specific basis for these differences would have precluded their identification to date. Further investigations into parasite cell biology, employing genomic, proteomic and metabolomic studies to compare CQ response phenotypes within regional populations, are warranted to identify these molecules and their determinants. French Guinea may well provide an ideal set of geographically restricted isolates in which to define these factors, because of its complex history of antimalarial drug usage and the existence of mutant pfcrt strains with both resistance and tolerance phenotypes.
Informed consent was not required for this study as the collection of samples from malaria patients for drug susceptibility testing are part of the French national recommendations for the care and surveillance of malaria. As the Pasteur Institute French Guiana laboratory is the regional malaria reference center, blood samples are sent to the laboratory by practitioners (from health centers, private medical offices and hospitals) for drug susceptibility testing, as part of the national regular medical surveillance. This included in vitro drug susceptibility testing and assessments of molecular markers. This research is mandated by the French Ministry of Health, and has been approved by the Institutional Review Boards of the Pasteur Institute in Paris and in French Guiana.
pfcrt plasmid inserts were assembled from two contiguous sequences. The first 800 bp sequence, spanning 0.5 kb of the pfcrt 5′ UTR (denoted Δ5′) through to the intron 1/exon 2 junction (nucleotides 22960–23747 of the GenBank accession number AF030694), was amplified from Dd2 genomic DNA with the primers p251 and 10AE1-3′A (a list of these and all other primers used in this study is provided in Table S2). A 2.1 kb fragment corresponding to pfcrt exons 2–13 and the 3′ UTR of the P. yoelii ortholog pycrt (termed Py3′) was released following AvrII/BamHI digestion of the plasmids pBSD/AE123 -7G8, -GC03, and -SC01 (the latter has the Dd2 sequence) [31]). These two sequences were assembled in pCR2.1 (Invitrogen) to generate a 2.9 kb pfcrt fragment containing Δ5′, exon 1, intron 1, exons 2–13, and Py3′. This insert was subcloned as a SacII/BamHI fragment into the pCAM-BSD transfection plasmid. This plasmid expresses the bsd selectable marker, which is under control of a 0.6 kb P. falciparum calmodulin (cam) 5′ UTR and a 0.6 kb P. falciparum hrp2 3′ UTR. The resulting 7.2 kb plasmids were designated pBSD-7G8, pBSD-GC03, and pBSD-Dd2.
The P. falciparum 3D7, D10, and GC03 strains were cultured in human erythrocytes, transfected as described [21], and selected with 2.0 mg/ml blasticidin HCl (Invitrogen). Upon integration, recombinant parasites were cloned by limiting dilution and identified using Malstat assays [31]. The isolates from French Guyana were collected from malaria patients referred to the reference malaria laboratory of the Pasteur Institute of Guyana, in Cayenne, France. Each year this work was reviewed and approved by the Pasteur Institute Surveillance Committees of Guyana and Paris. The institutional review board of the Columbia University Medical Center also reviewed and approved the P. falciparum culture work.
PCR-based detection of plasmid integration into transfected parasites (Figure 1) used the pfcrt 5′ UTR-specific primer p1, the pfcrt exon 5-specific primer p2, the Py3′-specifc primer p3, the pfcrt intron 2-specific primer p4, and the plasmid-specific primer p5. For Southern blot analysis, 1 µg of DNA was digested with EcoRI/BglII, electrophoresed, and transferred onto nylon membranes. Hybridizations were performed with a hexamer-primed [32P]-labeled probe prepared from the 0.8 kb fragment spanning Δ5′, exon1 and intron 1, and released following SacII/AvrII digestion of the transfection plasmid pBSD-Dd2. The full-length sequence of pfcrt was determined from the complete coding sequence amplified from cDNA using the primers p251+BB116C and sequenced internally with the primers CF5C, BB84, AF12, AB22, AB25, and BB116B. For sequencing of the upstream pfmdr1 polymorphic residues at positions 86 and 184, genomic DNA was amplified with the primers p423+p231, and the resulting 0.7 kb products were sequenced with p231. For the downstream polymorphic residues at positions 1034, 1042, and 1246, the 0.8 kb amplification product of p426+p215 was sequenced with p238. pfmdr1 copy number was measured by Taqman quantitative real-time PCR and quantified with the ΔΔCt method as described elsewhere [79]. Genomic DNA samples were run twice in triplicate.
The expression of pfcrt in the recombinant clones was assessed by quantitative real-time PCR assays performed with the QuantiTect SYBR Green PCR Kit (Qiagen) on an Opticon2 (BioRad). Expression from the different alleles (endogenous and genetically introduced) was analyzed utilizing primers specific for the two different 3′ UTRs, designated Py3′ and Pf3′. For the loci containing Py3′, the primers p1752 and p1753 were used to generate a 182 bp amplicon. For the locus containing Pf3′, a 191 bp amplicon was generated using the primers p1754 and p1756. PCR conditions were optimized so that the relative efficiencies of the Pf3′ and Py3′ amplifications were equal. Reactions were performed in 25 mL volumes with 300 nM of each primer, 3 mM Mg2+, and 1/80th of the oligo(dT) primed cDNA generated from 1.5 µg of total RNA. As a control for each sample, a 150 bp amplicon of β-actin was amplified using the primers A129 and A130, using the same conditions as for Py3′ and Pf3′ except that the Mg2+ concentration was 3.5 mM. All amplifications were performed with 15 minutes of hot start at 95°C, followed by 40 cycles of denaturing for 30 seconds at 95°C, annealing for 30 seconds at 49°C, and extension for 30 seconds at 62°C. Melting curve analysis was performed for each assay to verify that a single melting peak was produced, indicating a single specific PCR product for each reaction. A standard curve for each reaction was generated with 10-fold serial dilutions of genomic DNA, spanning the range of 5 to 5×105 genome copies). This genomic DNA was prepared from D10C, a recombinant clone shown by Southern hybridization to have a single copy of each locus (Py3′ and Pf3′, Figure 1B). Each sample was run in triplicate on three separate occasions.
Protein extracts were prepared from sorbitol-synchronized trophozoite-stage parasites. For each sample, protein from ∼1×106 parasites was loaded per well, electrophoresed on 12% SDS-PAGE gels, and transferred onto polyvinylidene difluoride membranes. Membranes were probed with rabbit anti-PfCRT antibodies (diluted 1∶2,500) [11], followed by incubation with horseradish peroxidase-conjugated donkey anti-rabbit IgG (1∶10,000; Amersham Biosciences). Rabbit anti-PfERD2 antibodies (diluted 1∶1,000) [80] were used as an independent loading control. Bands were visualized by enhanced chemiluminescence (Amersham Biosciences) and quantified by densitometric analysis of autoradiograph data using NIH ImageJ 1.38× (http://rsb.info.nih.gov/ij). PfCRT band intensities were normalized against the PfERD2 bands to correct for minor differences in protein loading.
Parasite susceptibilities to antimalarial drugs were measured in vitro by [3H]-hypoxanthine incorporation assays, as described [81]. Briefly, predominately ring-stage cultures were seeded in duplicate in 96-well plates at 0.4% parasitemia and 1.6% hematocrit. Parasites were exposed to a range of drug concentrations, or no drug controls, for 72 hr, with 0.5 µCi per well of [3H]-hypoxanthine added at the 48 hr time point. IC50 and IC90 values were extrapolated by linear regression, as described [81]. Compounds were tested in duplicate on 4–11 separate occasions for CQ and mdCQ and 3–12 separate occasions for the other drugs. In some assays, VP was included at 0.8 µM final concentration. Statistical analyses comparing mutant pfcrt-modified lines against recombinant control lines of the same genetic backgrounds were performed using one-way ANOVA with a Bonferroni post-hoc test for CQ and mdCQ, or unpaired student t tests for quinine (QN), artemisinin (ART), monodesethyl-amodiaquine (mdADQ), lumefantrine (LMF), and piperaquine (PIP).
Parasites were assayed for their ability to grow under short-term exposure to high CQ concentrations. Predominately ring-stage cultures were seeded in 96-well plates at 0.2% parasitemia and 1.6% hematocrit. Parasites were exposed for 6 days to no drug, 50 nM CQ, or 80 nM CQ, with daily media changes. Drug pressure was then removed on day 7 and parasite growth was measured using Malstat assays ([31]). From days 7 through 30, media changes and Malstat assays were performed every two days, and the cultures cut 1∶2 into fresh erythrocytes weekly until the detection of positive wells. As part of this experiment, cultures of 3D77G8-1 and D107G8-1 were exposed to 50 nM CQ for 6 days and maintained until parasites became microscopically detectable, at days 15 and 20 respectively. These CQ-pretreated cultures were assayed for recrudescence alongside 7G8, 3D7C, 3D77G8-1, D10C, D107G8-1, GC03C, and GC037G8-1. Data were pooled from two independent experiments in which each line was assayed in duplicate for the no drug controls and in triplicate for the 50 nM and 80 nM CQ treatments.
These were performed as described [82], with minor modifications. Briefly, 100 µL of Malstat reagent was added to 50 µL of culture supernatant and incubated for 1 hr. Absorbance at 595 nM was measured on a VICTOR3 Multilabel Plate Reader (Perkin-Elmer). Wells positive for parasite growth were identified based on absorbance values greater than twice those obtained from control wells with uninfected erythrocytes. Positive wells were verified by microscopic evaluation of Giemsa-stained thin smears.
pfcrt: MAL7P1.27; pfmdr1: PFE1150w; pycrt: PY05061; b-actin: PFL2215w. Pfmrp: PFA0590w. All numbers are from www.plasmodb.org.
|
10.1371/journal.ppat.1004396 | MHC Class II Restricted Innate-Like Double Negative T Cells Contribute to Optimal Primary and Secondary Immunity to Leishmania major | Although it is generally believed that CD4+ T cells play important roles in anti-Leishmania immunity, some studies suggest that they may be dispensable, and that MHC II-restricted CD3+CD4−CD8− (double negative, DN) T cells may be more important in regulating primary anti-Leishmania immunity. In addition, while there are reports of increased numbers of DN T cells in Leishmania-infected patients, dogs and mice, concrete evidence implicating these cells in secondary anti-Leishmania immunity has not yet been documented. Here, we report that DN T cells extensively proliferate and produce effector cytokines (IFN-γ, TNF and IL-17) and granzyme B (GrzB) in the draining lymph nodes and spleens of mice following primary and secondary L. major infections. DN T cells from healed mice display functional characteristics of protective anti-Leishmania memory-like cells: rapid and extensive proliferation and effector cytokines production following L. major challenge in vitro and in vivo. DN T cells express predominantly (> 95%) alpha-beta T cell receptor (αβ TCR), are Leishmania-specific, restricted mostly by MHC class II molecules and display transcriptional profile of innate-like genes. Using in vivo depletion and adoptive transfer studies, we show that DN T cells contribute to optimal primary and secondary anti-Leishmania immunity in mice. These results directly identify DN T cells as important players in effective and protective primary and secondary anti-L. major immunity in experimental cutaneous leishmaniasis.
| Although it is generally believed that CD4+ T cells mediate anti-Leishmania immunity, some studies suggest that CD3+CD4−CD8− (double negative, DN) T cells may play a more important role in regulating primary anti-Leishmania immunity. Here, we report that DN T cells extensively proliferate and produce effector cytokines in mice following primary and secondary L. major infections. Leishmania-reactive DN T cells utilize αβ T cell receptor (TCR) and are restricted by MHC class II molecules. Strikingly, DN T cells from healed mice display functional characteristics of protective anti-Leishmania memory-like cells: rapid and extensive proliferation, effector cytokine production in vitro and in vivo, and accelerated parasite control following secondary L. major challenge. These results directly identify DN T cells as important players in protective primary and secondary anti-L. major immunity in experimental cutaneous leishmaniasis.
| The spectrum of disease collectively called Leishmaniasis is caused by several species of protozoan parasites belonging to the genus Leishmania. The disease is currently endemic in 88 countries, affecting an estimated 12 million people with over 1.5–2 million new cases and 70,000 deaths each year [1]. Because Leishmania parasites reside mainly within macrophages, a strong cell-mediated immunity is required to control intracellular parasite replication and disease progression [2], [3], [4], [5], [6]. Experimental L. major infection in mice closely mimics the human cutaneous disease and is an excellent model for understanding the factors that regulate cell-mediated immunity. Resistance to cutaneous leishmaniasis is associated with strong IFN-γ response, which activates infected macrophages leading to nitric oxide and reactive oxygen species production and destruction of the intracellular parasites [4], [7], [8], [9].
Although it is generally believed that CD4+ T cells play a primary role in mediating anti-Leishmania immunity, a study suggests that they may be dispensable and that MHC II-restricted CD3+CD4−CD8− (double negative, DN) T cells are critical for regulating primary anti-Leishmania immunity [10]. In addition, several studies have reported increased numbers of DN T cells in blood of Leishmania-infected patients [11], [12], dogs [13], and in spleens of Leishmania-infected mice [14]. These cells have been proposed to contribute to primary and vaccine-induced immunity against Leishmania. However, direct evidence implicating DN T cells in anti-Leishmania immunity has not yet been clearly documented. Here, we report for the first time, that infection with L. major leads to activation and proliferation of DN T cells in the draining lymph nodes (dLNs) and spleens of infected mice. These cells produce effector cytokines (IFN-γ and TNF), display functional characteristics of memory-like cells and contribute to optimal primary and secondary protection against L. major infection.
Recovery from natural or experimental L. major infection is associated with strong T cell proliferation and IFN-γ production in spleens and dLNs. To investigate the contribution of CD4+ T cells in this process, we co-cultured CD8+ T cell-depleted splenocytes from healed mice with L. major-infected BMDCs in vitro. Surprisingly, we found in addition to CD4+ T cells, strong proliferative and IFN-γ responses by CD3+CD4−CD8− (DN) T cells (Fig. 1A and Fig. S1A and B). Proliferating DN T cells also produced TNF (Fig 1B), IL-17 (Fig. S2A) and little IL-2 (Fig. 1D), suggesting they are polyfunctional in cytokine production. Indeed, most of the IFN-γ-producing DN cells also co-produced TNF (Fig. 1C). Interestingly, although DN T cells proliferate significantly more than CD4+ T cells, their quantitative ability to produce IFN-γ and TNF was significantly lower than those of CD4+ T cells (Fig. S2B–D). In addition, DN T cells also produced GrzB (Fig. 1E), suggesting they may perform effector functions in L. major-infected mice. DN T cells from L. major-infected mice did not proliferate or produce IFN-γ following stimulation with OVA-loaded DCs (Fig. 1F), but were activated by DCs pulsed with SLA or freeze-thawed L. major (Fig. 1F). Collectively, these results suggest that the proliferation and cytokine production by DN T cells from healed mice are L. major specific.
Our co-culture system showed that Leishmania-specific DN T cells are activated following in vitro recall response. To determine whether DN T cells are activated in vivo, we adoptively transferred CFSE-labeled T cells from healed Thy1.2 mice into naive Thy1.1 mice that were then challenged with L. major the next day. Both CD4+ and DN T cells from healed donor mice showed extensive proliferation and IFN-γ production compared to those from naive mice (Fig. 2A–D). The in vivo relevance of DN T cell response was further confirmed by BrdU incorporation (Fig. 2E and F). Interestingly and similar to CD4+ T cells, the percentage of proliferating and IFN-γ-producing DN T cells in healed mice were significantly higher than those in naïve mice following L. major challenge, suggesting that DN T cells display functional characteristics of memory T cells (rapid proliferation and cytokine production). Indeed, we found that the percentage of DN T cells in lymph nodes (Fig 3A) of healed mice that express CD62LhiCD44hi (central memory-like) was significantly higher (p<0.05) than those in naive mice (Fig. 3B). Following adoptive transfer of whole T cells from healed mice and subsequent L. major challenge, almost all the proliferating donor CD4+ T cells downregulated their CD62L expression (i.e. were CD62Llo). In contrast, the proliferating DN T cells contained an almost equal proportion of CD62Llo and CD62Lhi populations (Fig. 3C). In addition, more DN T cells were CD62LhiCD44hi compared to CD4+ T cells (Fig. 3D).
In addition to αβ T cells, NKT and γδ T cells also do not express CD4 and CD8 molecules. To determine whether Leishmania-reactive DN T cells are NKT and γδ T cells, we assessed the expression of αβ, γδ and NK1.1 molecules on DN T cells by flow cytometry. As shown in Fig. 4A, DN T cells predominately (> 90%) expressed αβ TCR and not NK1.1 and γδ molecules, indicating that they are not NKT or γδ T cells. To further determine whether DN T cells are CD4+ or CD8+ T cells that have down-regulated their surface molecules following activation, we assessed highly enriched (> 99% purity, Fig. S3) DN, CD4+ and CD8+ T cells for CD4 and CD8 transcripts by RT-PCR. DN T cells did not express CD4 and CD8 mRNA (Fig. 2B), suggesting they are not CD4+ or CD8+ T cells that have down-regulated their surface molecules. In addition, highly purified CD4+, CD8+ and DN T cells maintained their respective phenotypes following in vitro restimulation for 5 days with L. major-infected BMDCs (Fig. 4C). To determine whether Leishmania-reactive DN T cells display regulatory properties as previously reported in other systems [12], [15], [16], we co-cultured CD4+ and DN T cells with L. major-infected BMDCs and assessed CD4+ T cell proliferation and IFN-γ production by flow cytometry. DN T cells did not affect CD4+ cell proliferation and IFN-γ production (Fig. 4D), suggesting that they do not exhibit regulatory/suppressive properties.
To determine whether Leishmania-reactive DN T cells are restricted by MHC II molecule, we co-cultured highly enriched T cells from healed mice with infected BMDCs in the presence or absence of anti-MHC II antibodies. Anti-MHC II antibodies blocked proliferation and IFN-γ production by both CD4+ and DN T cells in a dose-dependent manner (Fig. 5A). In addition, L. major-infected BMDCs from MHC II KO mice failed to induce proliferation and IFN-γ production by DN and CD4+ T cells (Fig. 5B). In contrast, proliferation and IFN-γ production by DN T cells were minimally affected following co-culture with infected BMDCs from CD1d KO mice (Fig. 5C), confirming that DN T cells are mostly restricted by MHC II molecules.
We found that Leishmania-reactive DN T cells are recalled in healed mice following L. major challenge in vitro and in vivo suggesting that they may be induced following primary infection. To determine this, we assessed CD4+ and DN T cells response in the dLNs and spleens of infected mice C57BL/6 mice at different times after infection corresponding to early, peak and resolution of lesion progression (Fig. 6A). As expected, there was strong CD4+ T cell response (proliferation and IFN-γ production, Fig. 6B) at all times (3, 6 and 12 weeks) post-infection. Similarly, DN T cells from infected mice also strongly proliferated and produced IFN-γ following restimulation with infected BMDCs (Fig. 6C). In contrast, CD4+ and DN T cells from naïve mice did not proliferate or produce IFN-γ upon stimulation with L. major-infected BMDCs (Fig. 6B and C). Collectively, these results show that Leishmania-reactive DN T cells are induced during primary L. major infection and could contribute to anti-Leishmania immunity.
To determine if DN T cells contribute to primary immunity against L. major, we selectively depleted CD4+ and CD8+ or all T cells by treatment with anti-CD4/CD8 or anti-Thy1.2 mAbs, respectively, during the course of primary L. major infection (Fig. S4). Mice depleted of both CD4+ and CD8+ T cells still had some IFN-γ-producing CD3+ DN T cells (Fig. 7A and 7B) and harbor significantly (p<0.01) lower parasite burden (Fig. 7C) compared to those depleted of all T cells (by anti-Thy1.2 mAb treatment), indicating that DN T cells contribute to optimal control of parasite proliferation during primary L. major infection.
Next, we used both in vitro and in vivo approaches to investigate whether DN T cells contribute to secondary anti-Leishmania immunity. Highly purified DN T cells from healed (but not naïve) mice significantly (p<0.05) inhibited parasite proliferation in infected BMDMs and this effect was comparable to those of CD4+ T cells (Fig. S5A and B). These results provide direct in vitro evidence that DN T cells could control parasite growth in L. major-infected BMDMs.
Next, we used two different experimental approaches to determine whether DN T cells contribute to secondary anti-Leishmania immunity in vivo. First, we selectively depleted CD4+ and CD8+ or all CD3+ T cells (as in Fig. 7A above) in healed mice and after 24 hr, rechallenged them with L. major. As shown in Fig. 7D, CD4+ and CD8+ T cells-depleted mice, which still had DN T cells, retained some level of infection-induced resistance as evidenced by significantly (p<0.01) lower parasite burden compared to naïve mice (primary infection). In contrast, depletion of all CD3+ T cells completely abrogated secondary immunity (Fig. 7D). Second, we assessed the ability of highly enriched (purity > 96%, Fig. S6) DN T cells from healed mice to protect naïve animals against virulent L. major challenge. Adoptively transferred DN T cells from healed mice protected naïve mice against virulent L. major challenge as evidenced by significantly lower parasite burden (Fig. 7E). Collectively, these in vitro and in vivo observations strongly implicate Leishmania-reactive DN T cells in contributing to optimal anti-Leishmania immunity in mice.
Apart from lacking CD4 molecules, DN T cells display functional characteristics similar to CD4+ T cells (MHC-II restriction, proliferation, IFN-γ production and parasite control). To further investigate how Leishmania-reactive DN T cells differ from CD4+ T cells, we compared the transcriptional profile of proliferating DN and CD4+ T cells following restimulation with L. major-infected BMDCs. Although most of the 84 mouse innate and adaptive immune genes showed similar pattern and level of expression in both cell types, some genes were preferentially upregulated or downregulated in DN T cells compared to CD4+ T cells (Fig. 8A). The gene transcripts showing ≥ 2 folds difference in DN T cells were further analyzed and validated by quantitatively real-time PCR (Fig 8B and C). Interestingly, most of the upregulated transcripts in DN T cells were genes associated with innate immune responses, including C3, Mac-1 (CD11b), myeloperoxidase (Mpo), lysozyme, etc. In contrast, the downregulated transcripts (relative to CD4+ T cells) included genes associated with adaptive immunity, including CCR4, Foxp3, Gata-3, etc. Collectively, these results suggest that despite mediating anti-Leishmania immunity (akin to CD4+ T cells), Leishmania-reactive DN T cells are phenotypically distinct from conventional CD4+ T cells.
We show here that DN T cells proliferate and produce effector cytokines in secondary lymphoid organs of mice following primary and secondary L. major challenges. DN T cells from healed mice display functional characteristics of anti-Leishmania memory-like cells: they rapidly proliferate and produce effector cytokines (TNF and IFN-γ) in response to L. major challenge in vitro and in vivo and mediate infection-induced immunity (rapid protection) following adoptive transfer in vivo. Leishmania-reactive DN T cells express predominantly αβ TCR, are restricted by MHC class II molecules, lack immunoregulatory properties and display transcriptional profile distinct from conventional CD4+ T cells. To the best of our knowledge, this is the first extensive characterization and demonstration of the protective ability of Leishmania-reactive DN T cells in vitro and in vivo.
It is generally believed that CD4+ T cells play a dominant role in anti-Leishmania immunity. However, the finding that CD4 deficient mice were resistant while MHC class II deficient mice were highly susceptible to L. major challenged this dogma [10] and suggests that MHC II-restricted CD4−CD8− T cells may be more important in regulating primary anti-Leishmania immunity. Indeed, several studies have reported the expansion of CD3+CD4−CD8− (DN) T cells in the blood of Leishmania-infected patients and dogs, and in spleens of Leishmania-infected mice [11], [12], [13], [14]. These cells have been proposed to contribute to primary and vaccine-induced immunity although a concrete evidence implicating them in immunity has not yet been demonstrated. Our studies directly show the importance of Leishmania-reactive DN T cells in mediating optimal primary and secondary anti-Leishmania immunity in mice.
The precise origin and development of peripheral DN T cells is not clearly understood and is controversial. Some reports suggest that DN T cells originate in the thymus by escaping negative selection [17], [18], [19]. In contrast, several reports suggest that DN T cells are generated in the periphery rather than in the thymus [19], [20], [21], [22]. These cells comprise about 1–5% of total T cells in non-transgenic mice and in humans [11], [23] making them difficult to isolate and subsequently study. TCR transgenic [24] or lpr (Fas mutation) mice [19], [25], which present increasing accumulation of DN T cells are widely used to investigate the function and developmental origin of DN T cells. DN T cells have been shown to influence long-term allograft survival [24], [26], [27], prevent the development of autoimmune disease [28], [29], [30], and contribute to control of intracellular pathogens [31], [32]. In addition, DN T cells have been shown to possess immunoregulatory and alloreactive properties, inhibit autoreactive CD4+ T cells and mediate MHC I-restricted killing of allogenic target cells [20], [24], [25]. Our studies show that Leishmania-reactive DN T cells are restricted by MHC class II and may not have immunoregulatory properties because they failed to suppress CD4+ T cell proliferation in vitro (Fig. 4D). Rather, a large percentage of proliferating DN T cells produced IFN-γ, TNF, IL-17 and GrzB, which is consistent with their effector functions as seen in other studies [11], [12], [29].
Previous studies that have reported the expansion and possible protective role of DN T cells in leishmaniasis focused mainly on primary Leishmania infection [11], [12], [13], [14]. We extend these studies during secondary immunity by showing rapid expansion and effector functions (cytokine production and parasite control) by DN T cells following challenge infection. Healed mice had more proliferating and IFN-γ-producing DN T cells compared with naive mice following L. major challenge (Fig. 2), and adoptive transfer of DN T cells from healed (but not naïve mice) rapidly protected naïve mice against virulent L. major change. Moreover, DN T cells from healed mice expressed high levels of CD44 and majority of them were CD62LhiCD44hi, which are characteristics markers expressed by central memory-like cells. Collectively, these results suggest that DN T cells display functional characteristics of memory cells and contribute to optimal secondary immunity against L. major.
How do DN T cells mediate their anti-Leishmania immunity? We speculate that this may be related in part to their ability to produce IFN-γ and TNF, key cytokines that activate infected macrophages leading to intracellular parasite killing. Indeed, we found that Leishmania-reactive DN T cells in the spleens and lymph nodes are highly proliferative and produce IFN-γ, TNF and granzyme B. Importantly, we also found that DN T cells from immune mice were recruited to and proliferate at the infected footpads (Fig. S7). In addition, our in vitro co-culture experiments with infected BMDMs and highly enriched DN T cells show that suppression of parasite proliferation was associated with increased nitric oxide production, a key effector molecule that mediate destruction of parasites in infected cells.
The findings that DN T cells mediate comparable (or even superior) protection against L. major in vitro and in vivo may challenge the dogma that CD4+ T cells are the major T cell subset that mediates anti-Leishmania immunity. Indeed, the proliferation of DN T cells was either comparable or sometimes higher than those of CD4+ T cells following in vitro or in vivo L. major challenge (see Figs. 1–3). Interestingly, although the percentage of IFN-γ-producing DN T cells was sometimes higher than those of CD4+ T cells, their MFI was significantly lower (Fig. 1C), an observation that explain the relatively lower IFN-γ transcripts in DN compared to CD4+ T cells (Fig. 8). In addition, the numbers of Leishmania-reactive CD4+ T cells were quantitatively (∼ 3–4 fold) higher than those of DN T cells. Thus, despite their superior proliferative response, DN T cells may still play a subordinate role to CD4+ T cells in vivo. Furthermore, it is conceivable that CD4+ T cells may be required for proper activation and effector functions of DN T cells. In line with this, we have observed that proliferation and IFN-γ production by highly enriched DN T cells is impaired in cultures devoid of immune CD4+ T cells in vitro and in vivo (Fig. S8). It is conceivable that DN T cells may assume increased roles in the absence of CD4+ T cells. For example, SIV infection in nonhuman primates does not result in immune dysfunction and progression to simian AIDS because DN T cells partially compensate for defective CD4+ T cell functions upon SIV-induced CD4+ T cell depletion in these animals [33], [34]. Similarly, a strong DN T cell-mediated HIV Gag-specific response has been associated with seronegativity in HIV-exposed individuals [35].
It is interesting that the expression of genes associated with innate immune responses including C3, were significantly higher in Leishmania-reactive DN T cells than in CD4+ T cells. While commonly associated with initiation of inflammation and critical molecule involved in first line of defense against pathogens, the complement proteins, particularly C3 and its degradation fragments are also known to prominently influence the adaptive immunity [36], [37]. Recent studies have been shown that some subset of T cells express C3 and that its intracellular activation is not only required for homeostatic T cell survival [38], but also in optimal Th1 induction and differentiation into effector cytokine (particularly IFN-γ) production [38], [39]. It is conceivable that C3-expressing DN T cells in L. major-infected mice might be involved in IFN-γ production leading to effective macrophage activation, nitric oxide production and parasite killing.
Collectively, our studies provide direct evidence for DN T cells in mediating anti-Leishmania immunity akin to CD4+ T cells. We propose that DN T cells complement CD4+ T cells to mediate efficient primary and secondary anti-Leishmania immunity in mice. In the absence of DN T cells, the induction of effective anti-Leishmania immunity may be either delayed or impaired. In a recent preliminary study, we observed impaired induction of DN T cells in spleens and draining lymph nodes of L. major-infected highly susceptible BALB/c mice. It would be interesting to determine whether the susceptibility of BALB/c mice to L. major infection is related in part to this impaired expansion of DN T cells. Collectively, our studies clearly identify DN T cells as important subset of T cells that contribute to optimal anti-Leishmania immunity.
All mice were kept at the University of Manitoba Central Animal Care Services (CACS) facility in accordance to the Canadian Council for Animal Care guidelines. The University of Manitoba Animal Use Ethics Committee approved all studies involving animals, including infection, humane endpoints, euthanasia and collection of samples.
Six to 8 wk-old female C57BL/6 (Thy1.2) mice were obtained from Charles River, St Constante PQ, Canada. Thy1.1 and MHC class II deficient (MHC II KO) C57BL/6 mice were purchased from The Jackson Laboratory (Bar Harbor, ME). Female CD1d deficient C57BL/6 mice were kindly supplied by Dr. Xi Yang from a breeding colony maintained at the University of Manitoba Central Animal Care Services (CACS) Facility.
Leishmania major parasites (MHOM/80/Fredlin) were grown in M199 culture medium (Sigma, St. Louis, MO) supplemented with 20% heat inactivated FBS (HyClone, Logan, UT), 2 mM glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin. For infection, mice were injected with 2×106 (primary infection) or 5×106 (secondary infection) stationary-phase promastigotes in 50 µl PBS suspension into the right (primary) or left (secondary) hind footpad. Lesion sizes were monitored weekly by measuring footpad swelling with calipers. Parasite burden in the infected footpads was determined by limiting dilution assay. Parasite titers were determined from the highest dilution at which growth was visible.
Bone marrow cells were isolated from the femur and tibia of naïve C57BL/c mice and differentiated into macrophages using complete medium supplemented with 30% L929 cell culture supernatant as previously described [40]. BMDCs were differentiated in petri dishes in the presence of rmGM-CSF (20 ng/ml; Peprotech, Rocky Hill, NJ). BMDMs and BMDCs were infected at a cell-to-parasite ratio of 1∶5 and after 6 hr, free parasites were washed away and infected BMDCs were used to stimulate purified CD3+, CD4+ or DN T cells from naïve or healed mice in vitro. To assess the ability of CD4+ or DN T cells to control parasite proliferation, infected BMDMs were co-cultured with CD4+ or DN T cells and parasite proliferation in infected BMDMs was determined at different times by counting Giemsa-stained cytospin preparations under light microscope at ×100 (oil immersion) objective.
Infected mice were sacrificed and spleens and dLNs were collected and made into single-cell suspensions. Cells were labeled with CFSE dye (1.5 mM; Molecular Probes, Eugene, OR) and resuspended at a concentration of 2×106 cells per milliliter in RPMI 1640 supplemented with 10% heat-inactivated FBS, 100 U/ml penicillin, 100 µg/ml streptomycin, and 5×10−5 M 2-mercaptoethanol (complete medium), plated with 100 µl per well in 96-well tissue culture plates, and stimulated with infected BMDCs (BMDC: T cell = 1∶100) or soluble anti-CD3/CD28 mAb (1 µg/ml; BioLegend, San Diego, CA). After 5 days, proliferation and cytokine production were determined by flow cytometry. In some experiments, CFSE-labeled T cells from spleens and dLNs of infected mice were co-cultured with L. major-infected WT, MHC II KO, or CD1d KO BMDCs for 5 days, stimulated with PMA, BFA and ionomycin for 4–6 hr and proliferation, IFN-γ, TNF, IL-2, IL-17 and GrzB expression by different T cell subsets were analyzed by flow cytometry. In some experiments, anti-MHC II antibodies were used to block MHC II-TCR interaction in vitro.
Healed (> 12 weeks post-infection) Thy1.2 C57BL/6 mice were sacrificed and single-cell suspensions from the dLNs and spleens were made. T cells (Thy1.2+) were enriched by positive selection using mouse CD90.2 (Thy1.2) selection kit according to the manufacturer's protocols (StemCell Technologies, Vancouver, BC). Enriched T cells (> 98% purity) were labeled with CFSE dye, and 107 cells were adoptively transferred into naive congenic (Thy1.1) mice by tail vein injection. After 24 hr, the recipient mice were challenged with 5×106 L. major, sacrificed after 7 days and cell proliferation and IFN-γ expression by donor (Thy1.2) cells in the dLNs and spleens were determined directly ex vivo.
For in vitro co-culture experiments, DN (CD3+CD4−CD8−) and CD4+ T cells were purified from pooled spleens and dLNs of healed or naïve mice by cell sorting (FACSAria III, BD Biosciences). For in vivo adoptive transfer studies, DN T cells were enriched using a combination of in vivo depletion and positive selection. Briefly, L. major-infected and healed mice (> 12 weeks post-infection) were first injected with 200 µl GK1.5 and TIB210 ascites (i.p) to deplete CD4+ and CD8+ cells. After 48 hr, DN T cells were purified using mouse CD90.2 selection kit (StemCell Technologies, Vancouver, BC). Enriched DN T cells were > 99% negative for CD4 and CD8 expression and > 95% positive for CD3 by flow cytometry.
To assess the numbers (percentages) and proliferation of DN T cells at the site of infection, CFSE-labeled whole T cells from L. major-infected Thy1.2 mice were adoptively transferred into naïve Thy1.1 mice that were then challenged with L. major. After 7 days, recipient mice were sacrificed and donor cells were recovered from the footpads as we previously described [41]. Briefly, the footpads were disinfected in 70% ethanol, the skins were peeled off and homogenized gently in PBS with tissue grinders. The crude homogenates were resuspended in 7 ml of cold PBS, carefully layered on top of 5 ml Ficoll and the infiltrating cells were separated by centrifugation according to the manufacturer's suggested protocols. The cells were collected, resuspended in 5 ml complete medium, counted, stained directly for expression of various cell surface markers and analyzed by flow cytometer by gating on Thy1.2+ donor cells.
For reverse transcription-PCR (RT-PCR), cells from spleens of healed mice were stained with fluorescent-conjugated anti-CD3, anti-CD4 and anti-CD8 antibodies. CD4, CD8 and DN T cells were sorted to high purity by gating on CD3+ cells. CD4, CD8 and GAPDH gene expression in sorted cells were analyzed by RT-PCR. For PCR array, CFSE-labeled whole spleen cells from healed mice were stimulated with L. major-infected BMDCs for 5 days and proliferating (CFSElo) CD4+ and DN T cells were purified by cell sorting. Eighty-four innate and adaptive immune genes in CD4+ and DN T cells were analyzed with Mouse Innate & Adaptive Immune Responses PCR Array kit (Qiagen, Frederick, MD). PCR array was performed by a real-time cycler (Bio-Rad CFX96) and analyzed with web-based PCR Array Data Analysis Software (Qiagen, Frederick, MD). To quantify gene expression levels, equal amounts of cDNA were mixed with SYBR Green PCR master mix (Toyobo, Osaka, Japan) and primers specific for the gene of interest (Table S1). 18S rRNA was amplified as an internal control.
Naïve and healed mice were injected with 2 mg of BrdU i.p. per mouse and then challenged with 5×106 L. major in the next day. BrdU solution was prepared in sterile water, protected from light exposure, and changed daily. The night before the assay, mice were injected i.p. with 0.8 mg of BrdU in PBS. The next day, mice were sacrificed, spleens were harvested and BrdU staining was performed using BrdU Staining Kit according to the manufacturer's suggested protocol (BD PharMingen).
Healed mice were depleted of CD4 and/or CD8 T cells by injecting i.p. 200 µl ascites containing anti-CD4 (GK1.5) or anti-CD8 (TIB 210) mAb (or both) per mouse or depleted of total T cells by injecting i.p. 100 µg anti-Thy1.2 mAb (TIB 107) per mouse, once a week, and then challenged with 5×106 L. major.
Data are presented as means and standard error of mean (SEM). Two-tailed Student's t-test or ANOVA were used to compare means and SEM between groups using GraphPad Prism software. Differences were considered significant at p<0.05.
|
10.1371/journal.pntd.0001162 | Latent Microsporidial Infection in Immunocompetent Individuals – A Longitudinal Study | Microsporidia (Fungi) have been repeatedly identified as the cause of opportunistic infections predominantly in immunodeficient individuals such as AIDS patients. However, the global epidemiology of human microsporidiosis is poorly understood and the ability of microsporidia to survive and multiply in immunocompetent hosts remains unsolved.
To determine the presence of latent microsporidia infections in apparently healthy humans in the Czech Republic, the authors tested sera, urine and stool originating from fifteen persons within a three month period examined on a weekly basis.
Sera, stool and urine samples originating from fifteen HIV-negative people at risk with occupational exposure to animals, aged 22–56 years, living in the Czech Republic were tested by indirect immunofluorescence assay (IFA) for the presence of specific anti-microsporidial antibodies, standard Calcofluor M2R staining for the detection of microsporidian spores in all urine sediments and stool smears and molecular methods for the microsporidial species determination.
Specific anti-microsporidial antibodies were detected in fourteen individuals, asymptomatic Encephalitozoon spp. infection was found in thirteen and E. bieneusi infection was detected in seven of those examined. While E. hellem 1A and E. cuniculi II were the major causative agents identified, seven different genotypes of E. bieneusi were recorded.
These findings clearly show that exposure to microsporidia is common and chronic microsporidiosis is not linked to any clinical manifestation in healthy population. Moreover, our results indicate much higher incidence of microsporidial infections among an apparently healthy population than previously reported. These results open the question about the potential risk of reactivation of latent microsporidiosis in cases of immunosupression causing life-threatening disease.
| Microsporidia are a group of obligate intracellular parasitic fungi that have risen over the past two decades from obscure organisms to well recognized human pathogens. Out of 14 species reported to infect humans and causing more severe symptoms in immunocompromised individuals, microsporidia of the species Encephalitozoon and Enterocytozoon bieneusi are the most frequent causes of life-threatening chronic diarrhea and systemic disease in HIV patients and acute, self-limited diarrhea in immunocompetent persons. Although the diagnosis and clinical management of microsporidiosis cases have improved significantly recently, the epidemiology of human microsporidiosis is still unclear. To identify the occurrence of latent microsporidia infections in apparently healthy people, the authors tested sera, urine and stool originating from fifteen persons within a three month period. They found specific antibodies against microsporidia in sera originating from fourteen individuals, and using molecular tools, they detected microsporidial infection intermittently in all tested people. The presence of detectable amounts of microsporidial spores demonstrated that exposure to microsporidia is more common than previously believed and microsporidiosis is not linked to any clinical manifestations in healthy people. This finding should make the clinician more aware of the risk of this unapparent infection, its potential reactivation after immunosupression and consequences leading to life-threatening disease.
| Microsporidia have emerged as causative agents of opportunistic infections in AIDS patients and other immunodeficient individuals. Several species of microsporidia can cause disease in humans. Intestinal microsporidiosis due to Encephalitozoon (Septata) intestinalis and Enterocytozoon bieneusi are most frequently reported among immunocompromised people including patients with acquired immune deficiency syndrome (AIDS) [1], [2] and other immunocompromised patients such as transplant recipients [3]–[6]. Encephalitozoon cuniculi and E. hellem are less prevalent among immunodeficient patients [7], [8]. Infections with microsporidia in immunocompetent individuals such as travelers have also been described [9], [10].
Although the most common clinical symptoms related to encephalitozoonosis among immunodeficient patients are chronic diarrhea and malabsorption, they can also cause systemic diseases. While immunocompetent persons often have mild or self-limiting disease, AIDS patients can experience weight loss and increased mortality [11].
Since the studies examining the prevalence of microsporidiosis have been limited to patients who are infected with human immunodeficiency virus (HIV) or who have AIDS, the global epidemiology of human microsporidiosis is poorly understood. Variation of spore shedding intensity of microsporidia was shown in both human and animals [12]–[16]. However, to our knowledge there have been no reports on the spore shedding pattern of microsporidia in immunocompetent humans. Therefore we aimed to study the pattern of microsporidia spore shedding in a cohort of asymptomatic apparently healthy people.
The study was approved by the Hospital České Budějovice ethics committee (protocol no. 202/07).
Written informed consent was obtained from every person prior to examination.
Between September and December 2007, a total of 180 individual stool and 180 urine samples were collected on the weekly basis for 3 months from fifteen HIV-negative people at risk of occupational exposure to various animals, such as farm ruminants, pigs, poultry and rodents. The male to female ratio was 8 (53%) to 7 (47%) with mean age of 35±11 years and range between 22–56 years. The samples were stored at 4°C in the dark without any conservation and examined immediately. Every specimen in the study was supplemented with data on the person's clinical symptoms (e.g., indigestion, abdominal pain).
Prior to the study, serum samples were obtained from all individuals included and the presence of specific anti-microsporidial immunoglobulin G was tested by indirect immunofluorescence assay (IFA). IFA was performed with purified whole spores of E. hellem, E. cuniculi or E. intestinalis grown in vitro in VERO E6 cells and semi-purified spores of E. bieneusi at the concentration 105 spores/well (spores kindly provided by Dr. G.S. Visvesvara, CDC Atlanta, GA, USA). Sera were serially diluted (1∶8, 1∶16, 1∶32, 1∶64, 1∶128 and 1∶256) in PBS and compared with negative and positive control sera. Sera with positive fluorescence at titers greater than 128 were considered positive.
Standard Calcofluor M2R staining [17] was used for the detection of microsporidian spores in all urine sediments and stool smears. Stained slides were examined by fluorescence microscopy using UV light with a wavelength of 490 nm and at a magnification of 1000×. Positive control slides were used for each examination.
The DNA was isolated from the stool and urine samples using homogenization by bead disruption using FastPrep–24 Instrument (MP Biomedicals, CA, USA) and DNA was extracted using commercially available isolation kit QIAamp DNA Stool Mini Kit (QIAGEN, Hilden, Germany) according to the manufacturer's instructions. Acquired DNA was stored at −20°C.
The nested PCR protocol by Katzwinkel-Wladarsch et al [18] amplifying the ITS region of Encephalitozoon spp. and Enterocytozoon bieneusi using microsporidia-specific primers was performed as described elsewhere [16]. As positive controls the following were used: DNA obtained from spores of E. intestinalis originally isolated from AIDS patients [19] and grown in vitro in VERO E6 cells in the Laboratory of Veterinary and Medical Protistology at the Institute of Parasitology ASCR, and DNA from spores of E. bieneusi of genotype D originally isolated from a pig [16]. PCR products were visualized on a 2% agarose gel containing 0.2 µg/ml ethidium bromide and directly sequenced on the ABI3730XL sequence analyzer (Applied Biosystems, Foster City, CA). Sequences were aligned and completed using programs ChromasPro (Technelysium, Pty, Ltd.) BioEdit and Clustal X 2.0.6 and compared with sequences in GenBank.
Specific anti-microsporidial antibodies were detected in fourteen out of fifteen tested people; 87% of sera reacted with E. cuniculi, 47% with E. hellem, 13% with E. bieneusi and none with E. intestinalis (Table 1). None of the individuals demonstrated any clinical symptoms (loose stool, indigestion, etc.).
While no positive finding was revealed among samples of urine using microscopy, four stool samples originating from 3 persons were positive for microsporidia spores, which were subsequently molecularly characterized as E. bieneusi.
During the twelve week long observation of spore excretion, microsporidia were molecularly detected in 34 urine samples (19%) and 39 stool samples (22%) originating from all fifteen tested people. Each of the person excreted microsporidial spores intermittently in irregular intervals (Figure 1). The concurrent infection with two species of Encephalitozoon, E. cuniculi and E. hellem, and Enterocytozoon bieneusi was detected in 7 individuals, co-infection with E. cuniculi and E. hellem in three cases and monoinfections with E. bieneusi, E. hellem or E. cuniculi in one or three individuals, respectively. No E. intestinalis infection was detected. Whereas both Encephalitozoon spp. infections were more often found in urine, E. bieneusi was detected equally in urine and stool samples (Figure 1). While E. hellem or E. cuniculi infection were caused mainly by predominant genotype 1A or II, seven different genotypes of E. bieneusi including novel genotypes CZ4–CZ6 were identified in both urine and stool samples. All other ITS sequences from the study samples were a 100% match to the reference genotypes from GenBank listed in Table 2.
The actual extent of microsporidian infections is unknown. Microsporidia were, and still are, often overlooked and misdiagnosed because they are not specifically searched for in most diagnostic labs, they are rather small, and their staining with hematoxylin and eosin is not sufficient. Most of what is now known about human microsporidiosis can be attributed to the experience with patients infected with HIV [1], [2]. However, with increased awareness and improved diagnostics, microsporidia have become more frequently reported also in immunocompetent individuals, producing asymptomatic infections [20]–[22]. Despite limited sample number our findings showed a well-supported correlation between spore presence in excretions and seropositivity, which discriminates the actual latent microsporidiosis from simple consumption and passage of spores through the intestinal tract.
Intermittent spore shedding for a long period has been experimentally demonstrated for several hosts including rabbits with E. cuniculi [12], wild-type mice with E. intestinalis [13], pigs with E. bieneusi [15], budgerigars with naturally acquired Encephalitozoon spp. infection [23] and HIV-positive patients with E. bieneusi [14]. The persistence of microsporidia despite resolution of the intestinal disorder suggests that microsporidia infection may cause clinical symptoms (e.g., diarrhea) during the early stages of infection that could be overlooked and resolved even though the microsporidia persist.
Our survey was performed on a limited sample size from a highly selected population, which could result in decreased statistical power. On the basis of present results it is obvious, that prevalence data of microsporidial infection reported by various authors reaching up to 38% the case of Encephalitozoon spp. and 51% for E. bieneusi, could be hampered by collection of only a single sample for diagnosis, especially in low level infections. While the twelve week sampling enabled us to detect E. cuniculi in 86% of tested people, E. hellem in 66% and E. bieneusi in 47%, the hypothetical individual single sampling performed at any day would identify E. cuniculi in only 0–27% of persons, E. hellem in 0–13%, and E. bieneusi in 0–13%.
Based on data in the literature and our experience, it seems that the incidence of microsporidial infections is much higher than previously reported and microsporidia may represent neglected etiological agent of more common diseases. However, it is not known how extensive such silent infections are in asymptomatic carriers, including both humans and animals, which have been reported increasingly to harbour various species and genotypes of microsporidia [16], [24], [25]. Moreover, the fact that microsporidia DNA were detected in urine sediments suggests, that microsporidia are able to disseminate also in immunocompetent hosts despite previously reported protective T-cell mediated adaptive immunity together with several components of innate immunity [26], [27]. Furthermore, the majority of prevalence studies currently rely on detection of spores in stool samples only. The results of this study clearly showed that infected seropositive person could excrete detectable amount of microsporidial DNA via urine, nevertheless examination of stool sample will be negative. Detection of specific antibodies seems to be more sensitive than one-shot detection of spores and can provide more accurate information about ongoing microsporidia infection.
In conclusion, studies focusing on the epidemiology of microsporidiosis will more clearly define the environmental sources of microsporidia that pose a risk for transmission so that preventative strategies can be implemented. Since no data exist about latent infection in immunocompetent carriers, possible infection reactivation in these individuals and person to person transmission risk via organ donation, such epidemiological data must be compared with experiments that could solve this question definitively. Moreover, using detection methods with a high sensitivity, such as PCR, and consecutive sampling from every individual is recommended to provide more precise epidemiological data.
|
10.1371/journal.pcbi.1000355 | An Analytically Solvable Model for Rapid Evolution of Modular
Structure | Biological systems often display modularity, in the sense that they can be
decomposed into nearly independent subsystems. Recent studies have suggested
that modular structure can spontaneously emerge if goals (environments) change
over time, such that each new goal shares the same set of sub-problems with
previous goals. Such modularly varying goals can also dramatically speed up
evolution, relative to evolution under a constant goal. These studies were based
on simulations of model systems, such as logic circuits and RNA structure, which
are generally not easy to treat analytically. We present, here, a simple model
for evolution under modularly varying goals that can be solved analytically.
This model helps to understand some of the fundamental mechanisms that lead to
rapid emergence of modular structure under modularly varying goals. In
particular, the model suggests a mechanism for the dramatic speedup in evolution
observed under such temporally varying goals.
| Biological systems often display modularity, in the sense that they can be
decomposed into nearly independent subsystems. The evolutionary origin of
modularity has recently been the focus of renewed attention. A series of studies
suggested that modularity can spontaneously emerge in environments that vary
over time in a modular fashion—goals composed of the same set of
subgoals but each time in a different combination. In addition to spontaneous
generation of modularity, evolution was found to be dramatically accelerated
under such varying environments. The time to achieve a given goal was much
shorter under varying environments in comparison to constant conditions. These
studies were based on computer simulations of simple model systems such as logic
circuits and RNA secondary structure. Here, we take this a step forward. We
present a simple mathematical model that can be solved analytically and suggests
mechanisms that lead to the rapid emergence of modular structure.
| Biological systems often display modularity, defined as the seperability of the
design into units that perform independently, at least to a first approximation
[1]–[5]. Modularity can be seen
in the design of organisms (organs, limbs, sensory systems), in the design of
regulatory networks in the cell (signaling pathways, transcription modules) and even
in the design of many bio-molecules (protein domains).
The evolution of modularity has been a puzzle because computer simulations of
evolution are well-known to lead to non-modular solutions. This tendency of
simulations to evolve non-modular structures is familiar in fields such as evolution
of neural networks, evolution of hardware and evolution of software. In almost all
cases, the evolved systems cannot be decomposed into sub-systems, and are difficult
to understand intuitively [6]. Non-modular solutions are found because they are
far more numerous than modular designs, and are usually more optimal. Even if a
modular solution is provided as an initial condition, evolution in simulations
rapidly moves towards non-modular solutions. This loss of modularity occurs because
there are so many changes that reduce modularity, by forming connections between
modules, that almost always a change is found that increases fitness.
Several suggestions have been made to address the origin of modularity in biological
evolution [5], [7]–[16], recently reviewed by
Wagner et al [17]. Here we focus on a recent series of studies that
demonstrated the spontaneous evolution of modular structure when goals vary over
time. These studies used computer simulations of a range of systems including logic
circuits, neural networks and RNA secondary structure. They showed that modular
structures spontaneously arise if goals vary over time, such that each new goal
shares the same set of sub-problems with previous goals [18]. This scenario is
called modularly varying goals, or MVG. Under MVG, modules
spontaneously evolve. Each module corresponds to one of the sub-goals shared by the
different varying goals. When goals change, mutations that rewire these modules are
rapidly fixed in the population to adapt to the new goal (Figure 1 A,B).
In addition to promoting modularity, MVG was also found to dramatically speed
evolution relative to evolution under a constant goal [19]. MVG speeds evolution
in the sense that it reduces the number of generations needed to achieve the goal,
starting from initial random genomes. Despite the fact that goals change over time,
a situation that might be thought to confuse the evolutionary search, the
convergence to the solution is much faster than in the case of a constant goal
(Figure 2 A,B).
Intriguingly, the harder the goal, the faster the speedup afforded by MVG evolution.
To summarize the main findings of [18],[19]:
Since these findings were based on simulations, it is of interest to try to find a
model that can be solved analytically so that the reasons for the emergence of
modular structure, and for the speedup of evolution, can be more fully understood.
Here we present such a simple, exactly solvable model. The model allows one to
understand some of the mechanisms that lead to modularity and speedup in
evolution.
The guiding principle in building the model was to find the simplest system that
shows the salient features described in the introduction. It turns out that many of these features can be
studied using a linear system, similar to those used in previous theoretical
work on evolution [8], [20]–[23].
Consider a system that provides an output for each given input. The input is a
vector of N numbers. For example, the input can represent the
abundance of N different resources in the environment. The
output is also a vector of N numbers, for example the
expression of the genes that utilize the resources. The structure that evolves
is represented by an N×N matrix,
A, that transforms the input vector v to a desired
output vector u such that:(1)
The matrix A can be thought of, quite generally, as the linearized
response of a biological regulatory system that maps inputs to outputs, taken
near a steady-state of the system. In this case the vectors u and
v represent perturbations around a mean level, and can have
negative or positive elements.
An evolutionary goal in the present study is that an input vector v
gives a certain output vector u. We will generally consider goals
G that are composed of k such input-output
vector pairs.
To evaluate the fitness of the system, we follow experimental studies in
bacteria, that suggest that biological circuits can be assigned benefit and cost
[24]. The benefit is the increase in fitness due to
the proper function of the circuit, and the cost is the decrease in fitness due
to the burden of producing and maintaining the circuit elements. In this
framework, fitness is the benefit minus the cost of a given structure
A.
We begin with the cost of the system, related to the magnitude of the elements of
A. We use a cost proportional to the sum over the squares of
all the elements of A:. This cost represents the reduction in fitness due to the need
to produce the system elements. A quadratic cost function resembles the cost of
protein production in E. coli
[24]–[26]. The cost tends to
make the elements of A as small as possible. Other forms for the
cost function, including sum of absolute values of
aij and saturating functions of
aij, are found to give similar conclusions as
the quadratic cost function (see Text S1).
In addition to the cost, each structure has a benefit. The benefit
b of a structure A is higher the closer the actual
output is to the desired output: (where Fo represents the maximal
benefit). In the case where the goal includes k input-output
pairs, one can arrange all input vectors in a matrix V, and all
output vectors in a matrix U, and the benefit is the sum over the
distances between the actual outputs and the desired outputs . In total, the fitness of A is the benefit minus
the cost:(2)
The first term on the right hand side represents the cost of the elements of
A, and the second term is the benefit based on the distance
between the actual output, AV, and the desired output,
U. The parameter ε sets the relative
importance of the first term relative to the second.
In realistic situations, the parameter ε is relatively
small, because getting the correct output is more important for fitness than
minimizing the elements of A. Thus, throughout, we will work in the
limit of ε much smaller than the typical values of the
elements of the input-output vectors.
Now that we have defined the fitness function, we turn to the definition of
modularity in structures and in goals.
A modular structure, which corresponds to a modular matrix A, is
simply a matrix with a block diagonal form (Figure 3). Such matrices have non-zero
elements in blocks around the diagonal, and zero elements everywhere else. Each
block on the diagonal maps a group of input vector components to the
corresponding group of output vector components. An example of a modular
structure is
In addition to the modularity of the structure A, one needs to
define the modularity of varying goals. In the present study, modularity of a
goal is defined as the ability to separate the input and output components of
u and v into two or more groups, such that the
outputs in each groups are a function only of the inputs in that group, and not
on the inputs in other groups. Thus, the inputs and outputs in a modular goal
are separable into modules, which can be considered independently (Figure 3). In the present
linear model we require that the outputs in each group are a linear function of
the inputs in that group. For example, consider the following goal
Go that is made of two input-output pairs:and
Here the first component of each output vector is a linear function of the first
component of the corresponding input vectors, namely the identity function. The
next two components of each output vector are equal to a linear 2×2
matrix,
L = [(0.5,0.5);(−0.5,
0.5)], times the same two components of the input vector. In fact, the
modular matrix A given above satisfies this goal, since
Av1 = u1
and
Av2 = u2.
Thus, the input-output vectors in Go can be
decomposed into independent groups of components, using the same linear
functions. Hence, the goal Go is modular. Note that
most goals (most input-output vector sets with N>2)
cannot be so decomposed, and are thus non-modular.
To quantify the modularity of a structure A we used the modularity
measure Qm based on the Newman and Girvan measure
[18],[27], described in [18]
and also in the Text S1. Under this measure, diagonal matrices have high modularity,
block modular matrices show intermediate modularity and matrices with non-zero
elements that are uniformly spread over the matrix have modularity close to zero
(Figure 3).
In the following sections we analyze the dynamics and convergence of evolution under
both fixed goal conditions and under MVG conditions. For clarity we first present a
two–dimensional system
(N = 2), and then move to present
the general case of high-dimension systems. Each of the sections is accompanied by
detailed examples that are given to help to understand the system behavior. The
third section describes full analytic solutions and proofs.
We studied a model for evolution under temporally varying goals that can be exactly
solved. This model captures some of the features previously observed with
simulations of more complex systems [18],[19]: MVG leads to
evolution of modular structures. The modules correspond to the correlations in the
goals. Furthermore, evolution is speeded up under MVG relative to constant goals.
The harder the goal is, the faster the speedup of MVG relative to evolution under a
constant goal. Most random non-modular goals do not generally lead to speedup or
evolution of modularity, but rather to evolutionary confusion. Although the modular
solution is sub-optimal, it is selected for its ability to adapt to the different
varying goals.
The speedup of evolution under MVG is a phenomenon that was previously found using
simulations, but lacked an analytical understanding. The present model offers an
analytical explanation for the speedup observed under MVG. The speedup in the model
is related to small eigenvalues that correspond to motion along fitness plateaus
when the goal is constant in time. These eigenvalues become large when the goal
changes over time, because in MVG, the plateaus of one goal become a high-slope
fitness region for the other goal. Switching between goals guides evolution along a
‘ramp’ that leads to the modular solution. This analytical
solution of the dynamics agrees with the qualitative analysis based on sampling of
the fitness landscape during the evolutionary simulations of complex models [19].
One limitation in comparing the present model to more complex simulations is that the
present model lacks a complex fitness landscape with many plateaus and local maxima.
Such plateaus and local fitness maxima make constant-goal evolution even more
difficult, and are expected to further augment the speed of MVG relative to constant
goal conditions. A second limitation of the present linear model is that it can
solve different MVG goals when presented simultaneously - a feature not possible for
nonlinear systems. This linearity of the model, however, provides a clue to how MVG
evolution works: whereas each goal supplies only partial information, all goals
together specify the unique modular solution. Under MVG evolution, the system
effectively remembers previous goals, supplying the information needed to guide
evolution to the modular solution, even though at each time point the current goal
provides insufficient information. This memory effect is likely to occur in the
nonlinear systems as well.
The series of studies on MVG, including the present theory, predict that organisms or
molecules whose environment does not change over time should gradually lose their
modular structure and approach a non-modular (but more optimal) structure. This
suggestion was supported by a study that showed that bacteria that live in
relatively constant niches such as obligate parasites that live inside cells, seem
to have a less modular metabolic network than organisms in varying environments such
as the soil [32],[33]. Another study considered modularity in proteins,
which corresponds to distinct functional domains within the protein. It was found
that proteins whose function is relatively constant over evolutionary time, such as
the ribosomal proteins present in all cells, are typically less modular in structure
than proteins that are specific to a few cell types and that repeatedly duplicate
and specialize over evolution [34]. Thus, one might envisage a tradeoff in
biological design between modularity and optimality. Modularity is favored by
varying goals, and non-modular optimality tends to occur under more constant goals.
In summary, the present model provides an analytical explanation for the evolution of
modular structures and for the speedup of evolution under MVG, previously found by
means of simulations. In the present view, the modularity of evolved structures is
an internal representation of the modularity found in the world [32]. The
modularity in the environmental goals is learned by the evolving structures when
conditions vary systematically (as opposed to randomly) over time. Conditions that
vary, but which preserve the same modular correlations between inputs and outputs,
promote the corresponding modules in the internal structure of the organism. The
present model may be extended to study additional features of the interplay between
spatio-temporal changes in environment and the design of evolved molecules and
organisms.
|
10.1371/journal.pcbi.1004100 | Resting-State Temporal Synchronization Networks Emerge from Connectivity Topology and Heterogeneity | Spatial patterns of coherent activity across different brain areas have been identified during the resting-state fluctuations of the brain. However, recent studies indicate that resting-state activity is not stationary, but shows complex temporal dynamics. We were interested in the spatiotemporal dynamics of the phase interactions among resting-state fMRI BOLD signals from human subjects. We found that the global phase synchrony of the BOLD signals evolves on a characteristic ultra-slow (<0.01Hz) time scale, and that its temporal variations reflect the transient formation and dissolution of multiple communities of synchronized brain regions. Synchronized communities reoccurred intermittently in time and across scanning sessions. We found that the synchronization communities relate to previously defined functional networks known to be engaged in sensory-motor or cognitive function, called resting-state networks (RSNs), including the default mode network, the somato-motor network, the visual network, the auditory network, the cognitive control networks, the self-referential network, and combinations of these and other RSNs. We studied the mechanism originating the observed spatiotemporal synchronization dynamics by using a network model of phase oscillators connected through the brain’s anatomical connectivity estimated using diffusion imaging human data. The model consistently approximates the temporal and spatial synchronization patterns of the empirical data, and reveals that multiple clusters that transiently synchronize and desynchronize emerge from the complex topology of anatomical connections, provided that oscillators are heterogeneous.
| The spontaneous or resting-state activity of the brain is organized into multiple spatial patterns of correlated activity. These patterns have been associated with functional interacting brain networks. Recent studies show that the correlations among brain regions are not stationary, but evolve over time, and have refocused the study of spontaneous brain activity on characterizing these time-varying functional interactions. In this article, we show that the synchrony between the BOLD activities of different brain regions displays global slow fluctuations that reflect the dynamical association and dissociation of functional synchronized clusters. Using a network of anatomically connected phase oscillators, we show that transiently synchronized patterns emerge from the interplay between nonlinear dynamics and the complex, but static, network topology. Our results suggest that the brain constantly explores its dynamical repertoire during rest, which allows for an all-around visitation of functional states.
| The spontaneous activity of the brain is organized into multiple spatial patterns of correlated activity across different brain regions, known as ‘resting-state networks’ (RSNs) [1–12]. The topography of the resting-state networks overlaps with functional networks observed during cognitive load, including the default mode network, the fronto-parietal network, and other attention, visual, auditory, and sensorimotor networks [13–16]. The temporal evolution of resting-state activity has only recently been subject to investigation. Recent studies have demonstrated that the correlations among brain regions, both within and between networks, evolve over time [11, 17–20]. These results suggest that spatial patterns are formed, dissolved, and reformed over time, so that resting activity can be divided into subsets or “communities” of brain regions that strongly interact for a period of time. Importantly, time-varying functional connectivity has been reported both in awake humans and anesthetized macaques [20], thus the temporal variability of functional interactions is not likely to be produced by transitions between different mental states, but it may be an emergent property of the complex brain network.
These results suggest that the interplay between space and time dimension is crucial to unravel the mechanistic origins of spontaneous activity. How non-stationary functional connectivity emerges in the brain network remains an open question though. Current large-scale models of spontaneous activity have been built to approximate the averaged functional connectivity and assume that the stationary long-range functional correlations rise from the interplay between the underlying anatomical connectivity structure and the neural dynamics [21–24]. It is unclear what mechanisms generate the temporal fluctuations in functional connectivity. Are the time-varying interactions governed by the topology of the brain anatomical connectivity? Or do they reflect the variation of the structural connections due to interplay between the dynamical states and the network couplings, such as, for example, in the case of short term synaptic plasticity?
In the present study we used phase synchronization to measure, with enhanced temporal resolution, the time-varying functional interactions of the resting-state fMRI BOLD signals of human subjects. This approach has been successfully applied to the study of fMRI data during rest and during viewing of natural scenes [25–27]. We found that global oscillations and multiple transient synchronized clusters (i.e. communities) are present in the data. By studying a network of phase oscillators in which connections are given by the brain’s anatomical connectivity, we showed that transiently synchronized networks emerge from the topology of anatomical connections and the heterogeneity among oscillators.
We analyzed the BOLD activity from a total of 48 scanning sessions, from 24 healthy human subjects, of 600s (sampled in T = 300 frames) during resting-state condition, and acquired using standard fMRI techniques (see Methods). The dataset consists of whole-brain BOLD signals, averaged over n = 66 cortical brain regions (see S1 Table).
The interaction between BOLD signals of different brain regions was measured using the instantaneous phase synchronization. For obtaining the phase of each brain region, the signals were first band-pass filtered within the narrowband 0.04–0.07Hz. Previous work has shown that this frequency band contains more robust and functionally relevant signals than the other bands [26]. Moreover, narrowband filtering is a methodological requirement for obtaining meaningful signal phases [26]. As shown in the Methods section, phase interactions fairly describe the interactions among the narrowband signals and, as shown in the following, they allow for a time-resolved analysis of interactions (see also S1 Fig.). As shown in Fig. 1a, the phase relation between two given BOLD narrowband signals changes over time, alternating periods during which the oscillations are in-phase and periods during which the oscillations are out-of-phase. To quantify these phase relations we computed the instantaneous phase φk(t) of each narrowband signal k using the Hilbert transform (HT) (see Methods). After this, the first and last 10 time steps were discarded to avoid border effects inherent to the HT, so that in the following T = 280. We next calculated the phase difference, Δφkl(t) = φk(t)–φl(t), for each pair of brain regions k and l, at each time step t. Across all pairs of signals and time steps, the probability density function (p.d.f.) of the pairwise phase differences, noted Pr(Δφ), is centred on zero (Fig. 1b). However, Pr(Δφ) is not constant, but evolves over time, going from a uniform distribution (i.e., the phases are independent) to a distribution that is densely concentrated around zero (i.e., high phase synchrony) (Fig. 1c, color plots).
To describe the temporal evolution of Pr(Δφ) we computed the order parameter, R(t)=|∑k=1nejφk(t)|/n, that measures the global level of synchronization of the n oscillating signals. Under complete independence, the n phases are uniformly distributed and thus R is nearly zero, whereas R = 1 if all phases are equal. The temporal evolution of R(t) effectively tracts the evolution of Pr(Δφ) (Fig. 1c, white traces). Phase synchronizations were significantly higher (p<10–10, t-test) than the expected accidental synchronizations of phase shuffled surrogates, designed to decorrelate the phases while preserving the power spectrum of the original signals (Fig. 1d) (see Methods). Notably, the power spectrum of R shows that the global level of synchronization evolves on an ultra-slow (<0.01 Hz) time-scale (Fig. 1e). The peak frequency averaged across all scanning sessions is equal to 0.006±3.10–4 Hz, i.e., one order of magnitude slower than the frequency of the narrowband signals (comprised between 0.04–0.07 Hz). This also robustly observed in individual scanning sessions for which the peak frequency of R ranges between 0.003–0.010 Hz. Finally, we evaluated the statistics of the order parameter R obtained using the broadband signals. We found that both the distribution of R and the peak frequency of its time evolution was preserved using the broadband signals (Fig. 1d, f), indicating that the observed slow fluctuations of the global synchronization are not a product of the narrowband filtering.
In the following, we show that the spatiotemporal patterns of synchronization reflect the formation and break up of various different clusters of synchronized brain regions. To show this, we used a decomposition technique, recently proposed for the detection of communities in time-varying networks [28], to represent both the topology of synchronized clusters and their activation over time.
Specifically, we reduced the spatiotemporal distribution of the n phases into a binary three-dimensional matrix, or tensor, of size n×n×T, for each scanning session. At each time step t, a symmetric n×n synchronization matrix was constructed, with (i, j) elements equal to 1 if |φj(t)-φi(t)|<π/6 and 0 otherwise (1 ≤ i, j ≤ n). The synchronization matrix evolved in time and displayed different topological patterns that last over several seconds and reoccur over time (Fig. 2). To detected these patterns, each constructed tensor was decomposed into a sum of K rank-one tensors, or “components”, using a non-negative tensor factorization (NNTF) technique, a higher-order analogue of Principal Component Analysis (see Methods) (Fig. 3a). Each of the K components is a rank-one tensor, namely an outer product of 3 vectors, ak, bk, and ck—in our case, due to symmetry we have: ak = bk. This decomposition separates space and time: the n elements of ak give the participation weight of each node in the component k, i.e. the community structure of component k, while the T elements of ck relate to the activation level of component k at time step t, i.e. the activation of community k over time.
For each scanning session we detected the communities and associated activations (Fig. 3b-e). The estimated number of components range between 6–13 for the individual scanning fMRI sessions and in the following, for simplicity, we chose the median value (K = 9) for all scanning sessions (see Methods). We found that similar community patterns were present in different scanning fMRI sessions (Fig. 3b, d). This is clearly shown by the modular structure of the correlation matrix between all pairs of components of all sessions (Fig. 3f), rearranged according to cluster membership (cophenetic correlation coefficient: 0.69). Communities are transiently activated over time (Fig. 3c, e, top), such that, most of the time, the synchronization matrix is closer to one community or a superposition of few communities. The averaged duration of activation (i.e., sk>0.1, see Methods) across all communities is 11.80±0.20s. Moreover, we found that the sum of activations of all communities, noted S(t), consistently mirrors the temporal evolution of the order parameter (Fig. 3c, e, bottom), the correlation coefficient between S(t) and R(t) averaged over all sessions being 0.76 ± 0.03. These results suggest that switching among multiple synchronized clusters that last ~10 seconds underlie the temporal evolution of phase synchronization.
We further examined the topological organization of the communities across all scanning sessions. For this population analysis, we concatenated in time the tensors of the scanning sessions. To test robustness of the detected communities we divided the dataset in two halves. The selected number of components was equal to 14 for the two tensors constructed by concatenating each half of all scanning sessions, respectively (see Methods). We found that most of the communities are consistently present in both half-datasets (Fig. 4a). Furthermore, we found that the communities represent functional networks, including the default mode network, the somato-motor network, the visual network, the auditory network, the cognitive control networks, the self-referential network, and combination of these and others RSNs. This can be better appreciated by comparing the synchronization communities with the spatial components obtained using spatial Independent Component Analysis (ICA) on the same data (see Methods). We found that the synchronization communities project to specific ICA RSNs or combinations of RSNs (Fig. 4b), i.e., some communities project to one ICA RSNs, in particular the default-mode network, while others clearly show combination of different ICA RSNs. Such combinations suggest that activations of synchronization communities may be seen as concomitant activations of multiple RSNs. The results were consistent using the communities extracted from the two half-datasets.
To test the effect of using a different synchronization threshold for constructing the binary synchronization tensors, we compared the obtained communities (i.e., the vectors ak) with the threshold equal to π/4 to the ones obtained previously with the threshold equal to π/6 (S2 Fig.). We found that similar community patterns are consistently found for different synchronization thresholds.
To get further theoretical insights on the emergence of transiently synchronized networks among segregated brain regions, we examined a model of oscillators interconnected through the brain anatomical connections estimated using diffusion imaging on human subjects (see Methods). Here, we used a version of the Kuramoto model, which is the canonical model for studying synchronization phenomena [29]. Within this model, each node of the network is model by a phase oscillator, with an intrinsic frequency ωi in the 0.04–0.07Hz band (i = 1, …, n). The intrinsic frequencies were estimated from the data, as given by the averaged peak frequency of the narrowband BOLD signals of each brain region. The state of each oscillating node i is determined by its phase, φi(t), and the interaction between nodes depends both on the structural couplings and the phase difference between the nodes. The model has one single parameter, G, that represents the global scaling of the anatomical connectivity matrix.
The model dynamics were compared to the empirical data, for varying values of G. As G increases the mean order parameter <R> goes from the independent scenario (<R>≈n-1/2) to a scenario in which all oscillators are synchronized, and for G ranging from 0.2 to 0.3 the model reproduces the value observed in the empirical data, equal to <R> = 0.335 ± 0.016 (Fig. 5a). Within this parameter range, the model maximally approximates the distribution of phase differences Pr(Δφ) (Fig. 5b)—the fitting of the distribution was evaluated using the inverse of the Kullback-Leibler divergence (1/DKL) between distributions (see Methods). Moreover, also within this parameter range, the model maximally approximates the distribution of synchronized pairs of nodes, i.e., the distribution of the number N(t) of (i, j) pairs such that |φj(t)-φi(t)|<π/6, noted Pr(N) (Fig. 5c). In addition to the previous global phase statistics of the system of oscillators, we also test whether the model replicates the detail phase relations between oscillators, as given by the matrix of phase locking values (PLV, see Methods). We compared the PLV matrices of the model and the empirical data by calculating the Pearson correlation coefficient between corresponding elements of the upper triangular part of the two matrices. We found that the maximal agreement between the model and the data is achieved for the parameter range G = 0.1–0.3 (Fig. 5d). Finally, we examined the temporal evolution of phase synchronization by computing the peak frequency of the order parameter. We found that, for all values of G, the order parameter fluctuates slowly over time, with a frequency comprised between 43.10-5–47.10-5 Hz (Fig. 5e-f). Although this frequency range is lower than the one empirically observed, the model replicates an important feature of the data, namely that fluctuations of the order parameter are one order of magnitude slower than the frequencies of the oscillators. Hence, there exists a range of the parameter G for which the model consistently approximates several global and detailed spatiotemporal phase statistics of the empirical data.
We next examined which features of the above model (model 1) contribute to the generation of the different statistics. We first perturbed the topology of the anatomical connectivity matrix by generating surrogate connectivity matrices, the elements of which were taken randomly from the original distribution of connection weights, while keeping the distribution of ω as in model 1. This model (model 2) fairly approximates both <R> and Pr(Δφ), but it leads to significantly lower prediction of Pr(N) compared to model 1 and cannot fit the PLV matrix (Fig. 5a-e, gray trace). This indicates that the topology of the connectivity is an important feature of the model to approximate the observed phase statistics.
Second, we keep the connectivity as in model 1 but imposed homogeneity in the intrinsic frequencies ω by setting its value to 0.05 Hz for all nodes (model 3). Homogeneity leads to the trivial state of complete synchronization of all oscillators, thus destroying the rich dynamics observed in the original model (Fig. 5a-e, green trace). Thus, heterogeneity is also a key ingredient of the model for predicting the observed phase statistics.
Finally, we tested the alternative scenario in which heterogeneity is produced by stochastic forces. Noise is ubiquitous in neural systems and plays an important role in neural dynamics [30], it is thus important to examine whether the observed phase statistics can be modeled by noise-driven dynamics. For this, we imposed homogeneity in the intrinsic frequencies as above and we added uncorrelated white noise to the model (model 4). For sufficiently small global couplings, the noise prevents the network from reaching full synchrony. This control model consistently predicts <R>, Pr(Δφ), and the PLV matrix, but it leads to a lower prediction of Pr(N) and to slower variations (<2.10–3 Hz) of the order parameter than in model 1 (Fig. 5a-e, yellow trace). The same results were obtained in an exhaustive analysis in the {G-σ} plane, where σ is the noise intensity (see S3 Fig.). Thus, the deterministic model, for which transient synchronization arises as a consequence of heterogeneity in natural frequencies, explains the empirical observations better than the model with noise. This indicates that the dynamics are less likely to be noise-driven than to be produced by intrinsic heterogeneities in the nonlinear deterministic system. In conclusion, both topology and heterogeneity in the model are essential to approximate the statistics observed empirically.
We next tested whether the previous results are also found in frequency bands other than 0.04–0.07 Hz. For this, we band-pass filtered the fMRI data in different narrow frequency bands within 0.01 and 0.13 Hz, we calculated the phase statistics for each frequency band, and we tested the predictive power of the Kuramoto model using the corresponding distribution of intrinsic frequencies and the anatomical connectivity. We found that the model consistently approximates the empirical phase statistics within the same parameter range (G = 0.1–0.3) (S4 Fig.).
Altogether, the above results show that the empirical statistics are well described in the region of partial synchronization, the parameter range between disorder (asynchrony) and complete order (full synchrony). This parameter region is characterized by metastability (Fig. 6). Indeed, it has been argued that step-like increases of the order parameter as a function of the global coupling is an indication of metastability [31]. This is clearly shown by the behavior of the relative phase between two given nodes of the network model (Fig. 6a). For small values of the global coupling (G), the phases run practically independently. For intermediate values of G, quasi-phase-locking events are indicated by the deflections of the relative phase’s trajectory. Such events appear transiently and are separated by periods during which the phase-lock is lost. For high values of G, the nodes are synchronized, leading to a stable relation between their phases.
In the following, we tested whether the model can generate the observed transient synchronized clusters, as a result of transitions among the metastable states. This was done by constructing the interaction tensor of the model and applying the NNTF (Fig. 7). We found that fluctuations in the global synchronization result from intermittent activation of synchronized communities (Fig. 7a-b). Moreover, the correlation maxima between the model’s communities and the empirical communities indicate a significant similarity between the model and the empirical community structures, for G ranging from 0.15 to 0.5 (Fig. 7c).
In conclusion, the simple model used here emulates several temporal and spatial synchronization characteristics of the human resting-state BOLD activity.
Our results demonstrate that the brain’s spontaneous activity can be decomposed into synchronization networks, or communities, that transiently emerge and dissolve, giving rise to a global synchronization that fluctuates on a characteristic ultra-slow time scale (<0.01Hz). Consistent with our findings, resting-state functional connectivity fluctuations between 0.005 and 0.015 Hz have been reported using sliding window analysis of the Pearson correlation matrix of fMRI signals [19]. We used a combination of phase synchrony measuring and NNTF that allowed us to track the synchronization networks with enhanced temporal resolution and without the need of using sliding windows. We showed that such synchronized communities reoccur in time and across scanning sessions and that they relate to previously defined functional networks or combinations of those networks, including the default mode network and other sensory, motor, and cognitive networks [32].
Here, we demonstrated that the slow variation of synchronization is an emergent collective behavior in an anatomically-constrained network of oscillators. Notably, the model consistently i) generates the observed fluctuations of the global synchrony at an order of magnitude lower time scale than the frequencies of the oscillators, ii) generates the intermittent appearance of transient synchronized networks, and iii) approximates multiple empirical phase statistics within the same parameter range. In particular the model efficiently approximates the mean level of global synchrony, the phase differences distribution, and the distribution of synchronized nodes, for the same parameter range. Nevertheless, the model’s prediction of the pairwise phase relations, given by the PLV matrix, is moderate, albeit highly significant within the parameter range for which the aforementioned statistics are approximated. This imperfect fitting is expected principally because the diffusion imaging tractography we used poorly captures the inter-hemispheric connections and does not provide information about the directionality of the connections [33].
Previous studies have used a version of the Kuramoto model to describe the entrainment between the spontaneous oscillations of distant brain areas [27, 34]. In these studies, it was assumed that neural populations oscillate in the gamma range (>30Hz), since this frequency has been associated to the information processing at the local circuit scale [35]. Opposite to this, here we used a network of slow oscillators, with intrinsic frequencies distributed in the frequency range between 0.04–0.07 Hz—but note that the model robustly approximates the empirical statistics in other low frequency narrow bands within 0.01 and 0.13 Hz. The choice of low frequency oscillators is consistent with recent studies showing that the slow electrical activity and the spontaneous BOLD signal are closely linked and produce similar spatial correlation patterns [36, 37], suggesting a common neural mechanism for both signals. Moreover, it has been shown that the slow components of cortical potentials play an important role in the coordination of large-scale networks [36–38]. Note that the slow oscillators used in the present study, allow us to neglect the effect of conduction delays between the different brain regions, which are orders of magnitude faster—tens of milliseconds [39]—than the periods of the model oscillators. The model presented herein describes the interaction between the mesoscopic dynamics among the different brain regions. Whilst the Kuramoto model is a canonical model that captures the collective synchronization phenomena, it should not be taken as a detailed biophysical model. Nevertheless, the Kuramoto model has been shown to approximate any network of sustained oscillators (limit cycles), as soon as the oscillators interact with a sufficiently weak coupling in order to not destroy the limit cycles, in which case only the phases are perturbed [40]. Thus, it is possible to envision a more biophysically detailed model that, after analysing how phases are perturbed (phase response function, PRF), can be reduced to its phase dynamics—provided that the PRF retains at least the first two Fourier components of the original neural interaction [41]. How the mesoscopic oscillations relate to the microscopic dynamics and to the local connectivity in networks of spiking neurons will be the subject of future investigation. In particular, a more detailed model that produces self-sustained oscillations would be very useful for investigating how oscillation amplitudes affect the phase dynamics [42], a question that is not addressed in the present work.
The findings presented here show a direct link between time-varying functional resting-state connectivity and nonlinear dynamics embedded in the complex large-scale brain network. Time-varying network interactions have been studied in the context of adaptive networks, where the structure and the dynamics can co-evolve through, for example, synaptic plasticity [43]. In contrast, here we showed that temporal synchronization networks emerge from the interplay between nonlinear dynamics and the static network structure. Here, transient synchronization is a result of metastability and it does not require any type of short term plasticity, or other time dependence of the network connections, or even transmission delays. However, temporal variations of couplings that are mediated by synaptic plasticity should not be excluded at faster time scales, which are not accessible with fMRI—in particular, synaptic plasticity might play an important active role in orientating the switching between metastable states. Our results are consistent with a recent study [44] showing that fluctuations of the global level of synchrony in complex modular networks, such as the metabolic and the metropolitan transport networks, result from network topology. Indeed, the brain’s anatomical connectivity has small-world and modular attributes [33] that are thus responsible for the temporal variability of the global synchronization. This topological organization has been related to function, since it might play a key role in the integration of information across functionally segregated brain regions [33]. Along with network topology, heterogeneity is necessary in the generation of multiple synchronized clusters. Heterogeneity in frequency content of fMRI time-series in different brain regions might be related to functional segregation since frequency power is significantly different across functional brain networks [45]. Hence, our study suggests that network topology and heterogeneity are two functionally relevant features, implicated in integration and segregation, which are responsible for the observed time-varying functional connectivity at rest.
Previous work argued that integration and segregation are reconciled in the case of metastability [46]. Consistent with this view, the simple model used here indicates that metastability may underlie the observed transient synchronized clusters for which sets of brain regions engage and disengage in time. This phenomenon does not require noise to produce the transient wandering between synchronization clusters, but it is the result of heterogeneous natural frequencies and nonlinear interactions. Metastability is a functionally relevant scenario since it facilitates the exploration of a larger dynamical repertoire of the brain and allows for the all-around visitation of functional states and dynamic responses to the external world. Indeed, there is evidence that the brain tends to enlarge its repertoire of potential states, as the variance/entropy of BOLD signals positively correlates with chronological age [47]. Furthermore, this repertoire can be flexibly modified to adapt to the cognitive demands during tasks [48, 49]. Synaptic plasticity, neuro-modulation, and input gating may provide a mechanism for the brain system to adjust the exploration of potential states in a context-dependent manner. How these mechanisms interact with large-scale dynamics and metastability is an open question that requires further development of more biophysically realistic models.
This research was conducted in agreement with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and informed consent was obtained from all subjects before performing the study, in accordance with institutional guidelines. The study design was approved by the local Ethics Committee of Chieti University and the local Ethics Committee of Lausanne.
The resting BOLD activity was measured in 24 right-handed healthy young volunteers (15 females, age range 20–31 years). Subjects were informed about the experimental procedures and provided written informed consent. The study design was approved by the local Ethics Committee of Chieti University. Subjects lay in a supine position and viewed a black screen with a centered red fixation point of 0.3 visual degrees, through a mirror tilted by 45 degrees. Each volunteer underwent two scanning runs of 10 minutes in a resting-state condition. Specifically, they were instructed to relax, but to maintain fixation during scanning. The eye position was monitored at 120 Hz during scanning using an ISCAN eye tracker system. Scanning was performed with a 3T MR scanner (Achieva; Philips Medical Systems) at the Institute for Advanced Biomedical Technologies in Chieti, Italy. Functional images were obtained using T2-weighted echo-planar imaging (EPI) with blood oxygenation level-dependent (BOLD) contrast using SENSE imaging. EPIs (TR/TE = 2000/35 ms) comprised 32 axial slices acquired continuously in ascending order covering the entire brain (voxel size = 3×3×3.5 mm3). For each scanning run, initial 5 dummy volumes allowing the MRI signal to reach steady state were discarded. The next 300 functional volumes were used for the analysis. A three-dimensional high-resolution T1-weighted image (TR/TE = 9.6/4.6 ms, voxel size = 0.98×0.98×1.2 mm3) covering the entire brain was acquired at the end of the scanning session and used for anatomical reference. Initial data pre-processing was performed using the SPM5 software package (Wellcome Department of Cognitive Neurology, London, UK) running under MATLAB (The Mathworks, Natick, MA). The pre-processing steps involved the following: (1) correction for slice-timing differences (2) correction of head-motion across functional images, (3) co-registration of the anatomical image and the mean functional image, and (4) spatial normalization of all images to a standard stereotaxic space (Montreal Neurological Institute, MNI) with a voxel size of 3×3×3 mm3. Furthermore, the BOLD time series in MNI space were subjected to spatial Independent Component Analysis (ICA) for the identification and removal of artifacts related to blood pulsation, head movement and instrumental spikes [7]. The BOLD artifact removal procedure was performed using the GIFT toolbox (http://mialab.mrn.org/software/gift/index.html). No global signal regression was performed. Finally, for each recording session (subject and run), we extracted the mean BOLD time series from the 66 brain regions of the brain atlas [33].
The anatomical connectivity matrix was estimated using Diffusion Spectrum Imaging (DSI) data and tractography from five healthy right-handed male human subjects [33]. The experimental design was approved by the local ethics committee at University of Lausanne. The grey matter was subdivided in 33 cortical regions per hemisphere (n = 66 areas in total) according to anatomical landmarks (same as for the BOLD signals). This was followed by a further parcellation into 998 regions of interest (ROIs). White matter tractography was used to estimate the fiber tract density connecting each pair of ROIs, averaged across subjects. Anatomical connectivity among the 66 cortical regions was calculated by summing all incoming finat strengths to the target region, and dividing it by its region-dependent number of ROIs, resulting in a non-symmetric connectivity matrix. The normalization by the number of ROIs—which have approximately the same surface on the cortex, i.e. the same number of neurons—is required because neuronal activity is sensitive to the number of incoming fiber per neuron in the target region. The connection of a region to itself was set to 0 in the connectivity matrix for the simulations.
We extracted the phases of the fMRI time series of each of the n brain regions (n = 66). Following [26], the BOLD signals were first band-pass filtered within the narrowband 0.04–0.07 Hz. This frequency band has been mapped to the gray matter and it has been shown to be more reliable and functionally relevant than other frequency bands [1, 26, 50]. The Hilbert transform (HT) was applied to the filtered BOLD signals to obtain the associated analytical signals. The analytic signal represents a narrowband signal, a(t), in the time domain as a rotating vector with an instantaneous phase, φ(t), and an instantaneous amplitude, A(t), i.e., a(t) = A(t)cos(φ(t)). The phase and the amplitude are given by the argument and the modulus, respectively, of the complex signal z(t), given by z(t) = a(t)+j.HT[a(t)], where j is the imaginary unit. Note that narrowband filtering is a requirement for obtaining meaningful phases and envelopes through the HT [26].
The global level of phase synchrony was quantified by the order parameter, R(t), given by:
R(t)=|1n∑i=1nejφi(t)|
(1)
Thus, R is the average phase of the system and takes the values 0 and 1 for the completely asynchroneous and completely synchronized cases, respectively. For a finite number n of independent phases, the expected value of R is equal to n-1/2. To quantify the pairwise phase relation between two given brain regions k and l, we calculated the phase-locking value (PLV), given as:
PLV(k,l)=|1T∑t=1Tej[φk(t)−φl(t)]|
(2)
Thus, the phase locking value ranges from 0 (complete phase independence) to 1 (perfect phase locking: φk(t)–φl(t) = const., for all t).
Phase shuffled surrogates were used to assess the significance of synchronization. For this, the Fourier transform (FT) of the original signals was first compute and the phase of the Fourier coefficients was substituted with uniformly distributed random numbers while preserving their modulus. Second, the inverse FT was applied to return to the time domain. This procedure effectively randomizes the phases of the signals while preserving the same power spectra as the original series. Specifically, let xi(t) be the original BOLD signal from region i (i = 1, …, n and t = 1, …, T). The discrete Fourier transform of xi is given by:
x˜i(k)=∑t=1Txi(t)e−j2πktT
(3)
Where j is the imaginary unit and k = 1, …, T. A realization of the phase randomized surrogate is given by:
xisurr(t)=1T∑k=1T|x˜i(k)|e−j(2πktT+φr)
(4)
Where φr is a random variable uniformly distributed between—π and π. Finally, the surrogate time series were band-pass filtered within the narrowband 0.04–0.07 Hz and the global phase synchronization (order parameter) was computed.
We used these surrogates i) to evaluate the global level of phase synchrony (R) on the null hypothesis of null synchronization and ii) to correct the bias of the PLV matrix by subtracting to each entry of the matrix PLV(k, l) the expected value when the phases of the time-series k and l are phase-randomized (estimated using 103 randomizations for each pair (k, l)).
We evaluated the contribution of the amplitudes and the phases to the observed correlations among the BOLD signals. For this, we compared the correlation matrix among the narrowband signals, the correlation matrix among the signal amplitudes and the phase-locking matrix of the signal phases. Amplitudes and phases were extracted from the analytic signals, a(t) = A(t)cos(φ(t)), where A(t) is the instantaneous amplitude and φ(t) is the instantaneous phase. Phase-locking values were given by Equation (2). The Pearson correlation coefficients were computed for all pairs of narrowband signals and for all pairs of amplitudes of the analytical signals that were transformed by taking logarithm of the squared amplitude envelopes to render the amplitudes statistics more normal before calculating the correlations [51]:
rsignal(i,j)=∑t=1T(xi(t)−〈xi(t)〉)(xj(t)−〈xj(t)〉)∑t=1T(xi(t)−〈xi(t)〉)2(xj(t)−〈xj(t)〉)2,
(5)
ramp(i,j)=∑t=1T(A^i(t)−〈A^i(t)〉)(A^j(t)−〈A^j(t)〉)∑t=1T(A^i(t)−〈A^i(t)〉)2(A^j(t)−〈A^j(t)〉)2
(6)
Where xi(t) is the original narrowband signal from region i (i = 1, …, n and t = 1, …, T), Â is the square-logarithmic transform of A, and <.> denotes the average over time.
For each pair of brain regions, we calculated the narrowband signals’ correlations (rsignal), amplitudes’ correlations (ramp) and phase-locking values (PLV), averaged across scanning sessions, and compared them (S1 Fig.). Note that the relation between both rsignal and ramp and between rsignal and PLV is not linear (S1 Fig.). To quantify how well the interactions among amplitudes and the interactions among phases represent the interactions among signals, we evaluated the uncertainty reduction (Δ) about the narrowband signals’ correlations given the amplitude correlations or given the phase-locking values. This was done by computing the normalized mutual information MI between the different measures, given by:
Δ(y)=MI(rsignal,y)H(rsignal)=1H(rsignal)[H(rsignal)+H(y)−H(rsignal,y)]
(7)
Where H denotes the entropy and y represents either ramp or PLV. We found that the uncertainty reduction is slightly higher for the phase interactions than for the amplitude interactions (Δ(PLV) = 0.37; Δ(ramp) = 0.21). When comparing individual scanning sessions we found that Δ(PLV) is significantly (p<10–10, t-test) higher than Δ(ramp). Thus, phase interactions convey more information about the signals’ correlations than do the amplitudes’ correlations. This difference is mainly due to the fact that only high narrowband signals’ correlations are well reflected in the amplitude correlations, while weak and moderate narrowband signals’ correlations are associated to near zero amplitude correlations. Hence, phase interactions are a good description of the interactions of the narrowband signals and, importantly, allow for a time-resolved analysis of interactions, as shown below.
We detected the community structure of temporal synchronization networks by using a non-negative tensor factorization (NNTF) approach, based on the canonical decomposition procedure, which can be seen as a higher-order analogue of the Principal Component Analysis (PCA) [52]. This approach has successfully been applied to detect the community structure of temporal networks [28]. Unlike PCA that uses a vector-based representation, tensor factorization represents the spatial interaction matrices within the network’s nodes as a 3-dimensional space-time tensor and seeks for the most parsimonious decomposition of the tensor onto sum of simpler tensors. Tensor factorization preserves the two-dimensional character of spatial interactions (something that is lost when interaction matrices are vectorized in a PCA) and extracts the temporal coherence of the spatial patterns.
A synchronization matrix Q was built at each time step t (t = 1, .., T) by calculating the phase difference of each pair of empirical analytic signals (or phase oscillators in the model) and imposing a synchronization threshold that, unless otherwise specified, was equal to π/6, i.e.: Qij(t) = 1 if |φj(t)–φi(t)|<π/6 and Qij(t) = 0 otherwise (1 ≤ i, j ≤ n). In this way, we obtained a three-dimensional matrix, or tensor T, of size n×n×T, with values equal to 0 or 1, given by T(i, j, t) = Qij(t). To eliminate the possibly accidental phase synchronizations, a link (i, j) was set to zero, for all time steps, if the phases of signals i and j was such that |φj(t)–φi(t)|<π/6 for less than 20% of the time steps. The tensor T contains the topological and temporal information of phase synchronization.
The tensor T was decomposed into K rank-one tensors in the form:
Τ=∑k=1Kak∘bk∘ck
(8)
Where ak, bk, and ck are vectors of dimension n, n, and T, respectively, and “о” represents the outer product of vectors, defined as uov = u.vT, where the superscript T denotes the transpose. The aim of the tensor canonical decomposition is to find the set of factor matrices, A = [a1 … aK], B = [b1 … bK], and C = [c1 … cK] that best approximates T (Fig. 3a). In particular, here we imposed a non-negative constrain on the factor matrices as this provides an additive representation of the tensor in terms of the factors, thus providing a physically meaningful interpretation of the decomposition [53]. The optimization problem is equivalent to minimize the difference between T and the approximation, given the non-negative constrains, i.e.:
min‖Τ−ΤA,B,Capprox‖2,with A,B,C≥0
(9)
Where ‖.‖ represents the Frobenius norm and ΤA,B,Capprox=∑k=1Kak∘bk∘ck. We used the recently proposed block principal pivoting method (www.cc.gatech.edu/~hpark/nmfsoftware.php) to achieve this optimization [54].
Since at each time step t, the matrix Q is symmetric, we have A = B. The community structure of the network and the temporal activation of the different communities are contained in the matrices A and C, respectively. The column vectors of A represent the K communities, with ak(i) being the associated participation weight of the node i in the community k. The column vectors of C represent the activity level of each community, i.e., ck(t) is the activation level of community k at time step t. The strength of a community k, noted sk(t), can be calculated as a combination of the level of activation of this community and its total participation weight [28]:sk(t)=ck(t)∑i=1nak(i).
Finally, the number of components was selected using the DIFFIT method [55]. This procedure calculates the goodness-of-fit for each K, given by F(K)=1−‖Τ−ΤA,B,Capprox(K)‖/‖T‖, and detects the number of components K after which the function F enters into a plateau by selecting the value that maximize the function: DIFIT(K) = [F(K)–F(K–1)]/F(K+1). The selected number of components range between 6–13 for the individual scanning fMRI sessions (median: 9). For simplicity, we chose the median value (K = 9) for all scanning sessions. The selected number of components was equal to 14 for the two tensors constructed by concatenating each half of all scanning fMRI sessions, respectively. Finally, in a second analysis we changed the synchronization threshold from π/6 to π/4 (S2 Fig.); the selected number of components using a threshold equal to π/4 was equal to 15 and 16 for the two tensors constructed by concatenating each half of all scanning fMRI sessions, respectively.
We compared the synchronization communities obtained using the NNTF with the spatial components obtained using spatial Independent Component Analysis (ICA). Spatial ICA is a standard technique that clusters the data into maximally spatially independent patterns of coherent fMRI activity [2, 32], i.e., voxels belonging to a given ICA pattern have higher temporal correlations among themselves than with voxels belonging to other ICA patterns. Here we used the GIFT toolbox to perform the ICA decomposition for each scanning session. The estimation of the number of independent components (ICs) was performed using the minimum description length criterion [2]. Each IC consisted of a spatial pattern of activity (z-score) and a corresponding time-course of the spatial pattern. Self-organizing group ICA (sogICA) [56] was used to average the ICs extracted from single scanning fMRI datasets. Fourteen ICs were identified, ICs were for the most part consistent, and were present in at least 75% of the subjects [32]. The ICA patterns at the voxel level were down-sampled onto the same parcellation of 66 cortical regions used in the present study by averaging the results among the voxels belonging to each cortical region. The ICA patterns were associated to different functional networks which are often seen in resting-state fMRI activity and that consist of sets of brain regions known to be engaged in sensory-motor or cognitive function, called resting-state networks (RSNs) [32]. The RSNs include the default mode, core, somatomotor, dorsal/ventral attention, vision, auditory, self-referential, language, cognitive control, and working memory networks. For further details about the ICA procedure see [32].
For comparison, the synchronization communities and the ICA-based RSNs were converted to 66-dimensional binary vectors by imposing a threshold, equal to 0.1, above which the elements were set to 1 and, otherwise, they were set to 0. As a measure of similarity between the binarized synchronization communities and the binarized ICA RSNs we used the Jaccard index. The Jaccard index is defined as the proportion of nonzero coordinates that are equal in two given binary vectors.
We considered a simple model of a network of n coupled phase oscillators, in which connections are determined by the anatomical connectivity matrix that was estimated using DSI (n = 66). The Kuramoto model is considered a canonical model of synchronous oscillations in many systems [57]. This model assumes that the oscillators interact with each other through their phase differences. Let φi(t) be the phase of the i-th oscillator (i = 1, …, n) at time t. The time evolution of the phases is governed by the following set of coupled differential equations:
dφidt=ωi+G∑j=1nCijanatsin(φj−φi)
(10)
Where ωi is the natural frequency of the i-th oscillator, Canat is the n×n anatomical coupling matrix, and the parameter G represents the global coupling strength. The interaction between two given oscillators i and j is modulated by the sinus of the phase difference, sin(φj—φi). Such interaction tends to synchronize the oscillators since an oscillator lagging behind another one (φj—φi>0) is speed up, whereas an oscillator leading another (φj—φi<0) is slow down. The model was numerically integrated using the Euler’s method with a time step equal to 0.01, equivalent to 10ms. The total number of simulation steps was 12.105 and the first 5.105 steps were discarded to remove the transient period.
In addition, we considered the effect of adding noise to the anatomically-connected homogeneous Kuramoto model. This model (called model 4 in the following) assumes that all the oscillators have the same intrinsic frequency ω0 = 0.05 Hz and receive independent uncorrelated white noise. The time evolution of the phases is governed by the following set of coupled stochastic differential equations:
dφi(t)dt=ω0+ξi(t)+G∑j=1nCijanatsin(φj(t)−φi(t))
(11)
Where i = 1, …, n, Canat is the n×n anatomical coupling matrix, scaled by the global coupling strength G, and ξi is uncorrelated white noise, i.e., <ξi(t)> = 0 and <ξi(t)ξj(t’)> = σ2δ(t’–t)δij, where σ is the noise amplitude. Noisy fluctuations can be interpreted as the modification of the phase of each oscillator due to the underlying neural noise, which is a consequence of random fluctuations of the background activity and inherent to neural networks of finite size [58].
We use the Kullback-Liebler divergence, DKL, to evaluate how well the Kuramoto model describes the empirical probability distribution of phase difference, Pr(Δφ), and the empirical probability distribution of the number of synchronized pairs, Pr(N). Given an observed (empirical) probability distribution g and a model probability distribution f, DKL measures the dissimilarity between the two distributions. It is defined to be:
DKL(g,f)=∫g(x)log(g(x)f(x)) dx
(12)
The smaller DKL means that the model distribution f is closer to the empirical distribution g. We used the inverse of DKL as the goodness-of-fit of the model.
|
10.1371/journal.pntd.0005890 | Control of Phlebotomus argentipes (Diptera: Psychodidae) sand fly in Bangladesh: A cluster randomized controlled trial | A number of studies on visceral leishmaniasis (VL) vector control have been conducted during the past decade, sometimes came to very different conclusion. The present study on a large sample investigated different options which are partially unexplored including: (1) indoor residual spraying (IRS) with alpha cypermethrin 5WP; (2) long lasting insecticide impregnated bed-net (LLIN); (3) impregnation of local bed-nets with slow release insecticide K-O TAB 1-2-3 (KOTAB); (4) insecticide spraying in potential breeding sites outside of house using chlorpyrifos 20EC (OUT) and different combinations of the above.
The study was a cluster randomized controlled trial where 3089 houses from 11 villages were divided into 10 sections, each section with 6 clusters and each cluster having approximately 50 houses. Based on vector density (males plus females) during baseline survey, the 60 clusters were categorized into 3 groups: (1) high, (2) medium and (3) low. Each group had 20 clusters. From these three groups, 6 clusters (about 300 households) were randomly selected for each type of intervention and control arms. Vector density was measured before and 2, 4, 5, 7, 11, 14, 15, 18 and 22 months after intervention using CDC light traps. The impact of interventions was measured by using the difference-in-differences regression model.
A total of 17,434 sand flies were collected at baseline and during the surveys conducted over 9 months following the baseline measurements. At baseline, the average P. argentipes density per household was 10.6 (SD = 11.5) in the control arm and 7.3 (SD = 8.46) to 11.5 (SD = 20.2) in intervention arms. The intervention results presented as the range of percent reductions of sand flies (males plus females) and rate ratios in 9 measurements over 22 months. Among single type interventions, the effect of IRS with 2 rounds of spraying (applied by the research team) ranged from 13% to 75% reduction of P. argentipes density compared to the control arm (rate-ratio [RR] ranged from 0.25 to 0.87). LLINs caused a vector reduction of 9% to 78% (RR, 0.22 to 0.91). KOTAB reduced vectors by 4% to 73% (RR, 0.27 to 0.96). The combination of LLIN and OUT led to a vector reduction of 26% to 86% (RR, 0.14 to 0.74). The reduction for the combination of IRS and OUT was 8% to 88% (RR, 0.12 to 0.92). IRS and LLIN combined resulted in a vector reduction of 13% to 85% (RR, 0.15 to 0.77). The IRS and KOTAB combination reduced vector densities by 16% to 86% (RR, 0.14 to 0.84). Some intermediate measurements for KOTAB alone and for IRS plus LLIN; and IRS plus KOTAB were not statistically significant. The bioassays on sprayed surfaces or netting materials showed favourable results (>80% mortality) for 22 months (IRS tested for 12 months). In the KOTAB, a gradual decline was observed after 6 months.
LLIN and OUT was the best combination to reduce VL vector densities for 22 months or longer. Operationally, this is much easier to apply than IRS. A cost analysis of the preferred tools will follow. The relationship between vector density (males plus females) and leishmaniasis incidence should be investigated, and this will require estimates of the Entomological Inoculation Rate.
| Integrated vector management (IVM) is of the important elements in the Regional Visceral Leishmaniasis (VL) elimination strategy. VL supposed to be eliminated from the Region (Bangladesh, India and Nepal) by 2015 which is extended up to 2017. There are several factors are responsible for not achieving the elimination goal and IVM is one of them. In Bangladesh, currently VL vector control activity is confined to indoor residual spraying with deltamethrin. Therefore, it is crucial to identify alternative VL vector control method(s) to enhance VL elimination programme. In the present study we have investigated the efficacy of (1) indoor residual spraying (IRS) with alpha cypermethrin 5WP; (2) long lasting insecticide impregnated bed-net (LLIN); (3) impregnation of local bed-nets with slow release insecticide K-O TAB 1-2-3 (KOTAB); (4) insecticide spraying in potential breeding sites using chlorpyrifos 20EC (OUT) and different combinations of the above. Study findings showed that the combination of LLIN and OUT was the most effective method against VL vector and among single interventions, IRS with two rounds of spraying found efficacious. In addition to IRS, VL elimination programme should consider LLIN plus OUT for vector control in relation to IVM.
| Visceral leishmaniasis (VL) [known as kala-azar in the Indian sub-continent] is a parasitic disease present in South-East Asia since before the early 1800’s [1]. VL appears to have spread along the Ganges and the Brahmaputra rivers, the major transport routs of Bengal and Bangladesh. In this area, VL was first described in 1824 in the Jessore district where about 75,000 people died [2]. An intensive control programme aimed at the eradication of malaria was mounted in the late 1950s and early 1960s throughout the South Asian sub-continent with the main effort based on indoor residual spraying (IRS) of DDT (Dichlorodiphenyltrichloroethane). During the malaria eradication programme the incidence of VL dropped dramatically as a collateral benefit with DDT spraying [3]. However, within a few years after the end of the Malaria eradication effort, VL returned to Bihar and Bengal on both sides of the borders of India and Bangladesh [4].
In Bangladesh, five districts, Sirajgang, Pabna, Mymensingh, Rajshahi and Tangail were more affected following the end of the malaria eradication programme to 1980s [5]. These districts continue to have the highest number of cases along with other districts reporting few cases. The Malaria and Vector Borne Disease Control Unit, Directorate General of Health Service (DGHS), Government of Bangladesh has reported 109,226 VL cases including 329 VL related deaths from 1994 to 2013 [6]. Mymensingh district alone contributed about 50% of total reported cases and the highest number of cases was reported from the Fulbaria upazila (sub-district) with several other upazilas in Mymensingh reporting cases [6]. To combat VL in the Indian Sub-continent, a common platform was developed through signing a memorandum of understanding (MoU) by the health ministers from the three affected countries (Bangladesh, India and Nepal) in 2005 [7]. In the MoU, a target was set to reduce the VL incidence to less than 1 case per 10,000 population at the sub-district (upazila in Bangladesh) level by 2015 [7]. This MoU has been extended up to 2017 with inclusion of Bhutan and Thailand in the group [8]. Integrated vector management is one of the most important pillars in the elimination strategy, however virtually no vector control activities were under taken in Bangladesh for VL vector control until 2010 [6,9].
Since 2011, the National Kala-azar Elimination Programme (NKEP) in Bangladesh conducted IRS for vector control using deltamethrin 5 WP in the affected communities. In addition to IRS, two commercially manufactured long lasting impregnated bed-nets (LLIN) were given to each patient who was treated in the government hospitals for the last three years (2011 to 2013). The IRS (using deltamethrin 5WP) and LLIN were associated to decrease the level of the Phlebotomus argentipes sand fly by 70–80% in Bangladesh [10]. In India and Nepal, deficiencies in the quality of IRS was observed when carried out by the national programme [11]. Mosquito nets impregnated with a slow release insecticide (K-O TAB 1-2-3; deltamethrin with a binder) resulted in a 65% reduction in sand fly levels in Bangladesh [12]. None of the studies however determined the effect of both LLIN and insecticide treated local nets in a single study. VL vectors usually breed in the shady places with loose soil where enough moisture is available around the houses [13]. There is lack of evidence in Bangladesh for controlling immature stages of P. argentipes sand fly by applying insecticide spraying on their breeding habitat around the houses. Hypothetically, it is believed that the combined application of two vector control methods will increase the effect on VL vectors but currently there is no evidence for this in South-East Asia.
The present study on single and combined vector control interventions was conducted to assist the NKEP by identifying suitable VL vector control method(s) to achieve the elimination target within the set time frame and maintain a low vector density during the maintenance phase of the VL elimination programme.
The sample size estimation was based on the vector densities (female and male P.argentipes sand fly counts per household), variations and distributions documented in previous entomological studies and sand fly reduction rates in similar intervention studies performed in Venezuela; Bangladesh, India and Nepal [19,20]. Based on our previous experience, male:female ratio was always about 50:50 throughout the year [10,20], and so we assume that joint counts will not affect the outcome of the analysis, but will produce more accurate estimate through regression model. We assumed that the distribution of sand fly counts would follow a negative binomial distribution with a dispersion coefficient of k = 0.05 and an intra-cluster coefficient of 0.03, a reduction from 20 to 5 vectors per trap/ night, and an average of 50 households per cluster. The minimum sample size was found to be 6 clusters per intervention arm, with a total of 60 clusters in the study to achieve 80% power and a significance level of 5%.
VL surveillance data for three years (2009–2011) was collected from Upazila Health Complex (UHC), Fulbaria. Based on the passive surveillance reports, endemic villages were identified (containing 300 to 600 households [HH]) and were selected where the national programme was not conducting routine vector control activities. The following villages were selected (Fig 1): Mahespur in Bhabanipur union (311 HHs); Mandolbari and Chalkgarbajail in Balian union (319 HHs); Patira in Kaladah union (323 HHs); Hurbari in Kaladah union (297 + 312 = 609 HHs); Dulma in Enayetpur union (309 HHs); Kathgarh in Naogaon union (307 HHs); Bisania and Natuapara in Kushmail union (300 HHs); Anuhadi in Rangamatia union (322 HHs); and Haripur in Rangamatia union (302 HHs), in total 3079 HHs). We had no indication or evidence that the study villages were atypical. The study was conducted from September 2012 to October 2014.
Based on their high endemicity levels, 11 villages were selected from seven unions. Eleven villages were divided into 10 sections with a minimum of 300 HHs each (Fig 2).
Each section with at least 300 HHs was divided into six clusters, each with about 50 HHs. In each study cluster, all HHs were numbered using enamel paint on the front door. There was a minimum of 50 meters distance between two clusters to avoid cross-contamination of the interventions. The total number of 60 clusters (10 sections x 6 clusters each) was assigned for implementation of interventions (Fig 2).
Five HHs from each cluster (5 HHs x 60 clusters = 300 HHs) were selected by simple random sampling for measuring sand fly densities at 3 weeks before intervention (baseline survey) and at 2, 4, 5, 7, 11, 14, 15, 18 and 22 months after intervention (follow-up surveys).
In order to achieve homogenous and comparable groups of clusters, the following approach was undertaken. Based on vector (P. argentipes) density during the baseline collection, the 60 study clusters were stratified into 3 groups: (i) high [range of P. argentipes– 59 to 143], (ii) medium [range of P. argentipes– 31 to 57] and (iii) low [range of P. argentipes– 1 to 30] vector density. Each group (vector density stratum) had 20 clusters. From these 60 clusters (20 clusters in each group), the different intervention and control arms were randomly selected and each arm had 6 clusters [about 300 HHs each] (Fig 2).
After the formation of the clusters, all HH heads were interviewed by trained field staff (Research Assistants [RA]) using a structured questionnaire to record the HH information, socioeconomic characteristics and some epidemiological information.
A standard data entry interface was designed using Microsoft Office Access for entering the study data. Data were checked and cleaned before analysis. Descriptive analysis was performed to determine the nature of the data. The main analysis was based on P. argentipes sand fly counts per HH collected by CDC light traps over two nights. No zero count was removed. All CDC traps worked perfectly. The average sand fly density among different interventions and control groups at baseline as well as follow-up time points was determined. Mean P. argentipes sand fly count between control and intervention areas were compared using a non-parametric approach (Mann Whitney U test). It was found that the negative binomial distribution fitted the data and all analyses were performed under that assumption. A generalized estimating equation (GEE) modelling technique was used to adjust for data correlations due to the longitudinal/ repeated measurements in cluster sampling. An interaction term for the intervention arm at follow-up was included in this model in order to estimate the effect of the intervention.
Technically, the regression model had the following structure:
Count=Intercept+a*Treatment+b*Time+c*Interaction+error
where treatment is one if it is the intervention and zero if it is the control; where time is one if follow up and zero if baseline; and where interaction is one if the intervention group at follow up. Intervention effect was measured using incidence rate ratio (IRR) of P. argentines sand fly count generated from the exponent of c-coefficient in the model. In the tables, IRR represented as the rate ratio (RR) and its p-value are given. Significances stated at 5% level and 95% confidence intervals are given. The main outcome variable was “P. argentipes sand flies per household” at before and 2, 4, 5, 7, 11, 14, 15, 18 and 22 months after the intervention. The following variables were controlled for in the full model: cattle shed, family members sleep in the bed room, number of bed nets in the house, type of house wall, presence of crack in wall and socio economic status. Economic status of the household was measured through the HH asset index. Household asset index was generated by the principal component method in factor analysis using the following variables: electricity, radio, television, mattress, bed net, motor cycle, bicycle, van, power tiller, shallow machine, chair/table, mobile phone, clock, sewing machine, and fishery. We categorized the index as low (score less than 33th percentile), medium (score between 33th to 66th percentile) and high (score greater than 66th percentile). We did not perform any sensitivity analysis as the study objective was to compare the efficacy of different vector control interventions against control arm. All analyses were performed by using STATA 10.1.
Investigators and external experts conducted all training activities to ensure the quality of the training. All study activities were monitored by the investigators to maintain the quality.
The study was approved by Bangladesh Medical Research Council (BMRC). Written informed voluntary consent was obtained from the HH heads/responsible family members before conducting any study related activities. All control (no intervention) arm HHs were donated one LLIN per family after completing the study.
The study was conducted in 3079 HHs with a population of 13,406 inhabitants in 10 sections of 60 clusters (Fig 2). Nine types of interventions were tested and one control arm. Within the study population, 49.4% were female and 39.6% were below 17 years of age (Table 1). About 5% and 0.6% of the total population had a past history of VL and PKDL respectively. The proportion of VL and PKDL varied from 3.6% to 5.8% and 0.1% to 1.1% respectively among the different study arms (Table 1). About 60% (1846/3079) of HHs had a cattle shed. The percentage of the population with cattle sheds varied from 53.5% (161/301) to 69.5% (214/308) among the different study arms. Almost all HHs had a non-impregnated bed net (99.1%) with an average of 2.33 per HH (SD = 1.35). About 30% of houses had precarious walls and about 15.0% had cracks in their walls. The study HHs were almost equally distributed among low (33.4%) medium (30.7%) and high (35.6%) asset index groups (Table 1).
A total 17,434 sand flies were collected during the entire study period including baseline and the 9 follow-up surveys (Table 2). Of all sand flies, 53.75% were P. argentipes and the remainder other species. Among P. argentipes, 4,443 (47.42%) were female including 16.01% gravid. At baseline (before implementation of intervention), 3,616 sand flies were captured of which 78.32% were P. argentipes. There were 83 (18.12%), 412 (36.52%), 1315 (42.57%), 1443 (58.52%), 383 (55.59%), 67 (35.26%), 331 (55.72%), 1419 (37.92%) and 1085 (74.21%) P. argentipes sand flies collected respectively in the first to ninth follow up measurements. There is no significant difference between male and female ratio of collected P. argentipes sand flies throughout the study period (S1 Table).
At baseline, the average P. argentipes density per household was 10.57 (SD = 11.51) in the control arm and 7.3 (SD = 8.46) to 11.53 (SD = 20.17) in the different intervention arms (Table 3). The difference of P. argentipes sand fly densities among intervention arms and control arm were not statistically significant at baseline except for the KOTAB intervention arm (p = 0.032). However, P. argentipes sand fly densities in most of the intervention arms were significantly lower than the control arm at different follow-up measurements except OUT (Table 3). Fig 3 shows that the mean P. argentipes sand fly density was always below the values in the control arm throughout all the measurements.
The efficacy of the interventions was measured through the reduction of P. argentipes sand fly densities in intervention HHs compared to the control HHs. The bioavailability of insecticides on treated surfaces (indicating how long the insecticide was capable of killing insect vectors) was determined using bioassay tests.
The Abbot corrected sand fly mortality at 24 hours of exposure on treated surfaces was as follows: in both cycles of IRS, the mortality was above 80% (which is the threshold level) even after 5 months following spraying (Fig 5). The mortality for LLIN was 82.59% at 20 months of use. The mortality on K-O TAB 1-2-3 impregnated nets dropped from 88.37% at 3 months to 69.12% at 20 months after use.
Our cluster randomized controlled trials on VL vector interventions are to our knowledge the largest ever conducted in the South-East Asia Region. In the present study a large number of sand flies was captured of which 45.9% were other than P. argentipes species which is three times higher than in a previous study conducted in the same sub-district [10]. We have tested four individual types of interventions and five combinations against VL vectors. In the present study, we used alpha cypermethrin 5WP for IRS as it is less expensive than deltamethrin and as efficacious as other pyrethroids [10,20]. IRS is however challenging in terms of operational complexity and cost and it is difficult to maintain a uniform quality of spraying. It was observed in India and Nepal that when IRS was applied by the research team under well controlled conditions, it was found to be very effective against VL vectors but when it was delivered by the national programme the efficacy dropped significantly [11].
At the current stage of the VL elimination initiative in the Indian subcontinent, Nepal has reached the target of less than one case per 10,000 population, Bangladesh has only a few sub-districts (upazilas) above this threshold and India has reached the goal in many areas but is still facing elevated VL endemicity in a number of districts [30]. In sub-districts where the final push towards elimination is still required, IRS alone or in combination with other measures are still needed. In sub-districts where VL/PKDL cases appear sporadically and case numbers are below the thresholds, new ways of active case detection (to reduce the transmission) and vector control (to prevent transmission) are required.
Regarding vector management in the post-elimination phase, recent studies have shown the potential of different vector control tools, including insecticide treated durable wall lining (DWL), commercially impregnated long lasting insecticidal nets (LLIN), slow release insecticides (K-O TAB 1-2-3,) treatment of existing bed nets (ITNs), as well as insecticidal paint (Inesfly company, Valencia, Spain). Prospects and limitations of these products include the following:
Our study contributes important information to what is known already. (i) The combination of different approaches leads to better results than single approaches in reducing the vector population. (ii) The combination of a chemical intervention in breeding and larval-development sites in and around rural houses together with measures against adult vectors using LLINs was particularly successful in significantly reducing vector populations for at least 22 months. Furthermore, this measure can be used in remote areas where sporadic cases appear as it lends itself to community actions. The use of ITNs would be even more feasible as it is independent of LLIN donations. Local bed nets to prevent mosquito bites are common in rural Bangladeshi communities and over 90% of HHs have nets [31,35]. Slow release insecticide impregnated bed nets might be a good alternative to prevent sand fly bites but the effect on the vector population was shorter and less marked compared to LLINs.
It will be a remarkable innovation if applications of insecticides in breeding places of sand flies around houses in endemic communities are able to reduce the vector density. In the current study, we tested chlorpyrifos 20EC to control immature stages of sand flies because this insecticide has no/very limited side effects on the environment or on human health [36]. The result following the first round of chlorpyrifos spraying was not promising but after the second round of spraying, it was effective in reducing the vector densities. This effect was considerably enhanced when combined with the treatment of bed nets.
In conclusion: The combination of LLIN and OUT (outdoor spraying of vector breeding sites) was the most efficacious measure among the different tools tested. This combination measure could play an important role during the maintenance phase of the VL elimination programme to maintain a low vector density particularly in remote areas where the community can take care of the measure. The relationship between vector density (males plus females) and leishmaniasis incidence should be investigated, and this will require estimates of the Entomological Inoculation Rate.
|
10.1371/journal.ppat.1002233 | Impairment of Immunoproteasome Function by β5i/LMP7 Subunit Deficiency Results in Severe Enterovirus Myocarditis | Proteasomes recognize and degrade poly-ubiquitinylated proteins. In infectious disease, cells activated by interferons (IFNs) express three unique catalytic subunits β1i/LMP2, β2i/MECL-1 and β5i/LMP7 forming an alternative proteasome isoform, the immunoproteasome (IP). The in vivo function of IPs in pathogen-induced inflammation is still a matter of controversy. IPs were mainly associated with MHC class I antigen processing. However, recent findings pointed to a more general function of IPs in response to cytokine stress. Here, we report on the role of IPs in acute coxsackievirus B3 (CVB3) myocarditis reflecting one of the most common viral disease entities among young people. Despite identical viral load in both control and IP-deficient mice, IP-deficiency was associated with severe acute heart muscle injury reflected by large foci of inflammatory lesions and severe myocardial tissue damage. Exacerbation of acute heart muscle injury in this host was ascribed to disequilibrium in protein homeostasis in viral heart disease as indicated by the detection of increased proteotoxic stress in cytokine-challenged cardiomyocytes and inflammatory cells from IP-deficient mice. In fact, due to IP-dependent removal of poly-ubiquitinylated protein aggregates in the injured myocardium IPs protected CVB3-challenged mice from oxidant-protein damage. Impaired NFκB activation in IP-deficient cardiomyocytes and inflammatory cells and proteotoxic stress in combination with severe inflammation in CVB3-challenged hearts from IP-deficient mice potentiated apoptotic cell death in this host, thus exacerbating acute tissue damage. Adoptive T cell transfer studies in IP-deficient mice are in agreement with data pointing towards an effective CD8 T cell immune. This study therefore demonstrates that IP formation primarily protects the target organ of CVB3 infection from excessive inflammatory tissue damage in a virus-induced proinflammatory cytokine milieu.
| The proteasome recognizes and degrades protein substrates tagged with poly-ubiquitin chains. Immune cells and cells activated by inflammatory cytokines/interferons express immunoproteasomes (IPs) that are characterized by unique catalytic subunits with increased substrate turnover. In infectious disease, the function of IPs is still a matter of controversial debate. Here, we report on a novel innate function of IPs in viral infection. We studied the murine model of acute enterovirus myocarditis, which represents one of the most common viral disease entities among young people. We found that IPs protect the pathogen-challenged tissue from severe injury, which was reflected in severe myocardial destruction and large inflammatory foci in mice lacking IPs. We show data that this prevention of excessive inflammatory tissue damage in viral heart disease is primarily attributed to preservation of protein homeostasis due to accelerated substrate turnover by IPs. Thus, a major innate function of IPs in viral infection is to stabilize cell viability in inflammatory tissue injury.
| Unfolded or misfolded proteins are potentially harmful to cells and have to be efficiently eliminated before they intoxicate the intracellular environment. This is of particular importance during proteotoxic stress as a consequence of intrinsic or extrinsic factors when the levels of misfolded proteins are transiently or persistently elevated (Dantuma, 2010 #1). In viral infection cytokine exposure and inflammation induce the generation of reactive oxygen species in both immune and non-immune cells [1], [2] with concomitant oxidant-protein damage and proteotoxic stress. An important defence mechanism is the specific destruction of these proteins by the ubiquitin-proteasome system (UPS) [3]. The UPS is among others involved in the regulation of protein quality control in cardiovascular pathologies [4]–[6], in neurodegenerative disorders and other human pathologies [7], [8]. The UPS with the 26S proteasome as central proteolytic unit represents the major ATP-dependent degradation system in eukaryotes responsible for the maintenance of protein homeostasis and the generation of the vast majority of antigenic peptides that are presented by MHC class I molecules to CD8+ T cells in infectious disease [9]. Short-lived regulatory proteins involved in cell differentiation, cell-cycle regulation, transcriptional regulation, or apoptosis, but also aberrant proteins are directed to proteasomal degradation through conjugation with the small protein modifier ubiquitin via a cascade of E1, E2, and E3 enzymes, thus forming poly-ubiquitinylated (poly-ub) proteins [10]. Poly-ub proteins are substrates for 26S proteasomes which are formed through the association of two 19S regulator complexes with the catalytic core complex, the 20S proteasome, that hydrolyzes proteins into shorter peptide fragments [11], [12]. Peptide hydrolyzing activity of the 20S core is restricted to three β-subunits, i.e. β1, β2, and β5, located in the two inner heptameric β-rings of the 20S proteasome [13].
Upon interferon (IFN)-exposure of cells or tissues, alternative catalytically active β subunits, i.e. β1i/LMP2, β2i/MECL-1, and β5i/LMP7, are induced. These so called immunosubunits are incorporated into newly formed 20S immunoproteasomes (IP) in a process that is driven by β5i/LMP7 [14]. β1i/LMP2 and β5i/LMP7 are encoded within the major histocompatibility II region and their incorporation into IPs induces altered proteolytic characteristics that result in many cases in more efficient liberation of MHC class I epitopes [15]–[17] particularly within the early phase of antiviral immunity [18], [19]. This increase in MHC class I peptide supply by IPs appears to be important for triggering an effective early CD8 T cell response [20]–[23]. However, controversial findings about the association between IP function and CD8 T cell priming raised some doubts with regard to the in vivo impact of these data. In fact, an alternative physiological function of IPs has been demonstrated recently by our group in that IPs protect cells against cytokine induced oxidative damage, thus preserving protein homeostasis. Substrate modification of oxidant-damaged proteins with poly-ubiquitin results in protein degradation particularly by IPs [24]. Nevertheless, conclusive studies investigating the role of IP in response to viral infection beyond the analysis of specific T cell immunity have not been performed. Also, the importance of this regulated protease in cardiac disease remains to be elucidated. Within the context of the murine model of ongoing coxsackievirus B3 (CVB3)-myocarditis, we recently reported on cardiac IP formation early upon infection in mice being resistant to chronic disease. The remarkably delayed induction of cardiac IPs in susceptible mice pointed towards a potential disease-modifying effect of this finding [19].
Here, we show that cardiac IP prevent exacerbation of acute CVB3-induced myocardial destruction and possess a protective function in viral heart disease expanding their role to the protection of cells against inflammation induced toxic effects thereby stabilizing cell viability in viral infection.
One well-established model to study myocardial inflammation is the induction of murine myocarditis with coxsackievirus B3 (CVB3) in C57BL/6 mice leading to acute heart muscle injury at d8 p.i. [19]. Here, we have challenged both β5i/LMP7+/+ and β5i/LMP7-/- mice on a C57BL/6 background with CVB3. Cardiac 20S proteasomes isolated from naive mice contain only very small amounts of β1i/LMP2, β5i/LMP7 and β2i/MECL-1 [19], [25]. To test whether β5i/LMP7 deficiency was indeed sufficient to negatively affect the incorporation of all three inducible catalytic subunits into cardiac IP in vivo, heart 20S proteasomes were isolated from naive mice and from CVB3-infected mice at the early stage of disease (d4 p.i.) and at the acute stage of myocarditis (d8 p.i.) from both β5i/LMP7+/+ and β5i/LMP7-/- mice. Whereas mRNA expression of β1i/LMP2 and β2i/MECL-1 was induced in both hosts (Fig. 1A), immunoblot analysis revealed strongly impaired incorporation of both β1i/LMP2 and β2i/MECL-1 into cardiac 20S proteasomes in acute myocarditis in β5i/LMP7-/- mice (Fig. 1B). Likewise, cytokine stimulation of primary cardiomyocytes from IP-competent mice with IFN-γ resulted in the efficient induction of all three IP subunits, whereas as expected incorporation of β1i/LMP2, β2i/MECL-1 and β5i/LMP7 was impaired in cardiomyocytes isolated from IP-deficient mice (Fig. 1C).
To obtain quantitative information on the content of each proteasome subunit within the 20S core complex in vivo, 20S proteasomes were subjected to reverse phase nano HPLC and LTQ-Orbitrap mass spectrometry (MS) analysis. All three inducible subunits β1i/LMP2 (3.5-fold induction; p<0.05), β2i/MECL-1 (2.6-fold induction; p<0.05) and β5i/LMP7 (2.0-fold induction; p = 0.10) were enhanced in CVB3-challenged hearts as early as at d4 p.i. in β5i/LMP7+/+ mice. In contrast, no improved incorporation of either β1i/LMP2 (1.3-fold; p = 0.67) or β2i/MECL-1 (1.5-fold; p = 0.38) was observed in cardiac 20S proteasomes from β5i/LMP7-/- mice at this stage of disease. Table 1 indicates relative quantitative expression levels of all three inducible subunits in cardiac proteasomes in wildtype mice with respect to IP-deficient mice. No significant differences were detected in the content of constitutive proteasome subunits as exemplarily shown in Table 1 for catalytic subunits β1, β2, and β5, and for constitutive non-catalytic subunits α2 and β4, respectively. Notably, in agreement with previous studies [26], β5i/LMP7-/- mice show baseline deficits in cardiac β2i/MECL-1 incorporation being aggravated during inflammation (Table 1). Therefore, β5i/LMP7-/- mice encounter a substantial impairment of IP assembly in inflammation-challenged hearts. This appears to be crucial for the interpretation of our data in terms of IP deficiency showing a severe impairment in the incorporation of all three inducible subunits in acute disease in vivo.
Histological analysis of acute myocarditis was done as previously published [27] defining acute myocarditis by lymphocytic infiltrates in association with myocyte necrosis, which we also see in patients with acute myocarditis [28]. CVB3-myocarditis was evaluated in IP-deficient mice at early stages of heart muscle infection (d4 p.i.) and at acute stages of myocarditis (d8 p.i.). Except for some scattered macrophages, no foci of inflammatory infiltrates were observed in the myocardium of both mouse strains at d4 p.i. (Fig. 2B). At this time point severe inflammation of the pancreas, the primary organ of viral replication, was observed in both β5i/LMP7+/+ and β5i/LMP7-/- mice (Fig. 2A). Pointing towards accelerated organ destruction in β5i/LMP7-/- mice, representative images of the pancreas at d8 p.i. illustrate final pancreatic islet destruction with fibrous and fatty tissue organ replacement in β5i/LMP7-/- mice, with necrotic cells and massive inflammation still being present in β5i/LMP7+/+ mice (Fig. 2A).
Clinically, myocardial tissue damage in CVB3-infection is utmost important since acute myocardial injury may result in severe acute arrhythmia and heart failure. Focussing on myocardial damage comprising cardiac cell necrosis and inflammation in acute myocarditis, IP-deficiency was found to be associated with severe acute myocarditis. As representatively depicted in Fig. 2B and 2C for independent experiments by HE staining, at the acute stage of disease large foci encompassing inflammatory cells and cardiomyocyte necrosis were detected in β5i/LMP7-/- mice, which is in striking contrast to small areas of myocardial inflammation and tissues damage in β5i/LMP7+/+ mice. To obtain quantitative information on heart muscle injury, myocarditis scores were determined yielding a score of 3.1±0.3 in β5i/LMP7-/- mice vs. 2.2±0.2 in β5i/LMP7+/+ mice (p<0.05, Fig. 2C).
Macrophages and CD3+ T lymphocytes represent the major fraction of invading inflammatory cells in acute CVB3-myocarditis in mice [27], [29]. To address the inflammatory infiltrate in our model in detail, CD3+ T lymphocytes, B220+ B lymphocytes and Mac-3+ macrophages were studied by immunohistology. As demonstrated in Fig. 3, the inflammatory infiltrate was primarily comprised of Mac-3+ macrophages and to a lesser extent of CD3+ T lymphocytes. B cells were scattered throughout the inflammatory lesions without significantly contributing to the invading cellular infiltrate. In agreement with myocarditis scores (Fig. 2C), quantification of invading cells revealed significantly increased macrophages and T lymphocytes in CVB3-infected β5i/LMP7-/- mice (Fig. 3D). As suggested by quantitative mRNA expression of CD8 and CD4 molecules in the infected myocardium, both CD4+ and CD8+ T lymphocytes were increased in CVB3-infected β5i/LMP7-/- mice (mRNA expression of CD3 revealed the same result, data not shown). In contrast, expression levels of NKP46, a marker for natural killer (NK) cells, did not differ in both hosts suggesting invasion of NK cells in both hosts to the same extent (Fig. 3E). In summary, immunohistological characterization of myocardial inflammation revealed that cardiac IP formation protected CVB3-challenged hearts from exacerbation of acute heart muscle injury.
The presence of infected cardiomyocytes adjacent to foci of mononuclear cell infiltrates is pathognomonic in viral myocarditis. Indeed, CVB3 in situ hybridization-positive cardiomyocytes were found within inflammatory lesions in acute heart muscle injury in both β5i/LMP7+/+ and β5i/LMP7-/- mice (Fig. 4A). However, despite the severity of myocardial tissue damage in CVB3-challenged β5i/LMP7-/- mice, scoring of CVB3 in situ hybridization-positive cardiomyocytes pointed towards equal viral replication in both hosts. Also, the titers of cardiac infectious viral particles were found to be within the same range in both β5i/LMP7+/+ and β5i/LMP7-/- mice in acute disease (Fig. 4B). To further investigate viral replication within the context of IP-deficiency, primary cardiomyocytes from β5i/LMP7+/+ and β5i/LMP7-/- mice were infected with CVB3 in vitro and CVB3 replication was determined by quantitative real-time PCR. These experiments were also carried out in the presence of type I IFN-stimulation to mimic the in vivo cytokine milieu in acute heart muscle injury. As shown in Fig. 4C, IP-deficiency revealed no influence on CVB3 replication in vitro.
In line with the finding of identical viral load in CVB3-challenged cardiomyocytes from β5i/LMP7+/+ and β5i/LMP7-/- mice, efficient virus elimination was observed at the chronic stage of disease at d28 p.i. revealing no relevant signs of ongoing disease in both hosts (data not shown). These findings pointed towards efficient induction of both innate and adaptive immunity also in mice lacking IP. To address this issue in detail, cytokine responses were determined in acute heart muscle injury (Fig. 5A). Our data demonstrate increased cardiac expression of pro-inflammatory cytokines as shown here exemplarily for TNF-α, IFN-β, IL-6 and IFN-γ in acute myocarditis in both β5i/LMP7+/+ and β5i/LMP7-/- mice. Also, the expression of type I IFN-induced antiviral pathways like the 2′ 5′-oligoadenylate synthetase-like protein-2 (OASL-2), the IFN-stimulated gene 15 (ISG15), the Myxovirus resistance protein (Mx) and the protein kinase K (PKR) pathway was efficiently induced in CVB3-challenged hearts from β5i/LMP7-/- mice (Fig. 5A).
To determine whether IP expression is required for maintaining homeostatic levels of B and T cells, we counted B220+/CD19+ B cells and CD4+ and CD8+ T cells in spleens isolated from β5i/LMP7+/+ and β5i/LMP7-/- mice. Hensley et al. recently reported on different absolute B cell and T cell numbers and impairment in humoral immunity in β1i/LMP2-deficient mice [30]. Here, we detected equal CD4+ T cell and B cell numbers in both CVB3-challenged β5i/LMP7+/+ and β5i/LMP7-/- mice (Fig. 5B, C). To further test humoral immunity in β5i/LMP7-/- mice, anti CVB3 IgG titers were determined at different time points p.i. CVB3-antibody titers in β5i/LMP7-/- mice were found to be within the range of anti CVB3 IgG in respective wildtype controls (Fig. 5B). In accordance with data by Fehling et al. [21], CD4+ and CD8+ T cells were not affected by β5i/LMP7 deletion in naive mice (Fig. 5C). We observed an overall decrease in splenic T cell levels early upon disease potentially reflecting migration of these cells to secondary lymphoid organs and to the target organ of infection; however, host-specific effects were precluded. Also, we did not observe any differences in the absolute number of splenocytes in both hosts in acute myocarditis.
Since β5i/LMP7-deficiency has been attributed to impaired CD8 T cell responses [21], [31], we also investigated CD8 T cell populations from both β5i/LMP7+/+ and β5i/LMP7-/- mice in vivo. Despite the fact that absolute frequencies of CVB3 VP2 [285–293]-specific CD8 T cells were rather low, no remarkable differences were detected in the absolute VP2 [285–293]-peptide specific CD8 T cell numbers in CVB3-infected β5i/LMP7+/+ and β5i/LMP7-/- mice (Fig. S2A). Given that CD8 T cells are crucial in the control of CVB3-myocarditis [29], [32], absolute CD8 T cell numbers for CVB3 epitopes are low and detection of specific CD8 T cell frequencies is limited in this infection model, adoptive transfer studies of CVB3-memory CD8 T cells from IP-deficient and IP-competent mice appeared to be the most reliable approach to address the effect of IP-deficiency on CD8 T cell function. Take of donor CD8 T cells from CVB3-infected mice is shown in Fig. S2B. To preclude effects of IP-deficiency on CD8 T cell survival, CD8 T cells were isolated from CVB3-challenged β5i/LMP7+/+ and β5i/LMP7-/- mice (both CD45.2) at d8 p.i. and transferred into naive B6.SJL-Ptprca Pepcb/BoyJ mice (CD45.1). The amount of transferred CD8 T cells was assessed at d8 after adoptive T cell transfer revealing no impairment of IP-deficiency on CD8 T cell survival (data not shown). Also, transfer of β5i/LMP7-deficient CD8 T cells (CD45.2) into CVB3-infected B6.SJL-Ptprca Pepcb/BoyJ mice (CD45.1) and vice versa revealed comparable CD8 T cell survival rates (Fig. 5D left panel). Next, CD8 T cells were isolated from CVB3-infected β5i/LMP7+/+ and β5i/LMP7-/- mice (both CD45.2) at d8 p.i. These cells were transferred into naive β5i/LMP7+/+ and β5i/LMP7-/- mice, which were then infected with CVB3 and sacrificed at d8 p.i. to assess myocarditis scores. Following adoptive T cell transfer of IP-deficient CD8 T cells into either IP-deficient or IP-competent recipients, we observed no effect on acute heart muscle inflammation in comparison to adoptive T cell transfer of CD8 T cells from IP-competent mice into both recipients (Fig. 5D right panel). Of note, adoptive T cell transfer of CD8 T cells from either CVB3-infected β5i/LMP7+/+ and β5i/LMP7-/- mice into β5i/LMP7-/- recipient mice resulted in a slightly milder acute disease than in non-transfected mice (respective myocarditis score from CVB3-infected donor mice are shown in the middle panel of Fig. 5D). However, these effects were detected for T cell transfer of both β5i/LMP7-/- and β5i/LMP7+/+ T cells, thus arguing against a detrimental effect of IP-deficiency on memory CD8 T cell function and being in accordance with the observation of equal virus titers and efficient viral clearance in both β5i/LMP7+/+ and β5i/LMP7-/- mice.
The data above illustrated severe tissue damage in CVB3-challenged hearts in mice lacking IPs and revealed large foci of inflammatory lesions in this host. Given that IPs preserve protein homeostasis and cell viability in response to cytokine stress [24], one may argue that viral infection induced cytokine release affects the cellular protein equilibrium in cardiomyocytes and invading inflammatory cells, which may further exacerbate heart muscle injury in IP-deficient hearts. To test this hypothesis, primary cardiomyocytes and B-cell depleted splenocytes (which represent the major populations of invading inflammatory cells) were isolated from β5i/LMP7+/+ and β5i/LMP7-/- mice and exposed to IFN-γ (cell purity is depicted in Fig. S1). Upon cytokine exposure, lack of IPs resulted in increased accumulation of poly-ub substrates in these cells (Fig. 6A). Failure of IP expression also coincided with increased accumulation of oxidant-damaged proteins in IP-deficient cardiomyocytes and inflammatory cells in response to prolonged cytokine exposure (Fig. 6A). IPs also contribute to the activation of NFκB transcription factor by accelerated turnover of IκBα, which is crucial for multiple processes in inflammation and apoptosis [24], [33]. Impaired activation of NFκB as shown here by reduced levels of NFκB p50 subunits in IP-deficient cardiomyocytes and inflammatory cells (Fig. 6B) reflected reduced proteasomal degradation of NFκB p105 precursor proteins, which is in concordance with impaired proteolysis in IP-deficient cells as shown in Fig. 6A.
These data suggested a role of IPs in regulating proteotoxic stress also in the infected myocardium. Indeed, IP-deficient mice failed to cope with accelerated protein turnover in CVB3 infection as reflected by increased accumulation of poly-ub proteins in acute inflammatory viral heart disease (Fig. 6C). The IP-deficient myocardium was not able to efficiently cope with the required protein turnover in acute CVB3 myocarditis (Fig. 6D + Fig. S3). Consequently, we observed significantly enhanced ALIS formation in the injured myocardium (evaluation of poly-ub-aggregates at the acute stage of myocarditis: β5i/LMP7+/+ mice: 13.5 ALIS / 1088 µm2±1.0 vs. β5i/LMP7-/- mice: 20.0 ALIS / 1088 µm2±1.8; p<0.05; n = 5 mice). These poly-ubiquitin conjugates were primarily detected within inflammatory lesions in invading inflammatory cells, and within the cytoplasm and nuclei of adjacent cardiomyocytes in acute myocarditis (Fig. 6E). Also, poly-ub signals were found to be increased in β5i/LMP7-/- mice in comparison to β5i/LMP7+/+ mice at d8 p.i. (Fig. 6E + Fig. S4). Since oxidant damaged proteins become substrates of the 26S IP upon tagging by poly-ub [24], the levels of carbonyl groups reflecting oxidant protein damage were monitored. As illustrated in Fig. 6F, oxidant protein damage was increased in acutely inflamed hearts in IP-deficient mice.
Since CVB3 titers were found to be within the same range in both hosts (Fig. 4), cytolytic effects of CVB3 do apparently not explain severe tissue injury as observed here in IP-deficient mice. However, oxidative-protein damage, inefficient degradation of poly-ub protein aggregates and reduced activation of NFκB transcription factor in CVB3-challenged hearts in mice lacking IPs may affect cell viability. Indeed, cytokine-induced cellular injury predominantly occurred in vitro in cardiomyocytes and macrophages that were isolated from IP-deficient mice (data not shown). To study cellular injury due to apoptotic cell death in vivo, DNA strand breaks as an early sign of apoptosis were assessed in cardiac tissue sections. No apoptotic cell death was detected in hearts from β5i/LMP7+/+ and β5i/LMP7-/- mice at d4 p.i. (Fig. S5). However, in acute heart muscle injury (d8 p.i.), increased levels of DNA strand breaks were detected particularly within inflammatory lesions and the surrounding tissue in CVB3-challenged β5i/LMP7-/- mice (Fig. 7, Fig. S5+S6). TUNEL positive staining was detected throughout the injured heart in IP-deficient mice; thus, apoptotic cell death was found to be quantitatively increased in β5i/LMP7-/- mice. Despite the fact that minor inflammatory lesions were also detected in CVB3-infected β5i/LMP7+/+ mice (d8 p.i.), here no significant apoptotic cell death occurred (Fig. 7). This observation is in agreement with previously published data [29]. These findings support the role of IP formation in cardiomyocytes and in inflammatory cells to protect the injured tissue from proteotoxic stress, which may exacerbate acute heart muscle injury in viral heart disease.
Proteasomes are responsible for the generation of peptides derived from pathogens or cellular proteins that are presented by MHC class I molecules on the cell surface to cytotoxic T cells (CTLs) [11]. Despite the fact that in vitro studies argued in favor of an impact of IP-dependent MHC class I antigen processing [15]–[17], [19], in vivo studies using IP-deficient mice reported conflicting data on the induction of CD8 T cell responses [17], [21], [23], [34]. CD8 T cells are crucial in virus elimination in CVB3-myocarditis [29], [32]. To study a potential contribution of IPs to the generation of CD8 T cell responses in CVB3-myocarditis, here adoptive memory CD8 T cell transfer experiments were performed since limited knowledge on immunodominant CVB3-specific CD8 T cell epitopes restrains solid quantification of CD8 T cell responses in murine enterovirus myocarditis [19], [35]. Transfer of CVB3 memory CD8 T cells from IP-competent mice did not reveal a beneficial effect on CVB3 myocarditis in comparison to transfer of CD8 T cells from IP-deficient mice. Likewise, CVB3 titers were within the same range in IP-deficient and wildtype control mice in acute myocarditis (Fig. 4) and the virus was efficiently eliminated in both hosts at d28 p.i. These findings support the induction of efficient CD8 T cell responses also in CVB3-challenged IP-deficient mice, which is in agreement with observations in other infection models: the kinetics of lymphocytic choriomeningitis virus clearance were similar in both β5i/LMP7+/+ and β5i/LMP7-/- mice [34]. This strongly supports the notion that the key innate function of IP in enteroviral heart disease lies elsewhere. Here, we have illustrated that IPs in CVB3-induced heart muscle injury preserve protein homeostasis and maintain cell viability in order to protect the inflammation-challenged myocardium from severe damage.
The UPS adapts to stress induced requirements by increased substrate turnover exerted by IPs, which possess improved peptide-hydrolyzing activity [15], [17], [35], and poly-Ub-substrate turnover [24]. Indeed, cardiac IP formation in CVB3-myocarditis resulted in enhanced proteasomal peptide-hydrolyzing activity [19]. One of the pivotal functions of the UPS is to limit the accumulation of potentially toxic misfolded proteins and protein aggregate formation, which as consequence of cellular stress represent a constant threat to normal cell function and cell viability [7], [36]. Also, CVB3-infection [2] as well as cytokine stress in inflammation [1] induce oxidative stress. Likewise, the pathogenesis of severe CVB3 myocarditis has been attributed to increased oxidative stress [37]–[39]. In agreement with our previous study reporting on the activity of the 26S IP for efficient elimination of oxidatively modified, poly-ub proteins in response to cytokine stress [24], here we demonstrated that IP function in both residing host cells and invading inflammatory cells is crucial for the efficient degradation of poly-ub proteins in acute viral heart disease. Elimination of nascent oxidant-damaged, poly-ub proteins by the 26S IP prevented the accumulation of harmful protein aggregates [7], [36], which may exacerbate acute heart muscle injury.
Moreover, as a consequence of impaired IP function, poly-ub proteins accumulated in ALIS in CVB3-infected hearts. Such aggregates, which are at least partially comprised of CVB3 proteins [40], are not inert metabolic end products, but may actively influence the metabolism of cells [41]. As shown here, degradation of oxidant-damaged, poly-ub proteins by cardiac 26S IPs in CVB3-challenged hearts resulted in ALIS degradation and likewise protected cardiomyocytes and invading inflammatory cells from proteotoxic stress. This first histological demonstration of severe tissue injury in virus infection in mice lacking IPs is in agreement with findings in experimental acute encephalomyelitis (EAE). IPs prevented accumulation of toxic protein aggregates in EAE, coinciding with less severe disease manifestation in IP-competent mice [24]. The detection of both poly-ub proteins in concert with apoptotic cell injury within inflammatory lesions in viral heart disease (Fig. 6+7) also supports the association between proteotoxic stress and cellular injury in this model [7], [24]. In fact, cellular injury as shown here by increased apoptotic cell death of invading inflammatory cells and adjacent cardiomyocytes in IP-deficient mice may result in the release of endogenous molecules, which as damage-associated molecular patterns (DAMPs) signal the threat of infection and injury to the organism. High levels of DAMPs have been linked to the pathogenesis of many inflammatory diseases, drive cellular activation and immunoreactivity [42]. This may in fact exacerbate acute inflammation and also result in killing of non-infected cardiomyocytes. Thus, IPs prevent excessive proteotoxic stress and cellular injury, which in consequence may limit additional effects like DAMP-associated activation of immunopathology.
Remarkably, this is the first study illustrating a detrimental effect of IP-deficiency in a viral infection-induced phenotype despite a lack of a significant effect of IPs on pathogen load. Our findings are in agreement with the observation that absence of β5i/LMP7 expression impairs the beneficial effects of IFN-β in patients suffering from multiple sclerosis [43]. In contrast, absence of IP-function either as a result of β5i/LMP7 deficiency or inhibition of β5i/LMP7 catalytic activity by PR-957 have recently been associated with attenuated experimental colitis [44]. Depending on the pathogenesis of the underlying disease, IP deficiency seems to exert either protective effects or to aggravate the consequences of inflammation in a disease or tissue-specific manner. Indeed, cytokine responses in IP-deficient mice differ considerably and are strikingly dependent on the disease entity being studied. CVB3 infection of IP-deficient mice revealed a cytokine induction in viral heart muscle injury comparable to that observed in wildtype mice (Fig. 5A). In contrast, amelioration of experimental colitis has been connected to limited induction of proinflammatory cytokines and chemokines [33], [44]. This was attributed to impaired activation of transcription factor NFκB, a central regulator of inflammation in inflammatory bowel disease [33]. In line with impaired NFκB activation in TNF-α stimulated murine embryonic fibroblasts [33], this study demonstrated reduced NFκB activation in cytokine-stressed cardiomyocytes and inflammatory cells lacking IPs (Fig. 6B). This may be attributed to the fact that IκBα, a specific inhibitor of NFκB activation, is degraded much faster in cells expressing IPs [45]. Also, IκBα has been identified to be oxidatively-modified upon cytokine stress, which supports the role of IPs in degradation of this specific substrate [24]. Whereas increased activation of NFκB is believed to exert detrimental functions in immune and non-immune cells in tissues affected by chronic inflammation, NFκB inhibition can also be harmful for the organism, and trigger the development of inflammation and disease. These findings suggested that NFκB signaling has important functions for the maintenance of physiological immune homeostasis and for the prevention of inflammatory diseases [46]. Studies with specific inhibitors of NFκB nuclear translocation and activity revealed induction of apoptosis, thus argueing in favor for anti-apoptotic effects of this prosurvival transcription factor as well [47]. In agreement with these findings, the here described impaired activation of NFκB may additionally contribute to the effects of proteotoxic stress, which resulted in cellular injury as shown in Fig. 7. Whereas interference with two major pathways leading to NFκB activation exerts beneficial effects in experimental colitis and anti-cancer treatment, our data indicate that activation of NFκB-mediated responses protects cytokine-challenged cardiomyocytes and inflammatory cells and argues against a significant contribution of NFκB to cytokine induction in viral heart disease. In conclusion, our findings support the view of a distinct tissue specific contribution of IP function driven by the pathogenesis of the underlying inflammatory disease.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the German animal welfare act, which is based on the directive of the European parliament and of the council on the protection of animals used for scientific purposes. This study conforms to the Berlin State guidelines for animal welfare. The protocol was approved by the Committee on the Ethics of Animal Experiments of Berlin State authorities (Permit Number: G0311/06). All efforts were made to minimize suffering.
CVB3 (cardiotropic Nancy strain) used in this study was prepared as previously described [48]. C57BL/6 mice were initially from Jackson Laboratory. C57BL/6 β5i/LMP7-/- mice originally obtained from HJ Fehling [21], who backcrossed them with C57BL/6 mice 10 times. Breeding pairs were kindly provided by U Steinhoff (Berlin, Germany). Mice were kept at the animal facilities of the Charité University Medical Center. Six week-old mice were infected i.p. with 1×105 PFU CVB3. Hearts were perfused with PBS, weighted and quickly frozen in liquid nitrogen before storage at -80°C. For some lymphocyte transfer experiments, B6.SJL-Ptprca Pepcb/BoyJ were purchased from Jackson Laboratory.
Primary cardiomyocytes (CM) were isolated from fetal mouse hearts (E13). Hearts were incubated in EDTA/Trypsin at 4°C overnight, followed by 15 min incubation at 37°C. Cardiac cells were resuspended in standard medium and transferred into cell culture flasks. Purity was determined by flow cytometry using cardiomyocyte-specific troponin I antibodies (abcam # 47003). Inflammatory cells were taken from whole spleen cell suspensions, which were B cell depleted by MACS (Miltenyi Biotec). All cell lines were cultivated under standard conditions in Dulbecco's MEM (MEFs) each containing 10% fetal calf serum (FCS), 2 mM L-glutamine, 100 U/ml penicillin and 100 mg/ml streptomycin. Cells were treated with either 100 U/ml IFN-β, 100 U/ml IFN-γ (Sigma-Aldrich) or 30 ng/ml TNF-α (all Sigma-Aldrich). Cells were infected with CVB3 (cardiotropic Nancy strain) MOI 0.1 or 0.5 as indicated. Plaque assay was performed as described previously [19].
In situ hybridization of genomic CVB3 RNA, histological staining with hematoxylin / eosin (HE) and immunohistochemistry for detection of CD3+ T lymphocytes and Mac-3+ macrophages were carried out and analysed as described [27]. Immunohistochemical stainings for ubiquitin and B cells were performed on a Ventana Benchmark stainer using the Vectostain Elite ABC Kits (Vector Laboratories; Burlingame, CA). The following primary antibodies were used: anti-ubiquitin 1∶1000 (DAKO Cytomation, Glostrup, Denmark) and CD45R/B220 1∶100 (BD Pharmingen, Heidelberg, Germany). Biotin-labeled secondary antibodies (goat-anti-rabbit and goat-anti-rat) were purchased from Jackson Immuno Research (Dianova, Hamburg, Germany) and used at a dilution of 1∶100. All slides were counterstained with hematoxylin.
Processing of cryo-sections and Hoechst staining was performed as described [24]. Briefly, cryo-sections were fixed in 4% paraformaldehyde, washed and permeabilized with PBS/1% TritonX. Staining was performed with FK1 mAb (PW 8805, Biomol, Germany) at 4°C over night. Confocal images were acquired on a Leica TCS SP2 microscope (Leica Microsystems). Quantification of ALIS has been based on counting cells with accumulation of poly-ub conjugates (focused staining over background defined as ub-rich inclusions) in defined areas (1088 µm2) at 100-fold magnification.
RNA preparation and cDNA synthesis were performed as described recently [19]. TaqMan PCR was performed using primers and probes of TaqMan Gene Expression Assays (Applied Biosystems, Germany). mRNA expression was normalized to the housekeeping gene HPRT by means of the ΔCt method.
Cell or tissue lyses was performed in 20 mM TRIS-HCl, pH 7.5, 10 mM EDTA, 100 mM NaCl, 1% NP40, 10 µM MG132, 5 mM NEM, Complete protease inhibitor cocktail (Roche, Germany). Immunoblot analysis was performed according to standard protocols. ubiquitin: Z0458 DAKO; α6 (pc, K379), β5i/LMP7 (pc, K63), β2i/MECL-1 (pc, K65): lab stock (all generated against peptides of the respective protein), β1i/LMP2: Abcam #3328 for isolated proteasomes or β1i/LMP2: lab stock for cardiomyocytes (cross-reaction with β1, pc, K620/21); α-actin, GAPDH: Santa Cruz.
The detection of oxidatively-damaged proteins was performed indirectly by chemical derivatization: this derivatization captures the oxidative state immediately during or after homogenization of the tissue. Oxidized proteins were visualized with the OxyBlot protein oxidation detection kit (Chemicon International) via immunodetection of carbonyl groups. DNA strand breaks (TUNEL assay) were determined by in situ cell death detection kit, TMR red (Roche, Germany) or in situ cell death detection kit, POD (Roche, Germany) according to the instructions of the manufacturer. POD stained slides were counterstained by hematoxylin.
After induction of IPs with IFN-γ, primary cardiomyocytes and B-cell depleted splenocytes were stimulated with 30 ng/ml TNF-α for 30 min. p50 NFκB was determined in whole tissue homogenates by ELISA according to the manufacture's instructions (ActiveMotif, Rixensart, Belgium).
CVB3-specific IgG antibody titers were determined with Enterovirus ELISA Kit (Genzyme Diagnostics) according to the manufacture's instructions [alternative secondary antibody (POX anti-mouse IgG, Dianova) was used]. CVB3-specific antibody titers are presented as log2 of the maximum dilution of serum showing an optical density greater than the mean optical density of sera obtained from naive mice plus threefold SD as described recently [27].
After Fc-receptor blockade cells were incubated with different combinations of fluorescently labeled Abs (eBiosciences and BD Biosciences) and samples were analyzed using CYAN-ADP flow cytometer (Beckman Coulter, Germany) or BD FACSCalibur (Becton Dickinson).
At day 8 p.i., CVB3-infected mice were sacrificed and splenocytes were isolated according to standard protocols. CD8+ T cells were purified by positive selection using commercially available kits yielding a purity of at least 85% (Miltenyi Biotec). 1-2×106 CD8+ T cells from one donor were injected i.v. through the tail vein into one recipient. After T cell transfer, mice were injected i.p. with 1×105 PFU CVB3 and sacrificed at the acute stage of infection at d8 p.i. Myocarditis was assessed as described above.
20S proteasomes were isolated according to standard procedure as described [19]. The mixture of tryptic peptides was separated prior to mass spectrometric analyses by reverse phase nano HPLC on a 15 cm PepMap100-column (3 µl, 100 Å) using an Proxeon System (Odense, Denmark) at a flow rate of 1 µl/min. Separation was carried out in a linear gradient within 86 min using 0.05% acetic acid, 2% acetonitrile in water and 0.05% acetic acid in 45% acetonitrile as eluents. MS-data were generated on an LTQ-Orbitrap-MS equipped with a nanoelectrospray ion source (PicoTip Emitter FS360-20-20-CE-20-C12, New Objective). After a first survey scan (r = 60,000) MS2 data were recorded for the five highest mass peaks in the linear ion trap at a collision induced energy of 35%. The exclusion time was set to 30 s and the minimal signal for MS2 was 1,000. Peptide identification was achieved by searching the SwissProt database rel. 57.1 restricted to mouse entries using SEQUEST search engine (SageN Research) and further processed by PeptideTeller and ProteinTeller [49] within the Elucidator system (Rosetta Biosoftware, Seattle, WA, U.S.A.). ProteinTeller results were further used for annotation, with a predicted error rate of<5%. Quantitative analysis of label-free MS data was achieved with the Elucidator system using peptide intensities as proxies for label-free peptide abundance measurements. The following criteria for frame and feature annotation were used: retention time minimum cut-off 9 min, retention time maximum cut-off 80 min, m/z minimum cut-off 300, instrument mass accuracy 5 ppm, alignment search distance 10 min. For quantitative analysis, the data were normalized and further grouped (two biological and two technical replicates).
Results of continuous variables are expressed as mean±standard error of mean (SEM) if not indicated otherwise. Two group comparisons of non-parametric data were performed using the Mann-Whitney test. Statistical significance between multiple groups was determined using two-way ANOVA and post hoc analysis with a Bonferroni test. Significance was assessed at the p<0.05 level (* indicates significant differences).
TNF-α (Q0X0E6, P06804), IFN-β (P01575), IL-6 (P08505), IFN-γ (P01580), OASL-2 (Q9Z2F2), ISG15 (Q64339), Mx (P09922), PKR (Q03963), NFκB p105 (P25799), IκBα (Q9Z1E3), TLR7 (P58681), TLR8 (P58682), LMP2 (P28076), MECL-1 (O35955), LMP7 (P28063), CD8 (P01731), CD68 (P31996), CD3 (P22646), B220 (P06800), CD4 (P06332), CD19 (P25918)
|
10.1371/journal.pbio.1000305 | Shape Invariant Coding of Motion Direction in Somatosensory Cortex | Invariant representations of stimulus features are thought to play an important role in producing stable percepts of objects. In the present study, we assess the invariance of neural representations of tactile motion direction with respect to other stimulus properties. To this end, we record the responses evoked in individual neurons in somatosensory cortex of primates, including areas 3b, 1, and 2, by three types of motion stimuli, namely scanned bars and dot patterns, and random dot displays, presented to the fingertips of macaque monkeys. We identify a population of neurons in area 1 that is highly sensitive to the direction of stimulus motion and whose motion signals are invariant across stimulus types and conditions. The motion signals conveyed by individual neurons in area 1 can account for the ability of human observers to discriminate the direction of motion of these stimuli, as measured in paired psychophysical experiments. We conclude that area 1 contains a robust representation of motion and discuss similarities in the neural mechanisms of visual and tactile motion processing.
| When we physically interact with an object, our hands convey information about the shape of the object, its texture, its compliance, and its thermal properties. This information allows us to manipulate tools and to recognize objects based on tactile exploration alone. One of the hallmarks of tactile object recognition is that it involves movement between the skin and the object. In this study, we investigate how the direction in which objects move relative to the skin is represented in the brain. Specifically, we scan a variety of stimuli, including bars and dot patterns, across the fingers of non-human primates while recording the evoked neuronal activity. We find that a population of neurons in somatosensory cortex encodes the direction of moving stimuli regardless of the shape of the stimuli, the speed at which they are scanned across the skin, or the force with which they contact the skin. We show that these neurons can account for our ability to perceive the direction of motion of tactile stimuli.
| In both vision and touch, information about form and motion is inferred from a spatio-temporal pattern of activation across a two-dimensional sensory sheet (the retina and the skin). The early stages of form processing have been shown to be similar in these two modalities in that both involve decomposing the stimulus into a set of local oriented contours [1]–[3]. Furthermore, the tactile integration of local motion cues has been shown, in psychophysical experiments, to be analogous to its visual counterpart [4] and the visual and tactile perception of motion have been shown to interact (see e.g. [5],[6]). In previous studies of motion processing in primary somatosensory cortex (S1), a population of neurons has been identified whose responses are modulated by the direction of stimulus motion [7]–[10]. Directionally sensitive neurons have been found in three areas of what is traditionally considered S1, namely 3b, 1, and 2. The question remains how representations of motion are elaborated in these three cortical areas.
In the present study, we investigate the representation of motion in S1 using experimental paradigms inspired by vision research. To that end, we deliver three types of motion stimuli—bars, dot patterns, and random dot displays—to the fingertips of Rhesus macaques while recording the responses these stimuli evoke in neurons in areas 3b, 1, and 2. Stimuli are delivered using a 400-probe stimulator, the tactile analogue of a visual monitor [11]. Probes are indented into the skin in spatio-temporal sequences, analogously to pixels on a monitor, to produce verisimilar percepts of shape and motion. Because the spacing between adjacent probes is shorter than that between adjacent mechanoreceptors in the skin, the inherent pixelation of the array is not felt. The bars and dot patterns are scanned across the receptive field (RF) in each of 16 directions ranging from 0° to 360°. The random dot displays are the tactile analogues of stimuli that have been widely used in studies of visual motion [12],[13]. In the tactile version, a set of five hemispheric dots move across the skin surface (see inset of Figure 1C); the degree to which the dots move in a coherent direction can be varied. At one extreme (0% coherence), the direction of motion of each dot at each point in time is determined randomly. In this condition, the display cannot and does not yield any holistic percept of motion direction. At the other extreme (100% coherence), all the dots move in the same direction, and the display yields a robust percept of motion direction. At intermediate levels of coherence, the probability that individual dots move in the prescribed direction is set (at a level determined by the motion coherence) between 0% (chance) and 100%. In a paired psychophysical study, we measured the ability of human subjects to discriminate the direction of motion of these same stimuli presented to their left index fingertips.
The objective of the present study was to ascertain (1) whether a population of neurons in areas 3b, 1, and 2 conveyed motion information that was invariant relative to the spatial properties of the stimulus (i.e., its two-dimensional form); (2) whether the direction signal was modulated by motion coherence (in the case of random dot displays) as has been found for neurons in area MT; and (3) whether the motion signal conveyed by a subpopulation of neurons in areas 3b, 1, and 2 could account for psychophysical performance across paradigms and stimulus types.
We recorded the responses of 20 SA1 and 11 RA afferents, and 92, 148, and 37 neurons in areas 3b, 1, and 2, respectively (peripheral afferents were tested only with scanned bars; only a subset of cortical neurons was tested with all stimuli) (see Table 1). Figure 1 shows the responses of a neuron in area 1 to (A) scanned bars, (B) dot patterns and (C) random dot displays varying in coherence. The neuron responded most strongly when stimuli moved medial to lateral relative to the midline with a slight proximal to distal slant (preferred direction = 20°). Importantly, its preferred direction was approximately the same across stimulus types, demonstrating that this neuron conveys information about stimulus direction that is invariant with respect to spatial form. Furthermore, the neuron's responses to all the random dot displays were equal when their coherence was 0% (cyan rasters and tuning curve in C) but direction tuning emerged and then sharpened as the motion coherence increased.
To quantify the strength of tuning, we derived a direction selectivity index (DI) (see Materials and Methods) that increased from 0 to 1 as tuning strength increased. Figure 2A shows the cumulative histogram of DI obtained from the responses of peripheral afferents and neurons in areas 3b, 1, and 2 to scanned bars. Responses of individual SA1 and RA afferents were not tuned for direction as evidenced by the fact that they yielded DIs near zero (Figure 2A; also see Figures S2 and S3). In contrast, tuning for direction is evident at the earliest stage of cortical processing, namely in area 3b, which comprised a large proportion of neurons that were sensitive to the direction of motion of scanned bars. Direction tuning for bars was greater in areas 1 and 2, which contained a much larger proportion of neurons that exhibited strong direction tuning than did area 3b (Table 1). Although direction tuning in responses to dot patterns was present in areas 3b and 2, it was stronger for neurons in area 1 (Figure 2B, Table 1) (note that, although the numerical value of DI derived from responses to dot patterns tended to be higher for neurons in area 2 than in area 3b, many of the DIs derived from area 2 responses were not statistically reliable, as shown in Table 1). Finally, neurons in areas 3b and 2 exhibited only weak direction tuning in their responses to random dot displays at 100% coherence (Figure 2C), whereas the responses of a large proportion of area 1 neurons were strongly tuned for direction. Area 2 neurons exhibited particularly weak direction tuning to dot patterns, suggesting that these neurons are sensitive to edges; indeed, a large proportion of area 2 neurons are orientation selective (unpublished data). Despite the fact that area 2 is higher in the somatosensory pathway than area 1, it seems that the latter comprises a more robust representation of direction of motion than does the former.
Next, we examined the effect of motion coherence on direction tuning in direction-sensitive neurons. We found that the increase in tuning strength was marginal for neurons in area 3b, whereas it was substantial for neurons in area 1 (Figure 2D). The tuning strength of area 2 neurons exhibited an intermediate dependence on motion coherence, and direction tuning for these neurons only emerged at high levels of coherence (>70%).
We then wished to ascertain (1) whether individual neurons conveyed information about direction across stimulus types and (2) whether the direction signal conveyed by those neurons remained unchanged as the stimulus type changed. To this end, we identified a population of neurons that were significantly tuned for bars, dot patterns, and random dot displays. We found that no neurons in area 3b and 8% of the neurons (2 of 25 neurons) in area 2 that were tested with all three types of patterns exhibited significantly direction-tuned responses to all three stimulus types. In contrast, 30% (14 of 42) of the neurons in area 1 exhibited significant direction tuning independent of stimulus type, with a large majority having the same preferred direction for all three stimulus types (Figure 2E and 2F).
The direction signal conveyed by these neurons was also largely insensitive to changes in the stimulus amplitude (i.e., the indentation amplitude), or scanning speed over a wide range of behaviorally relevant amplitudes and speeds. The strength of direction tuning was not significantly modulated by stimulus amplitude (Figure 3A; F(2,705) = 0.4, p>0.6). Furthermore, with few exceptions, the preferred direction was the same across stimulus amplitudes (Figure 3B, 84% of direction selective neurons yielded preferred directions that differed by less than 45° across the two amplitudes, in contrast to afferents; see Figure S3). Similarly, while strength of tuning increased slightly but significantly with scanning speed across the population (Figure 3C; F(3,540) = 7.3, p<.01), the strength of direction tuning of individual neurons exhibited a wide variety of relationships with scanning speed (Figure S4), as did the strength of their responses (unpublished data). Importantly, the preferred direction of individual neurons was consistent across speeds (Figure 3D).
As shown above, a subpopulation of neurons in areas 3b, 1, and 2 conveys significant information about direction of motion for each stimulus type (Figure 2). Can the responses of these neurons account for our ability to discriminate direction of motion? We derived psychometric functions from clockwise-counterclockwise judgments obtained from human subjects and compared them to analogous “judgments” derived from the responses of individual neurons. Specifically, we used an ideal observer analysis to determine the extent to which stimuli moving in different directions could be discriminated on the basis of the responses these evoked in individual neurons. We found that the responses of the most direction-selective neurons in area 1 could account for psychophysical performance (Figure 4). (We carried out this analysis using data only from neurons that were significantly direction selective for bars, dot patterns, and random dot displays. No neurons in area 3b and only 8% of neurons in area 2 met our selection criterion; see above.). Indeed, the direction of motion of bars, dot patterns, and random dots could be distinguished on the basis of the responses of a subpopulation of neurons in area 1 with the same accuracy as that observed in human psychophysical experiments (Figure 4A–C, also see Figure S5). Furthermore, the sensitivity of the direction signal to motion coherence mirrored that of human subjects (Figure 4D). Our results are therefore consistent with the hypothesis that area 1 comprises a population of neurons whose responses underlie our ability to perceive the direction of tactile motion. However, the neuronal and behavioral data were obtained from different species; this hypothesis could be tested in future experiments by assessing whether electrically stimulating clusters of direction selective neurons systematically affects the animal's performance in a direction discrimination task [14] or by ascertaining whether the responses of direction selective neurons are predictive of a monkey's behavior [15],[16].
A hallmark of many visual neurons is that they are sensitive to both direction of motion and stimulus orientation. We ascertained the extent to which neurons exhibited this dual sensitivity by examining their responses to scanned and indented bars. Specifically, we gauged the strength of orientation and of direction tuning in the responses of each neuron in our sample to scanned bars. We found that these neurons spanned the spectrum of tuning properties (Figure 5A). Some neurons (15%) were sensitive to orientation and not direction (Figure 5B); others (36%) were sensitive to direction but not orientation (Figure 5C). A large proportion of neurons (32%), however, were sensitive to both; for example, the neuron shown in Figure 5D responded to a bar oriented perpendicular to the long axis of the finger regardless of whether the bar moved proximal to distal (90°) or distal to proximal (270°), but produced a more robust response in the latter direction than in the former. In area MT, the relative preferred directions and orientations vary widely. However, the preferred direction is often perpendicular to the preferred orientation [17]. We tested whether this was the case for area 1 neurons by comparing the preferred orientation, measured from the responses to indented bars, to the preferred direction (measured from the responses to scanned dot patterns). We found that, indeed, neurons that were sensitive to both orientation and direction exhibited a variety of relative orientation and direction preferences, with a tendency for orthogonal preferences (Figure 5E).
In summary, area 1 comprises a population of neurons that are strongly tuned for stimulus direction and whose tuning is invariant with respect to three major stimulus properties, namely spatial form, speed, or intensity. Furthermore, the responses of these neurons can account for the ability of human observers to discriminate the direction of moving stimuli across a range of conditions. Finally, a large population of neurons is tuned to both stimulus direction and orientation, with the preferred direction predominantly orthogonal to the preferred orientation. These neurons are specialized detectors for moving contours and thus have RF properties that are strongly analogous to those of neurons in primary visual cortex or area MT.
As individual mechanoreceptive afferents are not sensitive to stimulus motion, an explicit representation of motion must emerge at higher processing stages. Computational models have been proposed to describe how the isomorphic representation of the stimulus at the somatosensory periphery is processed to yield information about direction of motion. Direction selectivity has been thought to be conferred by asymmetries in the spatial layout of in-field inhibition (also referred to as replacing or lagging inhibition [2],[18],[19]). However, in-field inhibition is stronger in area 3b than it is in area 1 [19], while neurons in the former exhibit weaker tuning than neurons in the latter. Rather, we propose that direction tuning first emerges in area 3b, produced in part by in-field inhibition and perhaps by mechanisms such as those observed in early visual motion processing (see e.g. [20]). This direction signal is then elaborated in area 1 to yield a more invariant representation of motion direction. Models of the neural mechanisms that produce increasingly invariant motion representations with respect to other stimulus properties at successive processing stages have been developed for the visual system (e.g., [21],[22]). The similarity in the visual and somatosensory representations of stimulus motion suggests that similar mechanisms may be involved in developing these representations in the two modalities [4].
Interestingly, complex processing of motion signals, in some ways analogous to that observed in area MT, occurs in a primary sensory area. Note, however, that area 1 is not strictly a primary somatosensory area [23]. Indeed, thalamocortical projections to area 1 are sparser, target layer III rather than layer IV, and comprise finer fibers than do those to areas 3a and 3b [24]–[27]. Furthermore, neurons in area 1 also receive strong projections from area 3b [28]. Indeed, many neurons in area 1 have larger RFs than do neurons in area 3b and are thought to each receive convergent input from multiple 3b neurons [29]. They are also less linear in the stimulus displacement profile than are their 3b counterparts [3],[19], which may in part account for the invariance of the representation of motion direction they carry with respect to stimulus parameters such as spatial form and speed. Area 1 also comprises a strong representation of stimulus orientation [3] and texture [30], which suggests that it serves other functions and is not a dedicated area for motion processing. The contiguity of form, texture, and motion representations in somatosensory cortex is not surprising given that motion is a hallmark of tactile exploration. Information about motion direction may indeed be necessary to resolve the spatial relationships between stimulus features during scanning [31].
Stimuli were generated and delivered with a device consisting of 400 independently controlled pins arrayed over a 1 cm2 area [11]. This array allows us to generate complex spatio-temporal patterns that simulate the kind of stimulation generated when a finger contacts a surface or object. Bars, scanned dot patterns, and random dot displays at 100% coherence yielded strong motion percepts as reported in pilot psychophysical experiments. Subjects were also able to clearly discern the spatial structure of the bars and the dot patterns (see Text S1 for a discussion of tactile acuity and skin mechanics).
To gauge the strength of direction tuning in the responses to scanned bars and dot patterns, we used vector strength:(1)where R(θi) is the neuron's response to a stimulus (bar or dot pattern) scanned in direction θi [36]. Values of DI ranged from 0, if for example a neuron responded uniformly to all scanning directions, to 1, when a neuron only responded to stimuli scanned in one direction. The statistical reliability of DI was tested using a standard randomization test (α = 0.01).
The preferred direction was determined by computing the weighted circular mean:(2)
We wished to assess the degree to which individual neurons also conveyed information about stimulus orientation. To that end, we computed an orientation selectivity index (OI), analogous to the DI:(3)(see above for conventions). Values of OI ranged from 0, if for example a neuron responded uniformly to all orientations, to 1, when a neuron only responded to stimuli at a single orientation. The statistical reliability of OI was tested using a standard randomization test (α = 0.01).
The preferred orientation was then determined by computing the weighted circular mean with a moment of 2:(4)
We wished to ascertain the extent to which neural responses could account for our psychophysical data using a standard ideal observer analysis. For each neuron, we randomly sampled, with replacement, from one of the five neural responses evoked by a stimulus moving in its preferred direction and one of the 10 neural responses evoked by a stimulus moving in a direction shifted by |Δφ| degrees relative to the preferred direction (there were five presentations of each stimulus and we assumed that ±Δφ are equivalent). We repeated this procedure 500 times and computed the proportion of times the response was greater in the preferred direction than when the direction was shifted by Δφ for each Δφ (ranging from 0° to 180°). Thus, to obtain the relative distributions of neural responses at φp and φp±22.5° evoked in a cell whose preferred direction was 90°, we sampled, on each of 500 iterations, one response evoked by a stimulus moving in the proximal-to-distal direction (90°) and one response evoked by a stimulus moving at 67.5° or 112.5°. We then computed the proportion of times the former was larger than the latter. The resulting neurometric functions provide an estimate of how well one could discriminate direction of motion based on the responses of individual neurons.
A similar analysis was performed to compute the neurometric function for random dot displays at various coherence levels: At each coherence level, responses to stimuli at the neuron's preferred direction were compared to responses to stimuli at its anti-preferred direction.
For the purposes of illustration, we fit the data shown in Figure 5B–D with orientation, direction, and combined orientation/direction tuning functions, respectively. For Figure 5B, we used a von Mises function (circular Gaussian) with a moment of 2:(5)where R(θ) is the neuronal response to a bar scanned in direction θ, θp is the preferred orientation of the neuron, and α, β, and δ are free parameters representing the depth of modulation of the response, the width of tuning for orientation, and the baseline response, respectively. This function denotes orientation tuning. For Figure 5C, we used a von Mises function with a moment of 1, which denotes direction tuning:(6)(see above for conventions). Finally, for Figure 5D, we used products of von Mises functions of the form:(7)where α, β, γ, and δ are free parameters representing the depth of modulation of the response, the width of tuning for orientation, the width of tuning for direction, and the baseline response, respectively. The preferred orientations of the three neurons shown in Figure 5B–D were perpendicular to their preferred directions as reflected in the fitted function (there is only one parameter denoting preference, namely θp). We also fit summed von Mises and found that the fits were equivalent but required an additional parameter (because orientation and direction tuning needed to be weighted independently).
|
10.1371/journal.pgen.1002106 | CorE from Myxococcus xanthus Is a Copper-Dependent RNA Polymerase Sigma Factor | The dual toxicity/essentiality of copper forces cells to maintain a tightly regulated homeostasis for this metal in all living organisms, from bacteria to humans. Consequently, many genes have previously been reported to participate in copper detoxification in bacteria. Myxococcus xanthus, a prokaryote, encodes many proteins involved in copper homeostasis that are differentially regulated by this metal. A σ factor of the ECF (extracytoplasmic function) family, CorE, has been found to regulate the expression of the multicopper oxidase cuoB, the P1B-type ATPases copA and copB, and a gene encoding a protein with a heavy-metal-associated domain. Characterization of CorE has revealed that it requires copper to bind DNA in vitro. Genes regulated by CorE exhibit a characteristic expression profile, with a peak at 2 h after copper addition. Expression rapidly decreases thereafter to basal levels, although the metal is still present in the medium, indicating that the activity of CorE is modulated by a process of activation and inactivation. The use of monovalent and divalent metals to mimic Cu(I) and Cu(II), respectively, and of additives that favor the formation of the two redox states of this metal, has revealed that CorE is activated by Cu(II) and inactivated by Cu(I). The activation/inactivation properties of CorE reside in a Cys-rich domain located at the C terminus of the protein. Point mutations at these residues have allowed the identification of several Cys involved in the activation and inactivation of CorE. Based on these data, along with comparative genomic studies, a new group of ECF σ factors is proposed, which not only clearly differs mechanistically from the other σ factors so far characterized, but also from other metal regulators.
| Copper exerts a dual effect on living organisms. It is essential for life, but an excess provokes cell damage, forcing cells to maintain a regulated homeostasis for this metal. These two antagonistic biological effects of copper are clearly illustrated by two human genetic disorders, Menkes syndrome and Wilson disease, caused by deficiency or accumulation of this metal, respectively. Myxococcus xanthus, a soil-dwelling bacterium, also has to cope with changes in copper concentration in its environment. The large genome of this myxobacterium encodes many genes involved in copper homeostasis, all of which are differentially regulated, indicating that many regulators participate in copper homeostasis in this prokaryote. Here, we identify one of these regulators (CorE), which belongs to the family of the extracytoplasmic function (ECF) σ factors. We demonstrate that CorE represents a novel group of ECF σ factors and of metal regulators, because its activity is modulated by the redox state of copper. This ability resides in a Cys-rich domain, which has also been found in other σ factors of different bacterial phyla. Therefore, we propose that CorE is the first member of a mechanistically new group of ECF σ factors.
| Myxococcus xanthus is a soil-dwelling δ-proteobacterium of the group of myxobacteria used as a model to study multicellular behavior and differentiation, because it exhibits a complex developmental cycle triggered by starvation [1]. However, M. xanthus cells not only have to adapt their metabolism and behavior to changing nutritional concentrations, but also to other parameters, such as metals.
Copper is a transition metal that functions as an ideal biological cofactor due to its ability to alternate between the redox states Cu(I) and Cu(II). However, copper also generates reactive oxygen species that cause cell damage [2]. This duality forces organisms to maintain a strict homeostasis for this metal. The most representative examples of the effect of disturbances in copper homeostasis are two inherited human disorders, Wilson disease and Menkes syndrome, which are directly linked to overload and deficiency of this metal, respectively [3].
Copper is required by prokaryotes in trace amounts because it is used as a cofactor by a few proteins. Hence, most bacterial homeostatic mechanisms are devoted to conferring resistance to this metal. The most common mechanisms are copper-transporting P1B-type ATPases, copper chaperones, multicopper oxidases (MCOs), and Cus systems [4]. In bacteria such as Escherichia coli, one of each of these elements is encoded in the genome [4]. In other bacteria, the homeostatic mechanism is even simpler, consisting of two P1B-type ATPases and one chaperone (Synechocystis PCC6803, Enterococcus hirae, and Lactococcus lactis), or one ATPase and one chaperone (Bacillus subtilis) [4], [5]. In contrast, the large M. xanthus genome encodes a large number of paralogous genes to confer copper tolerance: three MCOs, at least two Cus systems, and three P1B-type ATPases, as well as the genes required for the biosynthesis of carotenoids [6]–[9]. This gene redundancy indicates that copper homeostasis in this myxobacterium is more complex than in other prokaryotes. All of these genes have been shown to be differentially regulated [6]–[9], suggesting that this sophisticated network must be finely regulated by specific metal sensors.
One of the signal transduction mechanisms used by bacteria to direct gene expression at the transcriptional level in response to stress signals is represented by alternative σ factors [10]. The largest group of alternative σ factors is the ECF (extracytoplasmic function) family, which corresponds to group 4 of the σ70 proteins [11]. ECF σ factors are small proteins, quite divergent in sequence, that contain only two regions (σ2 and σ4) required for interaction with the RNA polymerase core enzyme and recognition of the promoter [12]. Their ability to promote transcription relies on a protein that is normally cotranscribed with the σ factor, named anti-σ factor. In the absence of external signals, ECF σ factors are sequestered by their cognate anti-σ. After detecting the specific stimulus, the anti-σ factor releases the σ subunit, which can then promote gene expression after recruitment of the core RNA polymerase [13]–[16].
In this report, we identify a novel metal sensor involved in copper homeostasis in M. xanthus named CorE (for copper-regulated ECF σ factor). We demonstrate that CorE requires copper in order to bind to DNA and that its activity is modulated by the redox state of this metal. According to these data, we propose a new group of ECF σ factors, defined by a Cys-rich domain (CRD) located at the C terminus of the protein, which is essential for activation and inactivation of the protein.
Most M. xanthus genes involved in copper homeostasis are located in the genome in two clusters [8]. In copper region 2, and next to the MCO cuoB, a gene encoding a protein with high similarity to ECF σ factors was found (MXAN_3426), suggesting that it could regulate the expression of genes involved in conferring copper resistance. This σ factor has been designated as CorE. The analysis of the CorE sequence has revealed a domain architecture with the conserved regions σ2 (sigma70_r2, PF04542; E-value of 3.2e-14) and σ4 (sigma70_r4_2, PF08281; E-value of 4.3e-06) typical of this type of σ factors [11], [12].
To determine the role of CorE in copper homeostasis, a strain harboring a corE-lacZ fusion was constructed, and the analysis of this strain revealed that corE was up-regulated by copper (Figure 1A). Additionally, an in-frame deletion mutant (ΔcorE) was also generated, and the phenotypic analysis of this strain confirmed that this regulator conferred copper tolerance (Figure 1B).
To identify genes regulated by CorE, plasmids containing fusions between the genes that have so far been involved in copper and/or other metal homeostasis in M. xanthus and lacZ were electroporated into the ΔcorE mutant. When the expression profiles of these genes in the mutant were compared with those exhibited in the wild-type (WT) strain, it was observed that only the MCO cuoB and the P1B-type ATPase copB remained undetectable in the ΔcorE background in the presence of copper (Figure 2A and Figure S1), indicating that they are regulated by this σ factor. Interestingly, these two M. xanthus genes exhibit a characteristic expression profile, with a peak at 2 h after the addition of exogenous copper.
As ECF σ factors are usually autoregulated, corE expression was analyzed in the ΔcorE mutant. The results obtained showed that this σ factor is only partially responsible for its own up-regulation by copper, especially in the early stages after metal addition. However, some up-regulation by the metal still remains in the mutant (Figure 1A, red lines), indicating that although cuoB and corE are very close in the genome (Figure S2), their regulation exhibits some differences.
The comparison and analysis of the upstream regions of cuoB and copB has allowed the identification of two similar sequences that could function as the promoter elements recognized by CorE (Figure S3), one located upstream of copB, and the other upstream of a third gene genetically linked to cuoB and corE which encodes an outer membrane efflux protein (MXAN_3424). A manual search for homologous sequences to this putative CorE-binding site in the M. xanthus copper regions 1 and 2 [8] revealed the presence of two other matches, one upstream of the gene for the P1B-type ATPase CopA and the other upstream of the gene identifier MXAN_3427, which encodes a protein with a heavy-metal-associated domain (PF00403, with an E-value of 4.7e-14). To corroborate that these two genes are regulated by CorE, plasmids harboring fusions between these two genes and lacZ were introduced into the WT and ΔcorE backgrounds and β–gal specific activity was determined in these strains. The results obtained revealed that the gene for the heavy-metal-associated protein exhibits an expression profile in the WT strain very similar to those of cuoB and copB after copper supplementation (compare Figure 2A and 2B). Up-regulation by copper was completely eliminated in the ΔcorE mutant, demonstrating that this gene is also part of the CorE regulon. In the case of copA the result was less clear. The expression profile of copA in the WT strain clearly differs from those exhibited by the CorE-regulated genes (compare panel C with panels A and B in Figure 2), and instead of a peak at 2 h, a plateau is reached 24 h after copper addition. Accordingly, the expression profile of copA in the ΔcorE mutant is quite similar to that of the WT strain (Figure 2C). However, when the expression level of copA in these two strains was analyzed with greater precision at short intervals (Figure 2D), it could be observed that the rapid induction of this gene obtained in the WT strain was no longer observed in the mutant. This result suggests that copA is subject to double regulation by CorE and another unidentified copper-dependent regulator. Nevertheless, further work will be required to unambiguously demonstrate that copA is regulated by CorE. Finally, using the consensus sequence of the promoters for these four genes, we tried to determine which other genes could also be under control of CorE. By using the approach described in Materials and Methods, another 13 similar sequences were identified in the M. xanthus genome (Figure S3). However, the fact that only two of them contain the seven invariable residues found in the other promoters, and that none of the proteins encoded by the genes located downstream of these sequences exhibit similarities to other proteins known to be involved in copper handling and trafficking, preventing us from drawing the conclusion that they are indeed regulated by CorE.
As the activation of CorE by copper could be caused either by the general oxidative stress induced by this metal or by the direct binding of the protein to copper in either of its two redox states, cuoB expression in the WT strain was tested in the presence of several concentrations of the oxidants hydrogen peroxide and diamide, and the Cu(II) mimetic divalent metals Cd2+, Ni2+, and Zn2+. Similarly, Ag+ was used to mimic Cu(I). The results obtained revealed that only Cd2+ and Zn2+ could induce cuoB expression (Figure 3A and Figure S4). The fact that Ni2+ does not up-regulate cuoB is not surprising, because the same metals cannot always mimic the copper effect. As an example, the M. xanthus P1B-type ATPase copA has been reported to be induced by copper, Ni2+ and Co2+, but not by Zn2+ [9]. It is notable that the expression levels obtained with Cd2+ and Zn2+ were not only much lower than with copper, but also that the expression profiles were different. In the case of Cd2+, no peak was observed at 2 h; instead, a plateau was reached 24 h after metal supplementation (Figure 3B). Although Zn2+ also yielded a rapid cuoB induction, the peak at 2 h was not as evident as in the case of copper (Figure 3C). cuoB up-regulation by Cd2+ and Zn2+ is also dependent on CorE (Figure 3B and 3C). These data indicate that Cu(II) is the redox state of copper that activates CorE. It should be noted that the Cd2+ and Zn2+ concentrations needed to observe a clear cuoB induction are close to the maxima that M. xanthus cells can tolerate [7], while 0.3 mM copper has almost no effect on myxobacterial growth (Figure 1B). It should also be noted that the addition of metals to the media not only alters the growth rates of the cultures, but also inhibits cell motility, explaining why the morphology of the cell spots is not the same in all of the media tested.
Many ECF σ factors function with a cognate anti-σ which is genetically linked to the σ subunit [11]–[16]. Analyses of the genes located in the proximity of corE revealed that they encode either proteins located in the periplasmic space or in the outer membrane, or that they exhibit striking similarities to well-characterized proteins involved in specific functions, suggesting that no anti-σ factor is cotranscribed with corE. However, the possibility remained that it could be encoded in some other region of the M. xanthus genome. To test for the existence of an anti-σ factor, a strategy was designed consisting of the over-expression of corE [17]. If CorE were present in higher quantities than an unidentified anti-σ factor, it would be released from the antagonistic effect of the anti-σ, and cuoB should be expressed even in the absence of any stimulus. To follow this approach, corE was cloned under control of the oar promoter and introduced into the ΔcorE mutant harboring cuoB-lacZ to facilitate the analysis of cuoB expression. The oar promoter allows genes to be expressed constitutively at high levels [18]. As a control, a corE' cuoB-lacZ strain was also constructed, in which the corE gene was under control of its own promoter (Figure S2 displays the cuoB-lacZ fusions used in this study). Quantitative analyses of cuoB expression in both strains reported no expression of this gene in the absence of copper (Figure 4A), indicating that an excess of CorE was not sufficient to activate the transcription of cuoB. To corroborate that CorE expressed under the oar promoter was functional, copper was added to the media. In this case, up-regulation of cuoB was observed in both strains and with similar expression profiles (Figure 4B). Finally, to confirm that corE was over-expressed when cloned under the oar promoter, we constructed the same two strains described above but introducing a His tag at the N terminus of CorE (hCorE'). Western blot analyses using antibodies against the His tag confirmed that corE was indeed expressed at very high levels in the absence as well as in the presence of copper (Figure 4C and 4D). CorE migrates as a double band, which must correspond to different forms of the protein. Activity of the hCorE' protein was further tested by following cuoB expression. The results obtained indicated that the proteins holding the His tag could promote cuoB transcription in the same manner as the native ones (Figure S5). Although it cannot be completely ruled out that a cognate anti-σ factor for CorE is encoded in the M. xanthus genome, all of these results indicate that CorE functions in a different manner from the one reported for the other characterized ECF σ factors.
The fact that the over-expression of corE did not lead to the induction of cuoB unless copper was added to the medium suggested that CorE might require the binding of copper to promote transcription. Hence, the ability of CorE to bind DNA in vitro was tested by using electrophoretic mobility shift assays. CorE was expressed in E. coli with an N-terminal His tag and purified by affinity chromatography. Additionally, a 265-bp fragment containing the copB promoter was amplified and labeled with 32P to be used as a probe. As shown in Figure 5, an electrophoretic mobility shift was only observed in the reaction mixture containing copper and bathocuproine disulfonic acid (BCS), a specific chelating agent for Cu(I) [19], [20]. These results not only confirm that CorE uses copper as a cofactor, but also suggest that Cu(I) prevents CorE from binding to DNA, and hence, that CorE-Cu(II) is the active form of this σ factor. This is also supported by the fact that only divalent metals can mimic the effect of copper on cuoB up-regulation. No other σ factor has so far been reported to require any metal to bind DNA.
The expression profiles of the CorE-regulated genes exhibit a peak around 2 h after copper addition (Figure 2A and 2B), in spite of the fact that corE expression is maintained for 48 h (Figure 1A). This observation could be explained by proteolysis of the σ factor. To investigate this option, Western blot analyses were carried out using the hcorE' cuoB-lacZ strain. The data shown in Figure 6A demonstrate that CorE was stable for 24 h after copper addition. Another explanation could be that CorE underwent a cycle of activation/inactivation, whereby the regulator would only be in the active form for a limited period of time. As shown in Figure 5, to obtain binding of CorE to DNA, the reaction mixture must include not only copper, but also a chelating agent for Cu(I). Moreover, only other divalent metals can mimic the copper effect on cuoB up-regulation (Figure 3 and Figure S4). These data suggest that the redox state of copper could be the key in this process. To investigate this possibility, the expression of cuoB was assayed in vivo in conditions that favor the formation of Cu(I) and Cu(II). As shown in Figure 6B and 6C, cuoB up-regulation could only be observed when copper was added to the medium. However, the maximum expression levels were diminished when the reducing agent ascorbate was also included in the medium to favor the formation of Cu(I) (Figure 6C, brown line). Similarly, the addition of Ag+, which mimics Cu(I), also yielded expression levels lower than those obtained with only copper (Figure 6C, green line). In contrast, when copper was added with the Cu(I) chelators BCS or bicinchoninic acid (BCA) [19], [20], cuoB expression was around three times that of the control (Figure 6C, blue versus red and black lines). Moreover, the addition of copper with tetrathiomolybdate (TTM), a chelator of Cu(I) and Cu(II) [21], decreased the up-regulation mediated by this metal to a very basal level (Figure 6C, orange line). In contrast, when these three chelators were tested with Zn2+ as the inducer, the expression levels of cuoB were diminished as the concentrations of all the chelators increased (Figure S6), due to the fact they can also chelate Zn2+, although to a much lesser extent than copper. According to all these data, CorE requires copper for activation, and it only acquires an active conformation in the presence of Cu(II), while the reduced state of the metal leads to an inactive conformation. This notion agrees well with the lack of a peak when up-regulation of cuoB is achieved by Zn2+ and Cd2+, which are metals with only one redox state (Figure 3B and 3C).
To confirm the results obtained in vivo, the DNA-binding assay was carried out again including Ag+ or TTM in the reaction mixtures. As shown in Figure 7, these two additives overrode the electrophoretic mobility shift achieved by the addition of copper and BCS. It should be reminded that Ag+ mimics Cu(I) and that TTM chelates Cu(II). All of the data presented in this section demonstrate that CorE activity is modulated by the redox state of copper.
This mechanism of action implies that Cu(II) must be available in the cytosol during the next 2 h after copper supplementation. Although it is assumed that all the copper in the reducing environment of the cytoplasm is present as Cu(I) under normal circumstances [2], [22], it is also expected that the cytoplasm will become more oxidizing in the presence of agents such as copper [23], favoring the formation of Cu(II) until the reducing conditions are restored by the participation of the elements involved in copper detoxification. Furthermore, as free copper in the cells is estimated to be less than one atom per cell [24], it is plausible to speculate that CorE functions with an unidentified Cu(II)-specific metallochaperone, which would ferry the cupric form through the cytoplasm to activate this σ factor. Such an activator working upstream of CorE would explain why the expression levels of cuoB do not increase when corE is over-expressed (Figure 4). However, another possible explanation for this observation could be that CorE aggregates when produced in large amounts.
One paradox is the fact that CorE is activated by Cu(II) and inactivated by Cu(I), while genes under its control encode proteins, such as CopA, CopB, and CuoB, that utilize Cu(I) as a substrate [7], [9]. Nevertheless, this contradiction can be explained by considering two facts: i) Out of the two redox states of copper, Cu(I) is the most toxic form [2]. As CorE-regulated genes represent the first protective barrier against the deleterious effect of copper (please note that these genes are rapidly up-regulated after copper addition, as shown in Figure 2 and Figure S1) and this protein is activated by Cu(II), it is plausible to speculate that copper will initially get into the cytoplasm in the form of Cu(II), activating CorE, and preparing the cells to act on Cu(I) as soon as it appears. At this point, the CorE regulon will be inactivated by the presence of Cu(I) in the cytoplasm. ii) If the presence of copper persists in the environment, M. xanthus cells will obtain protection against the metal by means of at least two other mechanisms (first, by the P1B-type ATPase CopA and the MCO CuoA, and later, by the Cus2 and Cus3 systems), which are sequentially induced after copper addition [7]–[9] (see also Figure 2 and Figure S1).
Although many bacterial transcriptional regulators need metal to bind DNA [25], [26], none of them have been reported to be modulated by the redox state of the metal. Moreover, those that function with copper show selectivity for Cu(I) [24], [27]. Hence, CorE represents a novel type of bacterial copper sensor.
CorE contains a short C-terminal extension after the σ4 domain consisting of 38 residues named CRD. Six of these residues are Cys. As different arrangements of Cys have been proved to be key elements in several metal-binding proteins [22], [27], [28], we tried to determine whether CRD was involved in the activation/inactivation of CorE mediated by copper. An M. xanthus in-frame deletion mutant was constructed in which most of the CRD region was deleted. This strain, designated as ΔcorECRD, encoded a protein containing the two domains σ2 and σ4 of CorE, but none of the six Cys of CRD. To analyze the activity of CorECRD, the two fusions cuoB-lacZ and copB-lacZ were introduced in this mutant and β-gal activity was assayed in the absence and in the presence of copper. The data obtained revealed that neither cuoB or copB were up-regulated by copper (data not shown), a result identical to that shown in Figure 2A, when the entire corE gene was deleted. These data demonstrate that CRD is essential for the copper-dependent transcription of the genes controlled by CorE.
To determine which Cys are involved in CorE activity, each residue was individually mutated to an Ala by site-directed mutagenesis. The six mutated genes were introduced into the ΔcorE strain harboring the fusion cuoB-lacZ. The effect of the mutations was evaluated by analyzing the expression of cuoB in the absence and in the presence of copper. The results obtained showed different patterns. Mutations C181A and C206A exhibited transcription profiles very similar to those of the WT (Figure 8A), although some small differences regarding the maximum expression levels and timing were observed, indicating that these residues play a minor role in CorE activity.
More severe effects were obtained with the mutations C192A and C194A. In these mutants, the expression levels of cuoB in the absence of copper were higher than in the WT (Figure 8B, dashed lines), suggesting that both Cys play some role in CorE inactivation. Moreover, although cuoB expression was up-regulated by copper in both mutants, the rapid induction and the peak exhibited by the WT at 2 h were not replicated (Figure 8B, continuous lines). The effect of the mutation C189A was even more drastic, because no expression was observed in the absence of copper and the up-regulation by the metal was almost completely non-existent (Figure 8C and 8D). Accordingly, it can be concluded that these three residues are important in the CorE activation process.
Cys184 was clearly required for CorE inactivation, because mutation C184A yielded a constitutive expression in the absence of copper (Figure 8E) and the addition of copper provoked a rapid up-regulation of cuoB. Interestingly, the expression level did not peak at 2 h, but kept increasing until it reached a plateau at 24 h (Figure 8F).
The effect of each point mutation in cuoB expression was also analyzed in cells grown on media containing copper plus BCS or silver (Figure S7). Mutations C181A and C206, which in the presence of copper yielded expression profiles similar to that of the WT (Figure 8A), also exhibited higher expression levels in the presence of copper plus BCS, and lower levels with copper plus Ag+ (Figure S7). In the case of substitutions C189A, C192A, and C194A, BCS and Ag+ barely affected the expression levels obtained with only copper (Figure S7). However, it should be reminded that these three residues are important for activation, and that cuoB up-regulation by copper was impaired in these mutants (Figure 8B and 8D), suggesting that these three proteins might only have a limited affinity for the metal. Surprisingly, however, the protein with the mutation C184A (the most important residue in CorE inactivation as shown in Figure 8E and 8F), can still be modulated by the two redox states of copper (Figure S7). This result indicates that some other residues must also be involved in the inactivation of CorE. As mentioned above, two of them could be Cys192 and Cys194, because mutations C192A and C194A yield constitutive expression of cuoB. However, it is expected that other residue(s) of CRD might also be required for the proper functioning of the protein (see below).
Taken as a whole, the results demonstrate that at least four of the Cys of CRD form a coordination environment for copper. This domain is able to recognize copper and sense its redox state, allowing the binding of CorE to DNA to activate transcription in those conditions that favor the formation of Cu(II), and inactivating the σ factor in those that favor the formation of Cu(I). However, how exactly CorE distinguishes between Cu(I) and Cu(II) is not easy to predict, because all of the residues identified so far that modulate the activity of CorE are Cys. Although Cys are able to coordinate Cu(I) and Cu(II), they require the presence of other amino acids, such as His, Asp, Glu, or Met to exert this function [22], [29]. Accordingly, it is expected that other residues also participate in the coordination of copper in either of the two redox states. Moreover, thiols are known to allow different types of modifications in an oxidative environment [30]. The exact modification of each individual Cys might also be crucial in the CorE activation/inactivation process. Further genetic, biochemical, and structural studies will be required to elucidate this intriguing question.
The role of CRD in CorE resembles the function of the anti-σ domain present in many anti-σ factors [31]. Anti-σ domains require Zn2+ binding to sequester their cognate σ factor. However, the anti-σ domain and CRD differ in many aspects: i) CRD is an extra portion of the σ factor; ii) elimination of CRD does not activate the σ factor; and iii) CRD senses the redox state of copper to activate or inactivate the σ factor.
BLASTP analyses have allowed the identification of 21 ECF σ factors with CRD, which are distributed in only four phyla. Fourteen belong to Proteobacteria (9 α and 5 δ), four to Acidobacteria, two to Verrucomicrobia, and one to Nitrospira (Figure 9). As in the case of CorE, anti-σ factors are not linked to any of these σ factors.
The alignments of these CRDs have revealed that only 4 Cys (corresponding to residues 181, 184, 192, and 194 in CorE) are absolutely conserved among these σ factors (Figure 9A). Surprisingly, Cys181, whose mutation causes a minor effect on CorE activity, is conserved in all these regulators. In contrast, Cys189, which is the main residue in CorE activation by copper, is only present in 11 σ factors. Interestingly, however, several of the strains with σ factors that conserve this Cys exhibit some synteny in the regions where they are encoded. The surrounding genes encode proteins with high similarities to others known to be implicated in copper handling and trafficking (Figure 9B).
Due to the diversity of the ECF σ factors, Staroń et al. [13] have proposed a classification of this family of regulatory proteins into 44 groups based on sequence similarities and domain architectures. However, CorE did not fit into any of the groups they defined and it was excluded from this classification. The data presented in this report support the notion that a new group should be added to the list, which will include the 21 ECF σ factors that contain CRD.
So far, seven families of metal de-repressors, metal co-repressors, and metal activators are known [25], [26], [32], all of which clearly differ mechanistically from CorE. Hence, elucidation of the exact mode of action of CorE will offer new insights into our current knowledge of metal sensors. Moreover, identification of the factor(s) working upstream of CorE will also help to elucidate how this type of σ factors works and how the trafficking of metals in the bacterial cytoplasm occurs. Finally, characterization of CorE-like proteins identified in other bacteria will also contribute to understanding the role, mechanism of action, and distribution of this novel type of regulators.
Genotypes of the bacterial strains and plasmids used in this study are listed in Table S1 and Table S2, respectively. M. xanthus was grown in CTT medium at 30°C, supplemented with the additives indicated in each figure, as previously described [6]. E. coli was grown on Luria-Bertani (LB) medium at 37°C [33].
The methodologies used for obtaining the in-frame deletion mutants and the transcriptional lacZ fusion strains used in this study are the same as previously reported [7]. To generate the corresponding plasmids (listed in Table S2), the desired fragments were amplified by polymerase chain reaction (PCR), using WT chromosomal DNA as a template, the oligonucleotides listed in Table S3 as primers, and the high-fidelity polymerase PrimeSTAR HS (Takara) [33]. PCR products were ligated to vectors pBJ113 and pKY481 [34], [35] to generate in-frame deletion mutants and lacZ fusions, respectively. Plasmids were always introduced into M. xanthus strains by electroporation to obtain integration into the chromosome by homologous recombination. Southern blot analyses were carried out to confirm the proper recombination events. β-gal specific activity in cell extracts obtained by sonication of the strains harboring lacZ fusions was determined as previously described [7], and it is expressed as nmol of o-nitrophenol produced per min and mg of protein. Measurements shown are the averages of data from triplicate experiments.
Appropriate oligonucleotide pairs (Table S3) were used to amplify by PCR an 817-bp fragment upstream of the oar gene (MXAN_1450) using M. xanthus chromosomal DNA as a template [33]. Simultaneously, corE was also amplified by PCR. A BamHI site was introduced at the start codon of oar in frame with another BamHI site introduced at the start codon of corE. Both PCR products were cloned in a vector derived from pUC19 in which the ampicillin-resistance gene was substituted by one that encodes resistance to tetracycline (Tetr). The resulting plasmid, pNG06, was introduced by electroporation into an M. xanthus strain with the genotype ΔcorE cuoB-lacZ, and several kanamycin-resistant (Kmr) and Tetr colonies were analyzed by Southern blot to confirm the proper recombination event. β-gal specific activity was determined to quantify cuoB expression. As a control, the plasmid pNG00 was constructed, in which corE was cloned under control of its own promoter. This plasmid was also electroporated into the ΔcorE cuoB-lacZ to restore corE at its original genomic location (see Table S1 and Figure S2). To corroborate that CorE was being over-produced under the constitutive oar promoter, we constructed the same strains described above, but introducing an N-6His tag upstream of CorE, to obtain the strains hcorE' cuoB-lacZ and oar-hcorE' cuoB-lacZ (Table S1). Briefly, the corE gene with an N-6His tag was amplified with appropriate oligonucleotide pairs (Table S3) using pETTOPOCorE plasmid (see below) as a template. The PCR product obtained was cloned under control of the oar promoter and its own promoter as above, obtaining plasmids pNG08 and pNG05, respectively. Plasmids were introduced into the ΔcorE cuoB-lacZ strain, and Kmr Tetr colonies were also analyzed by Southern blot hybridization.
The corE gene was amplified by PCR using the M. xanthus chromosome as a template and the primers CorEcTopoR and CorEcTopoF (Table S3). The PCR product was cloned into pET200/D-TOPO using a Champion pET Directional TOPO Expression Kit supplied by Invitrogen to create pETTOPOCorE. The absence of unwanted mutations in the insert was confirmed by DNA sequencing. As a result, CorE contained an N-6His tag at the N terminus, and this protein has been named hCorE.
The resulting plasmid was used to transform the strain E. coli BL21 Star(DE3). The transformed cells were grown in LB medium containing 25 µg/ml kanamycin at 37°C until the optical density at 600 nm (OD600) reached 0.7 to 0.8. Induction was performed by the addition of 1 mM of isopropyl-β-D-thiogalactopyranoside. Induced cultures were incubated with shaking at 37°C for 6 h. One liter of the cell culture was collected by centrifugation and resuspended in 20 mM Tris-HCl (pH 7.5) containing the protease inhibitors leupeptine and antipain (2 µg/ml each), 10 µg/ml DNAse I, and 5 mM MgCl2. Cells were disrupted in a French pressure cell (at 9000 psi), followed by centrifugation (13000×g for 30 min at 4°C) to remove cell debris. The resulting soluble extract was loaded onto a HisTrapHP column (bed volume 5 ml; GE Healthcare) equilibrated with 20 mM Tris-HCl (pH 7.5) containing 0.5 M NaCl and 30 mM imidazole. Elution was carried out with a linear imidazole gradient (30–250 mM) in the same buffer. Protein fractions were analyzed by using SDS-PAGE. Those fractions containing hCorE were pooled, concentrated by ultrafiltration (cutoff of 10 kDa), and equilibrated to 20 mM Tris-HCl (pH 7.5). The purified hCorE protein content was determined by the Bio-Rad protein assay kit as specified by the manufacturers, using bovine serum albumin as standard. The purity of the samples was higher than 90%.
This methodology was used to detect hCorE either in M. xanthus or E. coli extracts. Cells were disrupted by sonication and centrifuged to remove cellular debris. Proteins were separated by SDS-PAGE and transferred onto a membrane of Immobilon-P at 0.8 mA/cm2 for 1.5 hours. hCorE was detected with an anti-His G-AP antibody (Invitrogen), which is conjugated with alkaline phosphatase, using nitro-blue tetrazolium chloride and 5-bromo-4-chloro-3′-indolyphosphate p-toluidine as substrates, following the instructions specified by the manufacturer.
A DNA fragment containing an upstream sequence of the predicted ribosome-binding site of copB was amplified from the M. xanthus genome using primers 3422EMSA265F and 3422EMSA265R (Table S3). After purification, the 265-bp PCR product was labeled with T4 polynucleotide kinase (MBI Fermentas) and [γ-32P]ATP, and purified through a ProbeQuant G-50 Micro Column (GE Healthcare). Binding reactions contained 20 mM Tris-HCl (pH 7.5), 2 mM MgCl2, 0.25 mg/ml of bovine serum albumin, 0.5 mM dithiothreitol, 15% glycerol (v/v), 40 mM KCl, 0.5 nM of labeled DNA (13000 cpm), and a 500-fold molar excess of competitor DNA (polydIdC). When indicated, 500 nM of hCorE protein, 0.1 mM CuSO4, 0.05 or 0.1 mM BCS, 0.1 mM AgNO3, or 0.1 mM TTM were also added to the reaction mixtures. After incubation for 10 min at 30°C, the mixtures were loaded onto a pre-run 5% polyacrilamide gel and run at 100 V for 1 h. The gel was dried under vacuum and exposed to an autoradiography film at −80°C.
Single amino acid substitutions in the CRD of CorE were generated using the QuikChange II site-directed mutagenesis kit (Stratagene). Plasmid pNG00 (containing the WT corE sequence) was used as a template. The primers were designed using the QuikChange Primer Design Program (http://www.genomics.agilent.com/). Oligonucleotides used to generate the six point mutations are listed in Table S3. The presence of the desired mutations in the resulting plasmids pNG181, pNG184, pNG189, pNG192, pNG194, and pNG206 (carrying the mutations C181A, C184A, C189A, C192A, C194A, and C206A, respectively), and the absence of unwanted mutations in other regions of the inserts were confirmed by DNA sequencing. These plasmids were electroporated into M. xanthus JM51EBZY (ΔcorE cuoB-lacZ) to obtain strains SDM181EBZY to SDM206EBZY (Table S1).
Genes encoding ECF σ factors with a CRD in the prokaryotes were identified by BLASTP analysis of all the genome sequences deposited in the database of the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/genomes/lproks.cgi) and the DOE Joint Genome Institute (http://www.jgi.doe.gov/). All the sequences obtained with an E-value<2e-10 that conserved at least 4 Cys in the C terminus were back-searched against Pfam (http://pfam.sanger.ac.uk/) [36] to unequivocally verify that they matched the σ2 (PF07638) and σ4 (PF08281) regions conserved in all the ECF σ factors [11], [12]. Protein sequence alignments of the 21 σ factors with CRD identified were performed using ClustalX [37]. The graphic representation of the multiple sequence alignment was adjusted and colored manually using the model generated at ESPript.cgi Version 3.06 CGI 3.05 (http://espript.ibcp.fr/ESPript/cgi-bin/ESPript.cgi). For synteny determinations, two upstream and two downstream predicted proteins from CorE were initially aligned with the other 20 predicted proteomes using the BLASTP program (E-value<1e-4). We then manually analyzed the gene organization of positive matches. In the case of conservation, we extended the BLASTP searches to other genes within the same region as already described by Pérez et al. [38].
First, the upstream regions of copB and cuoB were manually analyzed to find the sequence AAC, which is well conserved in the −35 regions of other known ECF σ-factor promoters in M. xanthus and other bacteria [39]. Alignment of the sequences found permitted the identification of two regions that could function as the CorE-binding site (Figure S3). Next, homologous sequences to these ones were manually searched in all the upstream regions of the genes of the copper regions 1 and 2 of the M. xanthus genome [8]. By using this strategy we found, in copper region 2, two other putative CorE-dependent promoters upstream of MXAN_3427 and MXAN_3415 (which encodes the P1B-type ATPase CopA). Experimental approaches demonstrated that the expression of these two genes is dependent on copper and that they are regulated by CorE. The alignment of the four sequences was used to define the CorE-binding motif. The consensus sequence of the −35 region given in the IUPAC code (defined by Nomenclature Committee of the International Union of Biochemistry) was used to analyze the whole M. xanthus genome at the Virtual Footprint server (http://prodoric.tu-bs.de/vfp/) [40] to determine which other genes might be part of the CorE regulon. 754 positive sequences were obtained with a maximum of two mismatches with respect to the defined consensus. All the sequences were manually examined for the proper strand orientation and the conservation of the invariant residues observed in the −35 region. The resulting positive matches were again screened to identify a conserved G in the −10 region, maintaining a distance of 16–18 residues from the AAC of the −35 region. Only 13 positive sequences were finally selected (Figure S3).
|
10.1371/journal.ppat.1002502 | HCMV Targets the Metabolic Stress Response through Activation of AMPK Whose Activity Is Important for Viral Replication | Human Cytomegalovirus (HCMV) infection induces several metabolic activities that have been found to be important for viral replication. The cellular AMP-activated protein kinase (AMPK) is a metabolic stress response kinase that regulates both energy-producing catabolic processes and energy-consuming anabolic processes. Here we explore the role AMPK plays in generating an environment conducive to HCMV replication. We find that HCMV infection induces AMPK activity, resulting in the phosphorylation and increased abundance of several targets downstream of activated AMPK. Pharmacological and RNA-based inhibition of AMPK blocked the glycolytic activation induced by HCMV-infection, but had little impact on the glycolytic pathway of uninfected cells. Furthermore, inhibition of AMPK severely attenuated HCMV replication suggesting that AMPK is an important cellular factor for HCMV replication. Inhibition of AMPK attenuated early and late gene expression as well as viral DNA synthesis, but had no detectable impact on immediate-early gene expression, suggesting that AMPK activity is important at the immediate early to early transition of viral gene expression. Lastly, we find that inhibition of the Ca2+-calmodulin-dependent kinase kinase (CaMKK), a kinase known to activate AMPK, blocks HCMV-mediated AMPK activation. The combined data suggest a model in which HCMV activates AMPK through CaMKK, and depends on their activation for high titer replication, likely through induction of a metabolic environment conducive to viral replication.
| Human Cytomegalovirus (HCMV) is a ubiquitous human pathogen that is a major cause of birth defects. HCMV can also cause severe disease in immunocompromised individuals including transplant recipients, leukemia patients and those infected with HIV. It is clear that upon infection, HCMV takes control of numerous cellular processes that are important for the virus to generate the next round of infectious virions. We have previously found that upon infection, HCMV reprograms the metabolic activity of the host-cell. Here, we find that this metabolic reprogramming largely depends on the viral activation of a cellular protein called the AMP-activated protein kinase (AMPK). AMPK is a central regulator of cellular energy production that is typically only activated when cellular energy stores are very low. Our results indicate that HCMV-mediated activation of AMPK is necessary to flip the metabolic switch thereby driving host-cell metabolic activation and viral replication. As inhibition of AMPK blocked viral replication, and had little impact on uninfected host-cell metabolism, targeting AMPK could have therapeutic potential to treat HCMV-associated disease.
| Upon infection, viruses must create a cellular environment conducive to viral replication. While there are many different aspects of this virally-induced environment, a critical component of this cellular reprogramming is the diversion of cellular resources such as energy and molecular building blocks to the production of viral progeny. Numerous viruses, ranging from small, non-enveloped RNA viruses to large enveloped DNA viruses have been reported to target the host-cell metabolic machinery [1]–[9]. This suggests that cellular metabolic function is a key virus-host interaction.
Human cytomegalovirus (HCMV), a member of the betaherpesvirus family, is a major cause of birth defects upon congenital infection as well as morbidity in immunosuppressed populations [10]–[12]. HCMV has been found to cause drastic changes to the host cell metabolic network upon infection, increasing the concentrations of select glycolytic enzymes, the steady state levels of glycolytic metabolites, and the fluxes through glycolysis and the TCA cycle [13]–[14]. Though the exact mechanisms through which HCMV induces glycolytic activation are not clear, it has recently been shown that virally-mediated activation of glycolysis can be blocked through inhibition of CaMKK and that HCMV infection induces the expression of the Glut4 transporter [15]–[16]. Here we explore the role of the AMP-activated protein kinase (AMPK) during HCMV replication and HCMV-mediated glycolytic activation.
AMPK is a heterotrimeric, serine-threonine kinase that functions as a major energy regulator for the cell. Low ATP levels result in increased concentrations of AMP through the action of adenylate kinase [17]. These increased AMP concentrations induce AMP binding to AMPK and subsequent stimulation of AMPK activity, primarily through either LKB1 or CaMKK-dependent phosphorylation [18]–[21]. Upon activation, AMPK works to restore the ATP pool by activating ATP-producing pathways while simultaneously inhibiting ATP-consuming pathways [22]–[23]. One pathway positively regulated by AMPK is glycolysis. Upon AMPK activation, glycolysis can be upregulated through several mechanisms. AMPK targets numerous key glycolytic enzymes including glucose transporters (Glut1 and Glut4), hexokinase and PFK-2, to increase glycolytic flux [23]–[26].
Here we show that AMPK activity is increased throughout viral infection relative to mock-infected fibroblasts. Additionally, high-titer HCMV replication requires activated AMPK as pharmaceutical or RNAi-based inhibition of AMPK severely attenuates the production of viral progeny. Consistent with a role in glycolytic activation, AMPK inhibition also leads to an attenuation of HCMV-induced glycolytic flux. These results suggest that AMPK is a critical cellular protein targeted by HCMV infection.
To determine if AMPK might be responsible for the metabolic induction observed during HCMV infection, we analyzed mock or HCMV-infected extracts for AMPK activity using a well described in vitro AMPK activity assay [27]. To help verify the specificity of the measured AMPK activity, we also performed assays in the presence of Compound C, a specific AMPK inhibitor [28]. At 24 h post-infection both mock and HCMV-infected cells exhibited similar amounts of AMPK activity (Fig. 1). Compound C treatment suppressed the observed AMPK activity to a background level of ∼2000 CPM, which was a consistent background level at all time points examined (Fig. 1). At later time points, i.e. 48 and 72 h post-infection, the AMPK activity associated with mock-infected fibroblasts fell significantly to background levels (Fig. 1). In contrast, the AMPK activity associated with HCMV-infected lysates increased over this time frame with much greater AMPK activity levels observed in HCMV-infected lysates than in mock-infected lysates at the same time points (Fig. 1). These results indicate that HCMV induces AMPK activity upon infection.
Previously we have found that HCMV infection increases the total levels of the fatty acid biosynthetic enzyme, acetyl-CoA carboxylase (ACC1) as well as the amount of phosphorylated ACC1 [29]. AMPK has been shown to phosphorylate ACC1 at Ser79, resulting in decreased ACC1 activity with the end-result being inhibition of fatty acid biosynthesis and conservation of ATP [17]. To determine if activated AMPK increases levels of phosphorylated ACC1 in MRC-5 fibroblasts, we treated cells with an AMPK activator, AICAR [30]. AICAR treatment resulted in substantial increases in the abundance of phosphorylated ACC (Fig. 2A), consistent with activation of AMPK. Subsequently, we analyzed the levels of total ACC1 and Ser79 phosphorylated ACC1 in the presence of the AMPK inhibitor, Compound C, during HCMV infection. At 24 h post-infection, HCMV-infected lysates contained more Ser79-phosphorylated ACC1 than mock-infected lysates (Fig. 2B). At this time, treatment with Compound C reduced the levels of Ser79-phosphorylated ACC1 to the levels found in uninfected cells, indicating that AMPK might be responsible for the increased abundance of phosphorylated ACC1 (Fig. 2B). At 48 and 72 h post-infection, treatment with Compound C reduced the levels of Ser79-phosphorylated ACC1 in HCMV infected lysates but also decreased the levels of total ACC1 (Fig. 2B). Using densitometry, the relative signal ratios of pACC to ACC in the HCMV infected lysates were examined (Fig. 2B). Treatment with Compound C had the largest impact on the pACC/ACC ratio at 24 hpi, reducing it ∼10-fold. At subsequent times post infection, Compound C reduced the pACC/ACC ratio at every time point, consistent with decreased phosphorylation, albeit to a much lesser extent than observed at 24 hpi (Fig. 2B). While the decreases in phospho-ACC1 relative to total ACC1 upon Compound C treatment are consistent with inhibition of AMPK-mediated phosphorylation of ACC1, Compound C treatment also impaired the HCMV-induced accumulation of total ACC1 suggesting that Compound C treatment could be impacting normal HCMV infection.
Previous reports indicate that HCMV infection induces the levels of the Glut4 glucose transporter and the tuberous sclerosis protein (TSC1), a negative regulator of mTOR signaling [15], [31]. Activated AMPK has been shown to increase Glut4 expression [32] as well as TSC1 levels through prevention of proteosome mediated TSC1 degradation [33]–[34]. Our results indicating that HCMV infection induces AMPK activity suggest the possibility that the induction of these proteins during HCMV infection may result from increased AMPK activity. As previously reported [31], in the absence of inhibitor treatment, HCMV infection has little impact on TSC1 levels at 24 h post-infection but significantly increased the levels of TSC1 at 48 h post-infection and 72 h post-infection (Fig. 2C). Treatment of HCMV-infected cells with Compound C substantially reduced the levels of TSC1 at all time points compared to DMSO treated controls (Fig. 2C) suggesting that AMPK activity is necessary for HCMV-mediated induction of TSC1 levels. As was reported previously [15], we also observed increases in Glut4 in HCMV-infected cells as compared to mock-infected cells at 48 and 72 h post-infection (Fig. 2C). Treatment with Compound C inhibited this induction of Glut4 levels (Fig. 2C) suggesting that AMPK activity is important for the viral induction of Glut4 expression. Taken together, our results indicate that HCMV-infection activates AMPK which in turn is necessary for the induction of TSC1 and Glut4 levels.
AMPK can be activated by phosphorylation at residue Thr172 mediated by either LKB1 or CaMKK [17]. As shown in Figure 2D, HCMV-infected extracts contained a higher level of total and phosphorylated AMPK then mock extracts at 24, 48 and 72 h post-infection. In both mock and HCMV-infected cells, the amount of Thr172-phosphorylated AMPK appeared to be the greatest at 24 h post-infection and subsequently declined as infection progressed (Fig. 2D). For uninfected cells, this decline in Thr172-phosphorylated AMPK correlated with the decreased AMPK activity observed in cellular lysates (Fig. 2D). For the HCMV-infected cells, the increase in AMPK activity observed as infection progressed did not correlate with the levels of Thr172-phosphorylated AMPK although the levels of total AMPK remained elevated. This combination of increased abundance of total AMPK and increased AMPK Thr172 phosphorylation likely contribute to the observed increases in AMPK activity during HCMV infection, although as AMPK is reported to be regulated by multiple phosphorylation events [35]–[37], other mechanisms of activation cannot be ruled out. Treatment of cells with Compound C did not appreciably impact AMPK Thr172 phosphorylation (Fig. 2D), not surprising given that Compound C inhibits AMPK through competitive inhibition at its ATP binding site [28]. Taken together, our results suggest that HCMV activates AMPK during infection likely in part due to an increase in both the phosphorylation at Thr172 as well as the total abundance of AMPK.
AMPK is a central metabolic regulator whose activation can activate glycolysis by targeting multiple steps within the glycolytic pathway including glucose uptake and phosphofructokinase activity [17]. To determine if AMPK is important for the induction of glucose import by HCMV, we treated mock or HCMV-infected fibroblasts with Compound C and analyzed glucose import using a radioactive glucose analog. Consistent with previous reports [15], [26], HCMV infection induced glucose uptake greater than 5-fold compared to mock-infected cells (Fig. 3A). Treatment with Compound C almost completely reversed this increase (Fig. 3A). Compound C had a negligible impact on glucose uptake in mock-infected fibroblasts (Fig. 3A). These data indicate that HCMV relies heavily on AMPK activity to activate glucose import whereas AMPK is not critical for glucose uptake in uninfected fibroblasts.
To further analyze how activation of AMPK contributes to HCMV-mediated glycolytic activation, we measured the rate of glycolytic labeling after treatment with Compound C using 13C-labeled glucose as a metabolic tracer. Specifically, utilizing LC-MS/MS we measured the rate of 13 C-fructose bisphosphate accumulation, a central glycolytic metabolite, after pulse with 13C-glucose. Treatment with Compound C led to an approximate 2-fold decrease in 13C-labeled FBP accumulation in HCMV-infected fibroblasts (Fig. 3B). In contrast, inhibition of AMPK had little impact on the labeling rate of mock-infected fibroblasts (Fig. 3B). Lastly, we measured how inhibition of AMPK impacted the most downstream glycolytic phenotype, accumulation of lactate in the media. Treatment with Compound C substantially reduced lactate secretion in HCMV-infected fibroblasts, but not mock-infected fibroblasts (Fig. 3C). In all of the glycolysis assays tested, the inhibition of glycolytic flux upon Compound C treatment was specific for HCMV-infected cells. This suggests that AMPK is important for HCMV-induced glycolytic activation, but does not contribute appreciably to the glycolytic rate in uninfected fibroblasts, which is consistent with AMPK's described role as a stress-induced metabolic regulator [17].
Addition of Compound C immediately following adsorption blocked HCMV-mediated activation of glycolysis (Fig. 3A–C). To determine whether this glycolytic activation was sensitive to AMPK inhibition after the establishment of infection, we treated cells with Compound C at 24 hpi, a time at which immediate early gene expression is peaking, early genes are being expressed and viral DNA replication is initiating [38]. Subsequently, we measured lactate excretion into the media from 40-58 hpi, and from 58–76 hpi. As compared to a DMSO-treated control, treatment with Compound C immediately following adsorption resulted in a ∼40% reduction in lactate excretion (Fig. 3D). Treatment of cells at 24 hpi resulted in a ∼20% reduction in lactate excretion whether measured from 40–58 or 58–76 hpi (Fig. 3D). These results suggest that while HCMV-mediated activation of glycolysis is more sensitive to AMPK inhibition at the very beginning of infection, the induction of glycolysis is still attenuated when AMPK is inhibited during an HCMV infection that has already been established.
Given that HCMV-infection induces AMPK activity (Fig. 1), and that AMPK activity is important for HCMV-mediated glycolytic activation (Fig. 3), we next tested if AMPK inhibition impacts viral replication. We treated fibroblasts with DMSO or two different concentrations of Compound C, and analyzed the production of viral progeny by plaque assay. Treatment with increasing concentrations of Compound C resulted in a dose-dependent decrease in viral titers. A greater than 20-fold defect in production of viral progeny was observed at 2.5 µM Compound C and a greater than 1000-fold defect was observed in cells treated with 5 µM Compound C (Fig. 4A). To exclude the possibility of toxicity from drug treatment, we also performed a Live/Dead assay which stains live cells green based on their esterase activity and stains the nucleic acids of dead cells red based on the breakdown of membrane integrity. Treatment with Compound C at the highest concentration (5 µM, Fig. 4B) resulted in little to no red staining in both mock- and HCMV-infected fibroblasts ( <1%) with ubiquitous green staining (>99%) suggesting that Compound C treatment is not toxic to MRC-5 fibroblasts up to a concentration of 5 µM. These results suggest that AMPK activity is important for HCMV replication and that inhibition of AMPK does not induce significant toxicity in MRC-5 fibroblasts.
We were interested in determining at what point during the infectious cycle AMPK activity might be required for high-titer HCMV replication. To address this issue, we treated cells with Compound C at various points post-infection and analyzed viral replication. Addition of Compound C at 24 and 48 hrs resulted in a greater than 10 and 5-fold reduction in the production of viral progeny, respectively, compared to DMSO treated (Fig. 4C). These are significant reductions in viral yield and suggest that AMPK is important throughout the duration of HCMV infection for peak viral production. However, the difference in viral growth between treatment at adsorption and treatment at 24 h is large (∼100-fold, Fig. 4C), suggesting that the major requirement for AMPK activity occurs during the first 24 h of infection. Addition of Compound C at 72 hrs post-infection had a negligible impact on HCMV viral growth (Fig. 4C). This suggests that AMPK activity is not necessary during the late stages of growth. Furthermore, this lack of inhibition when added at 72 hpi suggests that Compound C is not blocking infectious HCMV production through some artifactual interaction with newly produced HCMV virions.
With the observation that Compound C treatment decreases HCMV viral titers, we were interested in investigating how this inhibition impacted other aspects of the viral life cycle. We performed Western blot analysis and quantitative real-time PCR (qPCR) to determine the effects of Compound C on viral protein accumulation and viral DNA replication, respectively. The expression of the immediate early protein, IE1, was largely unaffected by Compound C treatment (Fig. 5A), however a large decrease in the abundance of the early protein UL44 and the late protein pp28 was observed (Fig. 5A). These results suggest that AMPK is required for the transition from the immediate early to the early stages of infection. As many early genes are involved in the regulation of viral DNA replication, we hypothesized that this defect in early gene expression could result in decreased viral DNA replication. As shown in Figure 5B, when virally infected fibroblasts are treated with Compound C, a marked decrease in viral DNA accumulation is observed, most notably at 48 and 72 h post-infection. These findings suggest that AMPK inhibition affects the viral life cycle at early stages of infection and inhibits viral DNA replication.
The finding that pharmaceutical inhibition of AMPK attenuates viral replication and HCMV-induced glycolytic flux suggests that AMPK plays an important role during viral infection. Despite reports that Compound C is a specific inhibitor of AMPK [28], the possibility for off-target effects is always an issue with pharmaceutical inhibitors. To confirm the importance of AMPK for HCMV replication, we employed an AMPK-specific RNAi to decrease AMPK expression during infection. Transfection of RNAi specific for AMPK resulted in an ∼40% reduction in AMPK abundance in both mock and HCMV-infected fibroblasts at 24 h post-infection in comparison to control RNAi transfected cells (Fig. 6A). Analysis of AMPK activity indicated that transfection of AMPK-specific RNAi prior to HCMV infection reduced AMPK activity by approximately 50%, comparable to mock levels (Fig. 6B). AMPK-specific RNAi had a much smaller impact on the AMPK activity of mock-infected cells (Fig. 6B), consistent with a relative lack of AMPK activity in mock-infected cells to start with. Analysis of media lactate accumulation indicated that AMPK-specific RNAi ablated the HCMV-mediated induction of lactate excretion, but had little impact on the lactate excretion of mock-infected cells (Fig. 6C). Analysis of how RNAi-mediated AMPK inhibition impacted viral replication indicated a greater than 250-fold reduction in viral progeny production as compared to control cells (Fig. 6D). In total, these results confirm our findings that AMPK is a critical cellular factor required both for HCMV-mediated glycolytic induction as well as for high-titer replication.
Previously, we have shown that glycolysis is upregulated upon HCMV infection, and that CaMKK appears to be required for both productive viral replication and virally-induced glycolytic flux [16]. CaMKK has been shown to be involved in the regulation of AMPK activity under various conditions [18], [39]. Given the observation that AMPK appears to be important for both productive viral replication and HCMV-induced glycolytic flux as well, it seemed likely that CaMKK could be responsible for activating AMPK during HCMV infection. In order to determine the importance of CaMKK for AMPK activation, we used the CaMKK-specific inhibitor STO-609 to treat fibroblasts and subsequently analyzed the impact on AMPK activity. As shown in Figure 7A, inhibition of CaMKK blocked the induction of AMPK activity during HCMV infection. Importantly, it has been previously reported that STO-609 treatment at a similar dosage does not impact AMPK activity directly or affect AMPK activation induced by another AMPK-activating kinase, LKB1 [19], [40].
To further explore the impact of CaMKK inhibition on AMPK activity, we examined the accumulation and phosphorylation of AMPK and its substrates by Western blot after treatment with STO-609. Pharmaceutical inhibition of CaMKK decreased the levels of phosphorylated AMPK, but also reduced the levels of total AMPK (Fig. 7B). Analysis of the relative ratio of phospho-AMPK to total AMPK indicated STO-609 treatment shifted the ratio towards the unphosphorylated AMPK by 30% at 24 hpi (Fig. 7B). The observed reductions in total as well as pAMPK upon STO-609 treatment likely contribute to the reduction in AMPK activity observed in STO-609-treated cells (Fig. 7A). Analysis of Ser79 phosphorylated-ACC upon STO-609 treatment indicated a similar trend. The amounts of Ser79-phosphorylated ACC and total ACC were both reduced upon STO-609 treatment, with a reduction of 40-50% in the relative pACC/ACC ratio. Similar to the observation with AMPK inhibition, treatment with STO-609 also blocked the increases in TSC1 and Glut4 levels observed during HCMV-infection (Fig. 7B). As we had previously analyzed the impact of STO-609 treatment on HCMV activated 13C-FBP labeling [16], we wanted to extend these observations with respect to measuring lactate production. Consistent with our previous 13C-FBP labeling results, STO-609 treatment blocked the induction of lactate secretion associated with HCMV infection yet had no effect on the accumulation of lactate production in mock-infected cells (Fig. 7C). Taken together, our results demonstrate that inhibition of CaMKK inhibits HCMV-mediated AMPK activation as well as the accumulation and phosphorylation of downstream AMPK targets which is consistent with a model in which CaMKK mediates AMPK activation during HCMV infection.
We have previously established that HCMV infection induces numerous changes to the host-cell metabolic network [13]–[14]. Induction of glycolysis has also been found to be critical for high-titer HCMV replication [16], [41]. Here we report that HCMV activates AMPK, a metabolic stress kinase, and that HCMV depends on its activity for high-titer replication. HCMV requires AMPK activation to increase glucose import and drive increased glycolytic flux (Fig. 8). Inhibition of AMPK attenuated both early and late gene expression and markedly reduced viral DNA replication. These results suggest that AMPK is an important cellular factor for HCMV replication.
Unstressed, uninfected cells do not normally utilize AMPK to activate glycolysis [17]. Our results support this view, as the inhibition of AMPK did not impact the import of glucose or the FBP labeling rate in uninfected cells (Fig. 3). In contrast to uninfected cells, HCMV infection induces the activation of AMPK, which is critical for HCMV-mediated glycolytic activation. Interestingly, activation of AMPK would be predicted to have several consequences that are detrimental to infection including inhibition of protein translation and fatty acid biosynthesis [42]. AMPK-mediated inhibition of translation occurs through induction of the TSC1/2 complex which in turn negatively regulates translation through inhibition of mTOR [33]–[34]. It has recently been shown that the HCMV UL38 protein can bind to the TSC1/2 complex and prevent its inhibitory activity on mTOR and translation initiation [31], [43]. Taken together, it appears that HCMV infection induces AMPK activation which in turn drives glycolytic activation, yet blocks the anti-viral effects of AMPK activation through the action of specific gene products such as UL38 (Fig. 8).
While the UL38 protein appears sufficient to block the inhibitory effects of AMPK activation on mTOR activity, it is less clear how HCMV infection blocks AMPK's inhibitory effects on fatty acid biosynthesis. We have previously found that HCMV induces fatty acid biosynthesis, and specifically induces the activity of acetyl-CoA carboxylase (ACC), the rate-limiting enzyme of fatty acid biosynthesis [14], [29]. ACC, and consequently fatty acid biosynthesis, is negatively regulated by activated AMPK [17]. Given that HCMV requires activated fatty acid biosynthesis and ACC activity for viral replication, it is likely that HCMV infection blocks the negative impact of activated AMPK on ACC activity, potentially through the activity of an HCMV viral protein.
Our results suggest that inhibition of CaMKK blocks the down-stream effects associated with activated AMPK (Fig. 7). Previous reports suggest that pharmaceutical inhibition of CaMKK using STO-609 blocks CaMKK-mediated activation of AMPK but does not impact LKB1- mediated AMPK activation or activation of AMPK upon energetic stress, for example, upon treatment with glycolysis inhibitors [19], [40]. Taken together, these results suggest that HCMV infection requires CaMKK activity to activate AMPK, though the exact mechanism responsible is unclear. Our results suggest that inhibition of CaMKK reduces the amount of Thr172-phosphorylated AMPK, a known CaMKK phosphorylation site, as well as the total amount of AMPK. Both of these effects would be predicted to contribute to a decrease in AMPK activity during HCMV infection. Other AMPK phosphorylation sites have been implicated in the regulation of AMPK activity [35]–[37], thus CaMKK could potentially be modulating AMPK activity through phosphorylation of sites other than Thr172 as well.
We have previously reported that inhibition of CaMKK blocks high-titer virus production [16]. Our current findings that AMPK inhibition blocks HCMV replication to a similar extent as CaMKK inhibition is consistent with a model in which HCMV-mediates AMPK activation through CaMKK (Fig. 8). How HCMV infection induces CaMKK activity still remains to be determined, although it has previously been reported that HCMV infection induces Ca2+ release from ER stores and we have found that HCMV infection induces CaMKK expression [16], [44], both of which would be predicted to increase CaMKK activity.
Glycolysis has been shown to be important for HCMV replication, and glycolytic inhibition has a similar impact on HCMV as inhibition of AMPK and CaMKK [16], [41]. The similarity is both quantitative, in terms of the magnitude of reduction in viral titers, as well as qualitative, in blocking viral DNA replication and late gene expression, attenuating early gene expression and having no detectable impact on immediate early gene expression [16], [41]. Given these similarities, and combined with the observed necessity of AMPK and CaMKK for HCMV-induced glycolysis, the simplest model would be that CaMKK and AMPK activation are important for HCMV replication due to their activation of glycolysis. Despite these correlations, it remains to be determined how much of HCMV's reliance on CaMKK and AMPK activity is due to their activation of glycolysis. The possibility that these kinases contribute to viral infection through phosphorylation of other cellular or viral targets cannot be ruled out.
In summary, we find that HCMV infection activates AMPK, which is required for HCMV-mediated glycolytic activation and high-titer HCMV replication. As AMPK activation signals metabolic stress to normal cells, HCMV has evolved mechanisms to block the anti-viral consequences of metabolic stress pathway activation. While some of these mechanisms are known, such as UL38 maintaining mTOR activation through interaction and inhibition of the TSC complex, others remain to be elucidated, such as maintenance of fatty acid biosynthesis. This stress response balancing act is representative of a recurrent theme in virus-host evolution. Replicating viruses must create a cellular environment conducive to viral replication, the efforts of which host cells have evolved to resist. In the case of AMPK, the evolutionary struggle is for the keys to the host-cell metabolic machinery. As the AMPK pathway is not normally activated in uninfected cells and inhibition of AMPK activity is tolerated in animal models [45]–[46], targeting this pathway clinically could be therapeutically beneficial for preventing HCMV-associated disease.
MRC-5 fibroblasts were cultured in Dulbecco modified Eagle medium (DMEM; Invitrogen) supplemented with 7.5% fetal bovine serum. Cells were grown to confluence in either 10 cm or 6-well tissue culture plates. Once confluent, medium was removed and serum-free DMEM was added. Cells were maintained in serum-free medium for 24 h prior to infection.
HCMV (strain Ad169) was used to infect cells at a multiplicity of infection (MOI) of 3 for all of the current experiments. Mock-infected controls were treated with an equal volume of medium containing the same serum concentrations as virus-treated cells. Virus adsorptions were carried out for 90 min at 37°C, after which viral innocula were aspirated and serum-free DMEM was added back. Production of infectious virus was measured by standard viral plaque assay.
STO-609 (EMD Biosciences), a specific inhibitor of CaMKK and Compound C (Calbiochem), a specific inhibitor of AMPK, were maintained in DMSO at concentrations of 5 mg/ml at −20°C and 10 mg/ml at 4°C, respectively.
Labeled DMEM was prepared from glucose-free media by adding 10 mM HEPES and either labeled (13C) or unlabelled (12C) glucose to a final concentration of 4.5 gL-1. For flux analysis, samples were switched to fresh, unlabelled medium 24 h and 1 h before final addition of 13C-labeled medium. Samples were labeled for 1 min and the reaction was quenched by the addition of 4 ml −80°C 80% methanol and incubation at −80°C for 10 minutes. Cells were then scraped in the methanol, centrifuged at 3000 rpm for 5 min at 4°C and the supernatant was collected. The pellet was extracted twice more in 500 µl cold methanol, adding the resulting supernatants to the previously collected supernatant. After extraction, the supernatants were dried down under nitrogen gas and resuspended in 175 µl of 50% methanol. Samples were subsequently spun down at full speed for 5 min at 4°C and the remaining supernatant was transferred to HPLC sample vials.
The accumulation of fully-13C-labeled fructose 1,6-bisphosphate was monitored using liquid chromatography-tandem mass spectrometry (LC-MS/MS) as previously described [14] and is briefly discussed below. LC-MS/MS was performed using a LC-20AD HPLC system (Shimadzu) and a Synergi Hydro-RP column (150×2 mm with a 5 µm-particle size; Phenomenex) coupled to a mass spectrometer. The LC parameters were as follows: autosampler temperature, 4°C; injection volume, 20 µl; column temperature, 40°C; flow rate, 15 µl/sec. The LC solvents were solvent A, 100% methanol; solvent B, 10 mM tributylamine and 15 mM acetic acid in 97:3 water:methanol. The gradient conditions were as follows: negative mode—t = 0, 100% B; t = 5, 100% B; t = 10, 80% B; t = 20, 80% B; t = 35, 35% B; t = 38, 5% B; t = 42, 5% B; t = 43, 100% B; t = 50, 100% B. Mass spectrometric analyses were performed on a TSQ Quantum Ultra triple-quadrupole mass spectrometer running in multiple reaction monitoring mode (MRM) (Thermo Fisher Scientific). Peak heights for fructose-1,6-bisphosphate-extracted ion chromatograms were analyzed using Excalibur software (Thermo Fisher Scientific).
Proteins from cell lysates were solubilized in 1X disruption buffer (50 mM Tris (pH 7.0), 2% SDS, 5% 2-mercaptoethanol, and 2.75% sucrose), separated by 10% SDS-PAGE and transferred to nitrocellulose in Tris-glycine transfer buffer. Blots were stained with Ponceau S to visualize protein and ensure equal sample loading. The membranes were blocked in 5% milk in TBST followed by incubation in primary antibody. After subsequent washes, blots were incubated in secondary antibody and protein bands were visualized using the ECL detection system (Pierce). Antibodies used were specific for the following viral proteins: IE1 (Shenk Laboratory, unpublished), UL44 (Virusys), and pp28 [47] and the following cellular proteins: tubulin (Epitomics), TSC1 (Millipore), Glut4 (Abcam), ACC and phosphor-Ser79-ACC (Cell Signaling Technologies), AMPK and phospho-Thr172-AMPK (Cell Signaling Technologies). Image densitometry of specific protein bands was performed with ImageJ, developed by Rasband, W.S. at the NIH (http://imagej.nih.gov/ij/), as per the ImageJ instructions.
AMPK was assayed largely as previously described [27]. Briefly, cells were washed 3X with warm Krebs-Hepes buffer (20 mM Na Hepes, pH 7.4, 118 mM NaCl, 3.5 mM KCl, 1.3 mM CaCl2, 1.2 MgSO4, 10 mM glucose, 1.2 mM KH2PO4, 0.1% BSA) and incubated in Krebs-Hepes buffer containing either DMSO, the AMPK inhibitor, Compound C (5 µM), for 1 h at 37°C, or the CaMKK inhibitor, STO-609 (10 µg/ml). Buffer was then aspirated and dishes were placed on ice with immediate addition of 0.25 ml ice-cold lysis buffer (50 mM Tris/HCl, pH 7.4, 50 mM NaF, 5 mM Na pyrophosphate, 1 mM EDTA, 1 mM EGTA, 250 mM mannitol, 1% Triton X-100, 1 mM DTT, protease inhibitors). Cells were scraped and the resulting lysates transferred to microfuge tubes and incubated on ice for 10 minutes. Lysates were then centrifuged for 5 min at 14000 xg and 4°C in preparation for use.
The AMPK assay was composed of a total reaction volume of 25 µl that was incubated for 10 min at 30°C. Each reaction consisted of 2.5 µl lysate assay buffer (62.5 mM Na Hepes, pH 7.0, 62.5 mM NaCl, 62.5 mM NaF, 6.25 mM Na pyrophosphate, 1.25 mM EDTA, 1.25 mM EGTA, 1 mM DTT, and protease inhibitor cocktail (Roche)), 2.5 µl of 100 µM [γ-32P]-ATP (1 µCi/µl) in 25 mM MgCl2, 2.5 µl of 2 mM AMP in lysate assay buffer, 5 µl of 1 mM SAMS peptide in lysate assay buffer, with either Compound C (5 µM final) or the equivalent volume of DMSO and 12.5 µl cell lysate. The reaction mixture was spotted on P81 phosphocellulose paper which was washed with 1% phosphoric acid, water, and acetone. The radioactivity of the phosphorylated SAMS peptide was quantified by scintillation counting. Non-AMPK-mediated phosphorylation of the SAMS peptide was estimated by performing the AMPK activity assay in the presence of saturating amounts of the AMPK inhibitor, Compound C.
MRC-5 fibroblasts were transfected with 150 pmol of either esiRNA (pooled endoribonuclease-prepared siRNA) specific to AMPK1 (Sigma-Aldrich) or a non-targeting siRNA (Dharmacon) using Oligofectamine per manufacturer's directions. Forty-eight hours after transfection, siRNA-transfected cells were serum-starved for 24 h and then either mock-infected or infected with HCMV (MOI = 3). Samples were harvested at 24 h and 72 h post-infection to monitor AMPK protein knockdown by Western blot. Additional samples were harvested 96 h post-infection to monitor viral titers by standard plaque assay.
Viral and cellular DNA was harvested at various time points post-infection in lysis buffer (100 mM NaCl, 100 mM Tris-HCl, 25 mM EDTA, 0.5% SDS, 0.1 mg/ml proteinase K and 40 µg/ml RNase A), and viral DNA was quantified using the UL26 primer set (below). Quantitative PCR (qPCR) was performed using Fast SYBR green master mix, a model 7500 Fast real-time PCR system and Fast 7500 software (Applied Biosystems). For quantifying viral DNA aUL26-HCMV specific primer set was employed: 5_-AACATCGCGTCGGTGATTTCTTGC-3_ (forward) and 5_-ACAGCTACTTTGAAGACGTGGAGC-3_ (reverse), GAPDH 5_-CATGTTCGTCATGGGTGTGAACCA-3_ (forward) and 5_-ATGGCATGGACTGTGGTCATGAGT-3_ (reverse).
Lactate was measured in media samples using the BioProfile 100 Plus/400 (Nova Biomedical), which employs an enzyme dependent amperometric electrode. MRC-5 fibroblasts were cultured in serum free DMEM for 24 h before infection and either mock or HCMV-infected. Lactate excretion into the media was measured over an 18 h interval, starting with a media change. After 18 h, 600 µl of media was removed from each sample dish and analyzed according to the manufacturer's instructions (Nova Biomedical).
Further information regarding the genes/ proteins studied in this manuscript can be found at the NCBI Gene Database (http://www.ncbi.nlm.nih.gov/gene). Specific database entries for cellular genes are as follows: AMPK = PRKAB1, PRKAA1, PRKAG1; CaMKK = CAMKK1, CAMKK2; TSC1 = TSC1; TSC2 = TSC2; Glut4 = SLC2A4; ACC1 = ACACA. The viral genes mentioned include the following from HCMV (also known as Human Herpesvirus 5): UL38 = UL38; IE1 = UL123; UL44 = UL44; pp28 = UL99.
|
10.1371/journal.pcbi.1003201 | Model-Based Analysis of HER Activation in Cells Co-Expressing EGFR, HER2 and HER3 | The HER/ErbB family of receptor tyrosine kinases drives critical responses in normal physiology and cancer, and the expression levels of the various HER receptors are critical determinants of clinical outcomes. HER activation is driven by the formation of various dimer complexes between members of this receptor family. The HER dimer types can have differential effects on downstream signaling and phenotypic outcomes. We constructed an integrated mathematical model of HER activation, and trafficking to quantitatively link receptor expression levels to dimerization and activation. We parameterized the model with a comprehensive set of HER phosphorylation and abundance data collected in a panel of human mammary epithelial cells expressing varying levels of EGFR/HER1, HER2 and HER3. Although parameter estimation yielded multiple solutions, predictions for dimer phosphorylation were in agreement with each other. We validated the model using experiments where pertuzumab was used to block HER2 dimerization. We used the model to predict HER dimerization and activation patterns in a panel of human mammary epithelial cells lines with known HER expression levels in response to stimulations with ligands EGF and HRG. Simulations over the range of expression levels seen in various cell lines indicate that: i) EGFR phosphorylation is driven by HER1-HER1 and HER1-HER2 dimers, and not HER1-HER3 dimers, ii) HER1-HER2 and HER2-HER3 dimers both contribute significantly to HER2 activation with the EGFR expression level determining the relative importance of these species, and iii) the HER2-HER3 dimer is largely responsible for HER3 activation. The model can be used to predict phosphorylated dimer levels for any given HER expression profile. This information in turn can be used to quantify the potencies of the various HER dimers, and can potentially inform personalized therapeutic approaches.
| A family of cell surface molecules called the HER receptor family plays important roles in normal physiology and cancer. This family has four members, HER1-4. These receptors convert signals received from the extracellular environment into cell decisions such as growth and survival – a process termed signal transduction. In particular, HER2 and HER3 are over-expressed in a number of tumors, and their expression levels are associated with abnormal growth and poor clinical prognosis. A key step in HER-mediated signal transduction is the formation of dimer complexes between members of this family. Different dimer types have different potencies for activating normal and aberrant responses. Prediction of the dimerization pattern for a given HER expression level may pave the way for personalized therapeutic approaches targeting specific dimers. Towards this end, we constructed a mathematical model for HER dimerization and activation. We determined unknown model parameters by analyzing HER activation data collected in a panel of human mammary epithelial cells that express different levels of the HER molecules. The model enables us to quantitatively link HER expression levels to receptor dimerization and activation. Further, the model can be used to support additional quantitative investigations into the basic biology of HER-mediated signal transduction.
| The HER family (Human Epidermal growth factor Receptor, also known as the ErbB family) of cell surface receptors plays critical roles in normal cell physiology, development, and cancer pathophysiology [1], [2], [3], [4]. The family consists of the four closely related transmembrane receptor tyrosine kinases HER1 (EGFR), HER2 (NEU), HER3 and HER4, which when activated initiate downstream signaling, and affect a range of cellular decisions including proliferation, survival and motility [4], [5].
The HER receptor expression profile is a critical determinant of cell behavior [6], [7], and outcomes in cancer pathology. Overexpression of EGFR, HER2 and HER3 is associated with decreased survival in cancer, while HER4 overexpression is correlated with increased survival [8], [9]. HER2 is overexpressed in 25–30% of all breast cancers, as well as in other solid tumors [10], [11] and is associated with poor prognosis [8], [12], [13], [14]. While this has led to the development of a range of therapeutics targeting the HER2 receptor [15], the use of these drugs can often lead to resistance through a diverse set of mechanisms [16]. The overexpression of HER family members and their ligands are key compensatory mechanisms responsible for the development of resistance to HER-targeted therapies [17], [18], [19], [20]. In particular, the importance of HER3 expression in driving tumorigenesis [21], [22], [23], [24], and in the development of drug resistance [17], [25] is being increasingly recognized leading to an increased focus on HER3-targeted therapies [3], [15], [26], [27], [28]. While the importance of HER expression levels has been established for clinical prognosis and drug resistance, the mechanistic link between receptor expression, HER activation and downstream consequences is not as clear yet.
HER activation is a complex process involving multiple sequential steps, which in general are as follows: the specific binding of ligands (growth factors) to HER receptors leads to conformational changes promoting dimerization between members of the family [29], [30], [31]; dimerization leads to the trans-phosphorlyation of receptor cytoplasmic tails via the kinase activities of the partners in the dimer leading to downstream signaling [31]. Although the HER receptors are homologous, there are key differences in their behavior. EGFR [32], HER3 [33], and HER4 [34] undergo ligand-induced conformational changes promoting dimerization. In contrast, HER2, which has no known ligand, has a structure that enables constitutive dimerization [35], [36]. HER3, on the other hand has impaired kinase activity, but can allosterically facilitate a partner's kinase activity following dimerization [37]. Further, HER receptors have different trafficking properties with EGFR showing increased ligand-induced internalization and degradation compared to the other members of the family [38]. All of these aspects have a bearing on the number and types of dimers that are formed between the HER receptors following ligand addition. Since the HER dimerization pattern is an important determinant of the consequences of HER activation [39], [40] it is important to quantitatively predict this as a function of the receptor expression profile.
Mathematical models have been extensively applied to understand HER activation dynamics [1], [41], [42]. Recent efforts have focused on a quantitative understanding of the interactions between multiple members of the HER family [43], [44], [45], [46]. Birtwistle et al. constructed a mathematical model for the early events (0 to 30 min) in HER activation and downstream signaling in cells coexpressing all four HER receptors [45]. They parameterized their model using HER1, HER2, Erk and Akt activation data in response to EGF and HRG stimulation in MCF-7 cells. Chen et al. constructed a more expanded model for receptor activation and signaling over longer time frames (0 to 120 min) and parameterized it using HER1, Erk and Akt activation data in response to EGF and HRG stimulation in three different cell lines – A431, H1666 and H3255 cells [43]. In each of these two manuscripts, the authors note problems with regards to parameter identifiability given the size of the models [43], [45]. Hendriks et al. focused on HER activation alone in cells expressing HER1, HER2 and HER3 [44]. They assumed parameter values based on the literature and compared simulations with receptor activation data collected in the H292 lung carcinoma cell line [44].
We have recently developed a panel of Human Mammary Epithelial (HME) cells that co-express EGFR with varying levels of HER2 and HER3 [47]. HME cells, like many epithelium-derived cell types require EGFR activation for proliferation and migration [48], and are an excellent system for developing physiologically relevant models of HER signaling. Importantly, our cell line library enables us to study the effects of varying HER expression levels in a common cellular background. We have published data on HER1-3, Erk and Akt activation in these cell lines for single doses of EGF and HRG [47]. Here, we focus further on the quantitative aspects of HER activation. We collect an expanded dataset for receptor activation that includes measurements of total and internal HER phosphorylation and HER receptor mass in four distinct HME cell lines in response to a range of EGF and HRG doses. We have identified the appropriate modeling approach (choices for model scope, granularity, etc.) for analyzing such datasets through a comprehensive model-based analysis of EGFR activation in cells that predominantly express this receptor alone [49]. Here, we expand this model by considering the co-expression of EGFR, HER2 and HER3, and parameterize it using receptor activation data collected in our cell line library. We explicitly consider model identifiability and show that the model can predict the dimer phosphorylation levels given the receptor expression level of a cell line. We note that the fourth member of the HER family is also very important in cancer [50] and should be included for completeness. However, our gene expression and proteomics studies (unpublished data) indicated that used HME cells do not express HER4. Therefore, it was not considered in this study.
Our objectives here are to quantitatively link HER (specifically HER1-3) expression levels to receptor activation, and to understand how differential interactions between the members of the HER family drive the process. Towards this end we constructed a parsimonious mathematical model for HER dimerization and receptor activation (Figure 1, see Methods for details), and parameterized it using the appropriate experimental datasets. The model includes the ligands EGF and HRG, ligand-bound and unbound receptor monomers, as well as the feasible combinations of receptor homo- and hetero-dimers (Figure 1A). The reversible biochemical reactions of receptor-ligand binding and dimer formation are represented explicitly via mass action kinetics (Figure 1B). As in our recent manuscript [49], we expressed the level of phosphorylated HER1-3 as a linear combination of the contributions from various dimer species with dimer-specific phosphorylation factors (pfs in Figure 1B) accounting for the relative contribution of each species. The pf can be thought of as a lumped phosphorylation efficiency factor that combines the characteristics of all possible tyrosine sites in a dimer [49]. It enables us to calculate the HER1-3 phosphorylation signals emanating from the various dimer types. The model includes 3 compartments: the cell surface, early endosomes and late endosomes. Biochemical reactions are allowed to occur in the first 2 compartments, while the late endosome is assumed to be a site for the accumulation of dephosphorylated receptors prior to degradation (Methods, and also [49]).
In all, the model consists of 51 species and 140 parameters. Given these parameter values, and a specified HER1-3 expression level, the model can be used to predict activated levels of HER1-3 at the cell surface and interior as a function of time in response to various concentrations of EGF and HRG. Values for several model parameters including receptor-ligand binding rates, receptor internalization, recycling and degradation rates are available in the literature (Methods; Tables S1, S2, S3 in Text S1). With these in place, there are 47 unknown model parameters (highlighted in red in Figure 1) that include: the compartment-specific dissociation rates for various receptor dimers, the compartment-specific phosphorylation factors that define the contribution of various dimers to the HER1-3 phosphorylation levels, and a parameter that defines how species are sorted (distributed) between the early and late endosomes (Methods).
In order to determine these unknown model parameters, we measured HER1-3 activation dynamics in a panel of HME cell lines with relatively constant levels of HER1, and different levels of HER2 and HER3 [47]. The complete set of data collected for one of the four cell lines used in our study that expresses all 3 HER receptors at significant levels (HER2+3+; designated with clone tag D20) is presented in Figures 2–4. The data includes measurements of the total levels of phosphorylated HER1-3 in response to various doses of EGF (Figure 2A–C, markers), and various doses of HRG (Figure 2D–F). We also obtained detailed time course measurements of total levels of phosphorylated HER1-3 (Figure 3A–C), levels of phosphorylated HER1-3 in the cell interior (Figure 3D–F), and HER1-3 total protein levels (Figure 4) in response to a single specific dose each of EGF and HRG, either added separately or in combination. Corresponding datasets for the parental (HER2−3−) cell line that expresses very low levels of HER2 and HER3; a cell line that expresses HER2 but not HER3 (HER2+3−, designated 24H); and a cell line that expresses HER3 but not HER2 (HER2−3+, designated B5) are presented in Figures S3, S4, S5, S6, S7, S8 of Text S1. Note that consistent y-axis scales are used in Figures 2–4 and Figures S3, S4, S5, S6, S7, S8 of Text S1 for each measurement type to enable comparison of total receptor phosphorylation, internal phosphorylation levels and receptor mass across the four cell lines. In all, the experimental data consisted of 999 distinct measurements, with N> = 2 for each measurement.
We estimated the 47 unknown model parameters by simultaneously fitting the model to all of the data described above. Distinct measurement types (receptor phosphorylation, receptor mass) were scaled appropriately to ensure that they contributed comparable amounts to the residual vector (Methods). We found that ∼188 of the optimization runs converged with an RMSE relatively close to the best overall RMSE (Figure S1 in Text S1). In order to better assess the location of the solutions in the 47-dimensional parameter space, we used k-means clustering to identify the existence of distinct solution clusters (Figure S2 in Text S1). Our analysis indicated that the solutions can be split into 7 distinct clusters (Figure S2A in Text S1). In order to examine the similarities and differences in the model's behavior for these solutions, for each cluster we selected the parameter set that yielded the best fit (smallest RMSE), and used these 7 solutions (Figure S2B–C in Text S1) for additional analysis.
In Figures 2–4 and Figures S3, S4, S5, S6, S7, S8 of Text S1 we compare the experimental data (markers) to model predictions (lines) generated using each of the 7 representative parameter sets. Predictions using the parameter set with the best overall RMSE are depicted using darker lines in these plots. As seen, the model predictions are in good agreement with the experimental data (compare lines vs. markers of same color). Further, predictions using the 7 distinct parameter sets are in good agreement with each other (compare lines of the same color). This can also be seen in Figure S9 of Text S1 where the mean and standard deviations of the 7 distinct model predictions are plotted against the experimental data for the various types of measurements. There is a strong linear relationship between model predictions and the experimental data with a slope between 0.83 and 1.1 for most measurement types (Figure S9 in Text S1). The exceptions are for the levels of phosphorylated HER2 in the cell interior (slope = 0.64), and the EGFR receptor mass (slope = 0.77). Overall, model predictions are in good agreement with the experimental data with each of the 7 solutions yielding comparable results.
Parameter values for the 7 representative solutions are presented in Table S4 of Text S1. Although these solutions result in comparable fits to the various measurements (see Figures 2–4) and have similar overall RMSEs (Table S4 in Text S1), they involve substantial differences in the values of several parameters, with 20 parameters displaying a greater than 2-orders of magnitude variability. Reliability of the estimated parameters can depend on the information content of the training data, which in turn depends on the experimental design because the choice of which conditions are changed in the experiments can favor the identifiability of certain parameters. In other words, certain parameters would be more sensitive to the changed experimental conditions and this would allow for their better determination. For this reason, while some of the parameters can be extracted from the datasets reliably, we can only determine broad ranges for the rest of the parameters. The results indicate that both the dimerization affinities and the pf values are estimated with reasonable confidence for the R11 homodimer and the R12 heterodimer (in our notation Rij refers to the HERi-HERj dimer). However, there is more than two orders of magnitude variability in the dimerization and phosphorylation parameters related to the R13, R23, R22, R33 dimers (Table S4 in Text S1). Model predictions for the abundances of various receptor dimers generated using the 7 representative solutions reinforce these findings: predictions for R11 and R12 abundances fall within a narrow range, while there is considerable uncertainty in the estimates for the other dimers (Figure S10 in Text S1).
Visualization of the correlation between the parameters calculated based on the 188 solutions with good RMSE values (Figure S11 in Text S1) revealed that the dimer dissociation rates and phosphorylation efficiencies (pf values) were strongly correlated with each other for the various dimer types (Figure S11 in Text S1). This is because the extent of receptor phosphorylation for each of the three HER receptors is determined by both the absolute number of dimers of each type, as well as the pf values for the dimers (e.g., see equation in Figure 1B). Thus, estimating both the dimerization affinities and pf values by fitting the model to receptor phosphorylation data is expected to be challenging. The ability to overcome this limitation in the case of the R11 and R12 dimers is likely to be related to the availability of dose response datasets with strong HER1 and HER2 receptor phosphorylation in cell lines that express EGFR/HER1 alone, or HER1 and HER2, but not HER3.
In order to independently validate the model, we collected additional data for receptor phosphorylation in various cell lines in the absence and presence of 2C4 (Pertuzumab) (Figure 5). This monoclonal antibody is considered to be a general inhibitor of HER2 dimerization due to its ability to bind the HER dimerization surface [51], [52], [53]. Model simulations for the antibody blocking experiments were performed by assuming that the addition of 2C4 renders 95% of the cellular HER2 unavailable for receptor dimerization. Note that the concentration of 10 µg/ml 2C4 used in our experiments is much higher than the Kd value for 2C4-HER2 binding [54]. Although none of the data in Figure 5 was used in “training” the mathematical model, the model does an excellent job of predicting these results. When model predictions for EGFR, HER2 and HER3 phosphorylation are plotted against experimental data from the validation experiments we obtain linear relationships with slopes of 0.92, 0.96 and 0.97, respectively (Figure S12 in Text S1). As before, model predictions of receptor phosphorylation based on the 7 distinct parameter sets are in excellent agreement with each other (see standard deviations of model predictions in Figure 5 and Figure S12 in Text S1).
HER receptors display unique patterns of site-specific phosphorylation, adaptor protein recruitment, and downstream signaling depending upon their dimerization partners [39], [40]. Therefore, it is of interest to quantify the relative contributions of various dimer types to HER phosphorylation. We calculated the phosphorylated levels of HER1-3 emanating from various dimers by multiplying the dimer abundances with appropriate phosphorylation efficiencies (see Methods). We found that although dimer abundances could be uniquely determined for only a subset of dimers (Figure S10 in Text S1), the contributions of various dimers to HER phosphorylation could be determined with much higher confidence (Figure S13 in Text S1). Predictions for the time-dependent phosphorylation signal from the various dimer types in the HER2+3+ cells using the 7 distinct parameter sets were in good agreement with each other (Figure S13 in Text S1). Since, the phosphorylation levels were found to be relatively stable beyond 1 hour of ligand addition (Figure S13 in Text S1), we chose the t = 60 min time point for all subsequent analysis. Model predictions for dimer contributions to HER1, HER2, and HER3 phosphorylation in the HER2+3+ cells at 60 min following the addition of saturating levels of EGF and HRG are presented in Figures S14, S15, and S16, respectively in Text S1. Predictions from the 7 different solutions, in general, were in good agreement with each other. The exception was for HER2 phosphorylation where the R12 dimer was found to contribute between 48–69% of the HER2 phosphorylation signal, with a 4–5% contribution from the R22 homodimer and the rest from the R23 dimer (Figure S15 in Text S1). Since, the predictions overall were in reasonable agreement we picked the best fit parameter set – the one with the lowest RMSE (Table S4 in Text S1) – and used it to generate predictions for the other cell lines in our panel.
To understand the effect of HER expression levels on the phosphorylation pattern, we calculated the relative contributions of the dimers to HER activation in each of the 4 cell lines used in our study (Figure 6). In the figure, relative dimer contributions are shown as pie charts where the size of the circles is proportional to the total phosphorylation level (Figure 6). HER1 phosphorylation was found to be consistently high in all four cell lines with the highest level in the HER2+3− cells (Figure 6A–D). As expected, ∼90% of HER1 phosphorylation in the parental (HER2−3−) cells was found to be due to the R11 homodimer, with most of this contribution coming from the species where both dimer partners were ligand-bound (R11EE; Figure 6A). In the HER2+3− (24H clone) cells, >60% of EGFR phosphorylation was from the R12 dimer (Figure 6C). Dimer contributions to EGFR phosphorylation in the HER2−3+ (B5 clone) and HER2+3+ (D20 clone) cells were similar to that in the parental and 24H cells, respectively. In other words, HER1 and HER2 expression levels were found to dictate the HER1 phosphorylation pattern, with HER3 expression having little to no effect.
As expected, HER2 phosphorylation levels were predicted to be much higher in the HER2+3− and HER2+3+ cell lines compared to the HER2− cell lines (Figure 6E–H). The HER2+3+ cells were found to have the highest HER2 phosphorylation. Whereas in the HER2+3− cells 90% of HER2 phosphorylation was due to the R12 dimer (Figure 6F), there were substantial contributions from both the R12 and R23 dimers in the HER2+3+ cells (Figure 6H). Since the EGFR expression level in the HER2+3+ cells is an order of magnitude higher than the HER3 level (Table S5 in Text S1), this suggests a stronger propensity to form activated R23 dimers compared to R12 dimers. In all cases, we found that the R22 homodimer contributed less than 5% to HER2 activation. Overall, HER2 phosphorylation can be driven via interactions with either EGFR or HER3 with the latter appearing to be the preferred dimer partner.
High levels of HER3 phosphorylation were found only in the HER2+3+ cell line (Figure 6I–L). The R13 dimers (EGF or HRG-bound) were found to contribute significantly to HER3 activation when HER2 levels were low (Figure 6I, 6K). In cell lines that expressed both EGFR and HER2, HER3 activation was found to be dominated by the R23 interaction, which points to the significance of this interaction for HER3 activation.
Our model can be used to predict the extent of HER phosphorylation and the receptor dimerization pattern for any combination of HER1-3 expression levels. We generated model predictions over a wide range of receptor expression levels (Figure 7). To ensure the relevance of this analysis, we obtained information on the HER expression levels of various cell lines from the literature (Table S5 in Text S1). Since, most HER3-expressing cells typically display a receptor expression level of ∼40,000 molecules/cell (Figure S17 in Text S1), we fixed HER3 at this level. We varied EGFR from 103 to 106 and HER2 from 103 to 3×106 to encompass the receptor expression levels observed in various cell lines (Table S5 and Figure S18 in Text S1). HER1-3 phosphorylation levels (Figure 7A–C) and the percentage contribution from the R11 homodimer to EGFR phosphorylation (Fig 7E), the R12 dimer to HER2 phosphorylation (Figure 7F) and the R23 dimer to HER3 phosphorylation (Figure 7G) are presented in Figure 7 as a function of EGFR and HER2 expression levels. The contribution of the other dimer types to HER1-3 phosphorylation is presented in Figure S19 in Text S1. Simulation results indicate that EGFR phosphorylation increases with both EGFR and HER2 expression, with HER2 expression having a stronger effect at low to moderate EGFR expression (Figure 7A). The EGFR homodimer contributes anywhere from 0–100% of the EGFR phosphorylation signal with the actual contribution increasing with EGFR expression and decreasing with HER2 expression (Figure 7B). The contribution of the R12 dimer to HER1 phosphorylation displays the opposite pattern (Figure S19A in Text S1), while the R13 dimer contributes <10% to HER1 phosphorylation in all cases (Figure S19B in Text S1).
HER2 phosphorylation increases with HER2 expression, with EGFR expression having only a minor effect (Figure 7B). The contribution of the R12 dimer to HER2 phosphorylation shows a broad range, increasing with EGFR expression (Figure 7G). The contribution of the R23 dimer shows the opposite pattern (Figure S19C in Text S1). The HER2 homodimer is predicted to contribute <15% of the HER2 phosphorylation signal in all cases (Figure S19D in Text S1). Interestingly neither the R12 contribution nor the R23 contribution is a strong function of HER2 expression (Figure 7G, Figure S19C in Text S1). Thus, the EGFR expression level is the strongest predictor of which dimer type dominates HER2 signaling.
HER3 phosphorylation is found to be strongly dependent on HER2, but not EGFR expression levels (Figure 7C). Over the range of expression levels seen in actual cells (see dots in Figure 7D), >80% of the HER3 signal is predicted to be due to the R23 dimer (Figure 7D) with the remaining from the R13 dimer (Figure S19E in Text S1).
In order to validate these simulation results, we compared model predictions for HER3 activation in two distinct cell lines – ADRr and ADRrE2 – with similar levels of EGFR and HER3, but distinct HER2 levels with previously published experimental data [27]. As seen, model predictions for the relative change in HER3 activation due to an increase in HER2 expression are in good agreement with the experimental data (Figure 7). Thus, our results indicated that the model predictions may be applicable to other cell lines as well. To enable cell type-specific comparisons, we have computed the phosphorylation levels of HER receptors and HER dimerization patterns for 52 distinct cell lines using their receptor expression levels compiled from the literature (Figures S20, S21, S22 in Text S1).
One key aspect of receptor signaling is the prediction of how the changes in receptor phosphorylation levels would alter the activation patterns of the downstream elements of the involved signaling pathways. Here, we briefly illustrate how the constructed receptor activation model can be used to quantitatively predict the relative contributions of the HER receptor types and their dimers to the activations of Erk and Akt kinases. Erk and Akt are important regulators of the cell proliferation and mobility processes, and their activation kinetics in HME cells were subject of our earlier investigations [47], [55].
We pursued multilinear regression analysis to determine the relationship between Erk and Akt activation and HER phosphorylation by fitting the coefficients of the regression model to the data collected in our HME cell lines. This analysis was pursued in two different ways by assuming a) that the total receptor phosphorylations are the predictors, i.e., pT(t) = b0+Σ bi * pRi(t), where pRi(t) is the contribution of receptor type i ( = HER1, 2, or 3) to the activation of the target protein T ( = Erk or Akt) and the sum is over the receptor types, and b) that the receptor dimer contributions are the predictors [55], i.e., pT(t) = b0+Σ bi_ji * pRi_ji(t), where pRi_ji(t) is the contribution of the dimer Rij to the phosphorylation of receptor type i and the sum is over the receptor dimer types. Comparison of the regression model predictions with the experimental data have shown that the EGFR/HER1 contribution to Erk phosphorylation (pERK) is the dominant predictor and that various receptor dimers could make comparable contributions to the prediction of pERK (Table S1; Figure S24 in Text S1). In contrast to Erk, Akt phosphorylation has a much stronger dependence to the activation through particular receptor dimers: regression analysis indicated that signaling through the HER1-HER3 receptor dimer was the dominant predictor of pAKT (Table S1; Figure S25 in Text S1). Results of this analysis were consistent with the results of other analysis methods such as clustering and targeted inhibition (Gong et al, in preparation).
We have constructed a parsimonious mathematical model for HER1-3 activation that incorporates the important biochemical/biophysical steps involved in the process, and have parameterized it using the data collected in a panel of HME cells that express varying levels of HER1-3. Despite using rate constants from the literature where available, and considering an extensive dataset including total and internal receptor phosphorylation levels and receptor mass measurements as a function of ligand dose, our analysis indicates that not all aspects of the model are equally identifiable. Specifically, we find that while there is considerable uncertainty surrounding the absolute dimer abundances of all but the HER1 homo- and HER1-HER2 hetero-dimer types (Figure S10 in Text S1), the phosphorylation signal from all the dimer types can be predicted with good confidence using the model (Figure S13 in Text S1). Since the tyrosine phosphorylation levels in various dimers are the relevant quantities to consider in the context of signal transduction, the obtained results provide the needed information for probing dimer specific downstream responses. That said the lack of complete model identifiability still highlights the challenges encountered in the construction and parameterization of models for biomolecular networks. Additionally, as in almost all of the earlier studies, possible location-dependence of the HER receptor kinetics was not included in our study. Receptor placement in membrane ruffles or the corralling role of the cytoskeleton elements [56] could be important factors but such complexities cannot be captured with the design of our experiments and hence were omitted.
Previous modeling studies of the co-expression of multiple HER receptors considered both receptor activation and downstream signaling (Erk and Akt activation) as part of an integrated analysis [43], [45]. They involved the use of a single cell type [45], or multiple distinct cell types [43]. These models are useful because they represent comprehensive quantitative frameworks for assembling information regarding HER-mediated signaling, and serve to document the various steps in the process. They have also been utilized in subsequent studies that have focused on therapeutic targets for HER3-mediated signaling [27], [28]. However, due to the large scope of these models, model identifiability is a challenge as noted by the authors themselves [27], [43]. Here, we adopted an alternate approach to establish the quantitative link between HER expression levels and downstream signaling: we constructed a relatively detailed mechanistic model for HER activation, since the available information and datasets allowed us to do so. As a next step, we have used the dimer phosphorylation levels predicted by the model (an aspect that is identifiable given the data), along with data on the activation of MAPK and Akt signaling pathways to quantify differential signaling by the various HER dimers (Supplementary Material). We and others have previously used this conceptual step-wise approach to analyze HER-mediated signaling in cells that co-express EGFR and HER2 [55], [57], [58], [59]. Interestingly, as briefly discussed in the Results section, our recent analysis also indicated that activation of the pro-survival Akt pathway correlated more with HER3 signaling from the smaller R13 dimer pool compared to the substantially larger signal from the R23 dimers (to be submitted).
The current model can predict the abundance of R11 and R12 dimers with much higher confidence than that of the other dimer types (Figure S10 in Text S1). The predictions for dimer abundance, and the associated parameters for these two dimer types, can be compared with previous results. We recently used a simpler model for the activation of a single receptor type (EGFR) to analyze the data for the parental HME cell line, and showed that even under saturating concentrations of EGF, <40% of the receptors dimerize and are phosphorylated [49]. These findings were in agreement with previous experimental data from our laboratory where we quantified the fraction of phosphorylated EGFR [59]. While our previous model [49] neglected the presence of low levels of HER2 and HER3 in the parental cell line, our current analysis explicitly accounts for this aspect (Table S5 in Text S1). Further, our current analysis involves the simultaneous optimization of the model using data from four different HME cell lines. Despite these differences, the current model also predicts that for the parental cell line <40% of the EGFR form homodimers following ligand stimulation, with much lower abundances for the other dimer types (Figure S23 in Text S1). The value of the phosphorylation efficiency factor pf for the R11 dimer with two bound EGF molecules estimated here (see the pf11ees values in Table S4 of Text S1) ranges from 2.8×10−2 to 3.4×10−2, which is in excellent agreement with the mean value of 2.97×10−2 estimated in our previous manuscript [49]. This suggests that the four HME cell lines behave in a consistent manner, since the simultaneous analysis of these cell lines yields findings that are consistent with the analysis of the parental cell line in isolation.
We have previously used a much simpler model for HER activation that neglected explicit consideration of receptor-ligand binding, receptor recycling, and sorting to analyze receptor activation in cells co-expressing EGFR and HER2 alone [59]. In that analysis we assumed that the formation of active dimers occurred in a single lumped step, and for each dimer type we used a single lumped pf value applicable to all cellular compartments [59]. The analysis indicated decreased stability of the R12 dimer compared to the R11 dimer, which contradicted the assumptions used in other modeling papers [57], [60]. Further, we found that the pf11 and pf12 values were comparable indicating that EGFR phosphorylation occurred with equal efficiency in R11 and R12 dimers, and that pf22 was an order of magnitude smaller than pf21 indicating much lower HER2 phosphorylation efficiency in the R22 homodimer compared to the R12 dimer [59]. Comparison of the dimer dissociation constants obtained in our current analysis (see ku11ees and ku12es in Table S4 of Text S1) also indicates that the R11 dimer is far more stable than the R12 dimer. However, the model predicts that EGFR phosphorylation is 6–35 times more efficient at the cell surface and 1.5–4 times more efficient in the early endosome vesicles when the EGFR is part of the R12 dimer as opposed to the R11 dimer (see pf11ees/pf12es and pf11eei/pf12ei ratios in Table S4 of Text S1). The model, in agreement with our previous findings [59], also predicts that the HER2 phosphorylation is far more efficient in the R12 dimer compared to that in the R22 homodimer (see pf21s/pf22s ratio in Table S4 of Text S1). One way of validating these results would be to measure the absolute abundances of R11 and R12 dimers in these cell lines. While it is possible to address these using FRET or co-IP experiments, these experiments would be challenging due to the difficulties in quantitative interpretation of the FRET signal, and possible differences in antibody pull down efficiencies, respectively.
We used the model trained on data from HME cells to predict the HER activation levels and dimer contributions for a range of cell lines (Figure 7; Figures S20, S21, S22 in Text S1). For these predictions we first had to obtain information on receptor expression levels in various epithelial cell lines (see Table S5 in Text S1). Interestingly, while the data revealed wide variability in the expression levels of EGFR and HER2 among the cell lines, cells that expressed HER3 did so in a relatively narrow range of ∼30,000 to 60,000 receptors/cell (Table S5 in Text S1). Perhaps, this indicates that the in vivo quantitative regulation of HER3 signaling occurs via control of the expression level of its main partner HER2 and/or via the differential regulation of its ligands [61].
We partially validated the ability of our model to predict HER phosphorylation dynamics in other cell lines by comparing our results with experimental data [27] for HER3 phosphorylation in the ADRr and ADRrE2 cell lines (Figure 8). In such extrapolations we make the implicit assumption that the rate constants for HER activation processes (receptor-ligand binding, dimerization, phosphorylation, trafficking) are similar across various cell lines, and that knowledge of receptor expression levels alone is sufficient to predict dimerization and phosphorylation patterns. We caution that the validity of the assumption may be questionable, and that extrapolations to other cell lines should be specifically validated (for e.g., by measuring the levels of HER2 phosphorylation relative to a benchmark cell line) when quantitative predictions of dimerization patterns are desired. That said our predictions appear to be in qualitative agreement with published results. For instance, we predict that the R12 dimer is an important component of EGFR and HER2 activation with the contribution of this dimer dependent on both EGFR and HER2 expression levels. This is in agreement with the finding of Defazio-Eli et al. who used the VeraTag™ (Monogram Biosciences, South San Francisco, CA) proximity-based ligation assay to quantify EGFR and HER2 expression levels, R12 dimer abundances and phospho-R12 levels in various cells [62]. Mukherjee et al. [63] also used VeraTag assays to quantify the phosphorylation levels of various HER receptors and relative dimer abundances in panel of breast tumors with particular focus on HER3 activation. They found that the level of HER3 phosphorylation correlated strongly with the level of the R23 dimer, and that the expression level of HER2 is a strong determinant of the level of HER3-PI3K signaling [63]. This is in agreement with our finding that the HER2 expression level is the strongest determinant of HER3phosphorylation (Figure 7C), and that it is the R23 dimer that contributes significantly to HER3 activation (Figure 7F).
To summarize, we have constructed and parameterized a mathematical model that can be used to predict the levels of HER phosphorylation, and the levels of various phosphorylated HER dimers as a function of the HER expression profile. We present predictions of HER1-3 phosphorylation levels and their dimerization patterns for 52 distinct cell lines (Figures S20, S21, S22 in Text S1). These results can be used to determine the dominant dimer type that contributes to HER signaling in each cell line, and hence to device optimal strategies to disrupt HER signaling in a cell lines with known HER expression levels. Importantly, model predictions can be used to determine the relative potencies of the various HER dimers to activate distinct downstream cell signaling pathways, and drive specific cell decisions. In this regard, this manuscript represents a critical piece in the effort to mechanistically link HER expression levels to receptor dimerization, activation, and eventually to the cell phenotype.
The parental human mammary epithelial (HME) cell line used in this study was originally provided by Martha Stampfer (Lawrence Berkeley National Laboratory, Berkeley, CA) as cell line 184A1-1. It expresses approximately 200,000 molecules of EGFR/HER1, and much lower levels of HER2 and HER3 [47], [57], and is designated here as the HER2−3− cell line. We used retroviral transduction to insert the HER2 gene and the HER3 gene into the parental cell line to obtain the 24H (HER2+3−) and B5 (HER2−3+) cell lines, respectively. The HER3 gene was then inserted into the 24H cell line to obtain the D20 cell line (HER2+3+) that expressed all 3 receptors. We have previously described the detailed protocols used for deriving these cell lines [47]. The parental cell were maintained in DFCI-1 medium supplemented with 12.5 ng/ml EGF (PeproTech, Rocky Hill, NJ) as described previously [64]. Growth mediums for the 24H cell line, the B5 cell line, and the D20 cell line were the same as the parental cell line except for the addition of antibiotics G418 (250 µg/ml; Invitrogen, Carlsbad, CA), puromycin (2 µg/ml; Sigma, St. Louis, MO), and both, respectively to ensure selection [47].
When cells grew to near confluency, DFCI-1 medium was replaced with bicarbonate-free DFHB minimal medium lacking all supplements but 0.1% bovine serum albumin. Cells were then brought to quiescence for 12–18 hours before treatment. Cells were activated through the HER receptors by the addition of known concentration of EGF and/or HRG (Peprotech, Rocky Hill, NJ) followed by incubation at 37°C for fixed amounts of time from 5 to 120 min. In the dimerization blocking experiments, cells were preincubated with 10 µg/ml of monoclonal antibody 2C4 (Pertuzumab; generous gift from Genentech, Inc, San Francisco, CA) for 4 hours prior to ligand stimulation. Following stimulation, cells were then solubilized with ice cold lysis buffer (1% NP-40, 20 mM pH 8.0 Tris buffer, 137 mM NaCl, 10% glycerol, 2 mM EDTA, supplemented with 1 mM heat activated sodium orthovanadate and 1% protease inhibitor cocktail III; Calbiochem, La Jolla, CA) for 20 min. Cell lysates were collected with a scraper. Lysates were centrifuged at 13,000 rpm for 10 min at 4°C, and the supernatants were transferred into fresh microtubes. Obtained cell lysates were either analyzed immediately or stored at a −80°C freezer until needed.
Phosphorylated receptor levels in the internal compartments were determined using an acid-stripping protocol, which selectively dephosphorylates cell surface receptors without altering the phosphorylation of internalized receptors [65]. Following cell stimulation with ligands, and acid stripping, cells were washed 3X with ice cold PBS and incubated at room temperature for one minute to allow surface receptor dephosphorylation. After another round of cold PBS washing, cells were solubilized and lysates were prepared as described in the previous paragraph.
ELISA assays to quantify the receptor mass and phosphorylation levels were performed using the R&D DuoSet IC ELISA kits (R&D Systems Inc., Minneapolis, MN). Two types of ELISA data were collected as a function of time following ligand addition for each of the four cell lines used in our study:
The ELISA results were normalized based on the total protein present in the cell lysate (measured using the Bicinchoninic Acid protein quantitation kit, Sigma, St. Louis, MO), and were expressed in units of picograms per microgram of total lysate protein. For each cell line and treatment condition at least two independent measurements were performed, with at least two biological replicates in each experiment.
The mathematical model for HER activation (Figure 1) is an extension of our recently published multi-compartment model for cells expressing EGFR alone [49]. Here, we consider the interactions between multiple members of the HER family. There are 17 types of species in the mathematical model (Figure 1A) including the ligands EGF and HRG, free and ligand-bound HER monomers, and the various possible homo- and hetero-dimers that can be formed following the addition of EGF and/or HRG. These species are allowed to exist in 3 distinct compartments – the cell surface, early endosomes (EE) and late endosomes (LE) – resulting in a total of 51 model variables. The model combines the key biochemical reactions underlying HER activation (Figure 1B) with receptor trafficking between the cellular compartments (Figure 1C) to predict receptor mass, dimerization and phosphorylation dynamics following ligand stimulation.
In the model biochemical reactions leading to receptor activation – receptor-ligand binding, dimerization and phosphorylation – are allowed to occur at the cell surface and in the EE. Following exit from the EE, receptors destined for degradation become part of multivesicular bodies (MVBs) where they undergo terminal dephosphorylation prior to degradation [66], [67]. To account for this process, we include an idealized LE compartment which is a site for the accumulation of dephosphorylated receptors prior to degradation [49]. We assume that receptors in the LE do not contribute to receptor phosphorylation measurements, but contribute to receptor mass measurements. Since it is unnecessary to track the receptor activation process in the LE, biochemical reactions for this compartment were excluded from the rate equations.
Our general approach is to construct a parsimonious model to avoid over-fitting of the data. We use lumped parameters or scaling factors where detailed kinetic information is unavailable. To ground the model in reality, and to facilitate parameter estimation, we employ previously determined values for rate constants where available. Explicit consideration of the various HER homo- and hetero-dimer types in their different ligand-bound states demands the specification of a large number of model parameters due to the combinatorial complexity. However, we choose this approach because quantitative information is available in the literature regarding the relative affinities of various HER dimer types for EGF and HRG [68], [69], [70]. Further, we can use reasonable simplifying assumptions regarding the trafficking properties of the different species to reduce the number of unknown parameters in the model (see below).
The different reaction types in the model are briefly discussed below along with their associated assumptions, known parameter values, and unknowns. The complete governing equations for the model are presented as part of the Supporting Information. Rate expressions and parameter values used for the biochemical reactions at the cell surface and the EE are in Tables S1 and S2, respectively of Text S1. The trafficking parameters are presented in Table S3 of Text S1. Estimates for unknown model parameters obtained here by fitting to the experimental data are tabulated in Table S4 of Text S1.
There are 47 unknown model parameters (described above), which include the dimer dissociation rates, the pf values for the various dimer types and the trafficking parameter δ1. Given values for the 47 parameters and the HER1-3 expression levels (Table S5 in Text S1), the model can be simulated for any given concentration of EGF and/or HRG to predict the total (pRt) and internal (pRi) receptor phosphorylation levels as well as the receptor mass (mRt) for the three HER receptor types. In order to estimate the unknown parameters, we simultaneously considered these 3 distinct measurement types for each of the 4 HME cells. We constructed scaled residual vectors (residual = model prediction−experimental data) for the pR and mR predictions by dividing each of these residuals by the maximum values measured in the phosphorylation and receptor mass measurements. This ensures roughly equal importance to the distinct measurement types during parameter estimation. We then concatenated the scaled residual vectors and used lsqnonlin – the MATLAB (Natick, MA) nonlinear least squares regression function to determine optimal parameter values.
During optimization, initial guesses for the unknown kinetic rate parameters were generated by sampling the parameters from broad uniform distributions: guesses for the dimer dissociation rates ranged from 10−3 to 103; the pf values from 10−6 to 1 and δ1 from 0.1 to 10. To ensure convergence, we adopted a progressive optimization approach. HER1-related parameters were first estimated using the parental (HER2−3−) cell line. These values were then used as initial guesses in the estimation of parameters related to HER1-HER2 interactions using the parental and HER2+3− cell lines; and parameters related to HER1-HER3 interactions using the parental and HER2−3+ cell lines. The parameter sets obtained from these simpler optimizations were used as initial guesses in the final set of optimization runs where all 47 parameters were estimated by simultaneously considering the data from all four cell lines. These optimization runs were repeated 500 times with randomly generated initial guesses for parameters related to the HER2-HER3 interaction. We used the overall root-mean-squared error (RMSE) between the experimental data and model predictions to assess the goodness of the fit.
Following parameter estimation, model predictions were generated for dimer abundances, the HER1-3 phosphorylation signal from various dimers (product of abundance and appropriate pf value), and the total HER1-3 phosphorylation levels (sum of the relevant dimer phosphorylation signals). These results were used to compute the fractional contribution of the various dimer types to the phosphorylation of the HER1-3 receptors. Predictions were generated both for HME cells as well as a panel of 48 cell lines for which HER expression levels were compiled from the literature (Table S5 in Text S1). Unless specified otherwise, all predictions represent the HER dimerization and activation pattern at t = 60 min following the addition of saturating doses of both ligands, specifically 30 ng/ml EGF and 100 ng/ml HRG.
|
10.1371/journal.ppat.1003274 | Experimental Human Pneumococcal Carriage Augments IL-17A-dependent T-cell Defence of the Lung | Pneumococcal carriage is both immunising and a pre-requisite for mucosal and systemic disease. Murine models of pneumococcal colonisation show that IL-17A-secreting CD4+ T-cells (Th-17 cells) are essential for clearance of pneumococci from the nasopharynx. Pneumococcal-responding IL-17A-secreting CD4+ T-cells have not been described in the adult human lung and it is unknown whether they can be elicited by carriage and protect the lung from pneumococcal infection. We investigated the direct effect of experimental human pneumococcal nasal carriage (EHPC) on the frequency and phenotype of cognate CD4+ T-cells in broncho-alveolar lavage and blood using multi-parameter flow cytometry. We then examined whether they could augment ex vivo alveolar macrophage killing of pneumococci using an in vitro assay. We showed that human pneumococcal carriage leads to a 17.4-fold (p = 0.007) and 8-fold (p = 0.003) increase in the frequency of cognate IL-17A+ CD4+ T-cells in BAL and blood, respectively. The phenotype with the largest proportion were TNF+/IL-17A+ co-producing CD4+ memory T-cells (p<0.01); IFNγ+ CD4+ memory T-cells were not significantly increased following carriage. Pneumococci could stimulate large amounts of IL-17A protein from BAL cells in the absence of carriage but in the presence of cognate CD4+ memory T-cells, IL-17A protein levels were increased by a further 50%. Further to this we then show that alveolar macrophages, which express IL-17A receptors A and C, showed enhanced killing of opsonised pneumococci when stimulated with rhIL-17A (p = 0.013). Killing negatively correlated with RC (r = −0.9, p = 0.017) but not RA expression. We conclude that human pneumococcal carriage can increase the proportion of lung IL-17A-secreting CD4+ memory T-cells that may enhance innate cellular immunity against pathogenic challenge. These pathways may be utilised to enhance vaccine efficacy to protect the lung against pneumonia.
| Pneumococcal carriage is an important step in the development of cellular and humoral pneumococcal immunity but paradoxically may lead to mucosal diseases such as pneumonia. The frequency of carriage and pneumonia in young healthy adults is very low despite frequent exposures suggesting the presence of appropriate mucosal defences. Lung mucosal immunity against the pneumococcus is poorly described in humans and lags behind recent advances in our understanding of protective cellular responses in mice. We have therefore developed a method to experimentally induce pneumococcal carriage in healthy adults in order to provide a mechanistic insight into the protective responses elicited at the lung surface. We were able to produce carriage in healthy adults and show that – in the absence of respiratory symptoms or local lung inflammation – pneumococcal-responding (adaptive) cellular responses are increased to a large extent. We also provide evidence of cellular cross-talk between lung sentinels and the pneumococcal-responding adaptive response that may help prevent lung infection in humans. Manipulation of this response may provide novel therapeutic approaches to prevent pneumonia. Furthermore these tools allow better interpretation of defective responses in at risk individuals such as the elderly.
| Nasopharyngeal colonisation with Streptococcus pneumoniae (the pneumococcus) peaks in prevalence at 2–3 years of age [1] and declines thereafter becoming 10% or less in adult-hood and undetectable in the elderly [2]. Perturbations in host defence and/or increased pneumococcal pathogenicity facilitate colonisation and increase the frequency of progression to mucosal diseases such as pneumonia [3]. Pneumonia is the leading cause of hospitalisation of children in the USA [4]. Elderly populations are also highly susceptible to pneumonia [5]. Pneumococcal carriage is critical in transmission and disease but paradoxically it is also essential for the development of adaptive immunity.
Pneumococcal nasopharyngeal colonisation leads to the establishment of antigen specific memory CD4+ T-cells [6], [7] and specific antibody [8], [9] at systemic and mucosal sites in mice. It is well established in mice that, in concert with specific antibody and innate immunity, pneumococcal-responding interleukin-17+ (IL-17A+) and not interferon-gamma+ (IFNγ+) CD4+ T-cells (Th-17 cells) are essential for protection against pneumococcal carriage [6], [7] but their role in the lung is less clear. Pneumococcal lung infection in mice leads to the significant recruitment of CD4+ T-cells into the lungs [3], [7], [10], [11]. T cells are associated with protection from pneumococcal pneumonia in some models [3] but not others [8], [12] possibly owing to variation in host genetic background and the murine bacterial challenge model used.
In humans, increased rates of pneumococcal carriage in children [13] and clinical cases of pneumonia in adults [14] were associated with a reduction in circulating Th-1 (IFNγ+) CD4+ T-cells. Polymorphisms in the IL-17A gene are associated with increased pneumococcal colonisation [15] and lung infection [16]. IL-17A and IFNγ can be detected in pneumococcal stimulated blood samples [17]–[19] and tonsillar mononuclear cells [20]. T cells with a Th-1 [21] and Th-17 [22] phenotype have been described in the human airway but their specificity for pneumococcus has not been shown and it is unknown whether they are directly elicited by pneumococcal carriage.
Many functions are attributed to IL-17A secreted from Th-17 cells [23]. It can enhance neutrophil recruitment and phagocytosis [18], increase antimicrobial peptide (beta defensin 2) production [24], iBALT formation [25], and enhance polymeric Immunoglobulin receptor expression on mucosal airway epithelial cells [26]. Human Th-17 cells persist for longer and are more resistant to apoptosis compared to Th-1 cells [27], making their increase an attractive goal for vaccinations relying on cellular immunity.
Nasopharyngeal pneumococcal carriage mediated alterations in the frequency and phenotype of pneumococcal-responding T-cell response(s) in the lung that could impact vaccination strategies to prevent acute lower respiratory tract infections or therapies designed to augment/modulate lung immunity. We have developed an Experimental Human Pneumococcal Carriage (EHPC) model to determine whether carriage could enhance cellular immunity to pneumococcus in the lung. We showed that carriage significantly increased lung and blood IL-17A+ CD4+ T-cell responses. Furthermore, rhIL-17A, dependent upon IL-17 receptor expression, can augment alveolar macrophage killing of pneumococci, to increase innate mucosal defences of the lung.
Written informed consent was obtained from healthy adult volunteers to participate in an approved study at the Royal Liverpool and Broadgreen University Hospitals Trust. Approval was obtained from Liverpool Central [REC 11/NW/0011] and Sefton [08/H1001/52]) NHS Research Ethics Committees.
This work built on other EHPC development studies [28]. In contrast to our previous pneumococcal challenge study design [28] we omitted pre-challenge bronchoscopy with lavage from these cohorts to increase our colonisation success rate. Pneumococcal inoculation was done as published on-line [29]. Briefly, volunteers (cohort details in Table 1) were challenged with a single intra-nasal dose of either 23F (P833 strain a gift from Prof. JN Weiser, University of Pennsylvania) or 6B (BHN418 strain a gift from Prof. PW Hermans, University of Nijmegen) grown in vegitone broth (Oxoid). The inoculation was performed while the volunteer was seated comfortably in a semi-recumbent position. The head was tilted back slightly and 100 µl of the bacterial inoculum was dispensed, using a Gilson pipette (P100), across the nasal mucosa. Serial dilutions of the inoculum were plated onto blood agar (Oxoid) both before and after inoculation to confirm the dose (Table 1).
Intra-nasal colonisation was assessed in nasal washes collected 48 hours, 7 and 14 days later. Sterile isotonic saline (5 mls) was instilled into each naris with the subject seated at 45° to the horizontal. Saline was held in the nasopharynx for 5 seconds, following which the subjects were asked to tip their head gently forward to allow the saline to run out of the nose and be collected into specimen pots. Collected, pooled nasal washes were centrifuged at 3345 g for 10 minutes and the pellet was resuspended in 100 µl of Skim milk tryptone glucose glycerol (STGG) medium. An aliquot (25 µl) was plated onto Columbia horse blood agar (Oxoid) containing gentamicin (Sigma) and incubated at 37°C, 5% CO2. After 24 hours plates were inspected for the presence of draughtsman-like pneumococcal colonies. Isolated colonies were subsequently sub-cultured to confirm pneumococcal phenotype using Optochin sensitivity, bile solubility tests and for serotype confirmation, latex agglutination kits (Statens Serum Institute) were used. A further aliquot was used to perform a serial dilution (Miles and Misra) and 3×10 µl drops per dilution were dropped onto blood agar for colony counting to determine the carriage density (Table 1). Carriage density was calculated by obtaining the average CFU per 100 µl (of STGG) and dividing this value by the volume (ml) of nasal wash recovered to obtain CFU/ml of nasal wash. Volunteers with a pneumococcal positive nasal wash that was of the same serotype as the original inoculum were defined as having established carriage. These volunteers were subsequently selected for blood and BAL collection and subject data is summarised in Table 1. We also recruited 9 age-matched healthy adults (without pneumococcal colonisation) to act as controls and obtained BAL and blood samples (Table 1 and presented in [28]) for comparison in a cross-sectional study.
PBMCs were processed by standard methods [28]. Briefly, PBMCs were seeded in 48-well tissue culture plates in RPMI 1640 media with 2 mM L-glutamine (both Sigma-Aldrich) and 10% human AB serum (complete media) lot 655272 (Invitrogen, UK), prior to stimulation.
BAL was obtained and processed as previously described [28]. BAL cells were plated out into standard 24-well tissue culture plates (Greiner, UK) to allow macrophages to adhere for 3 hours at 37°C, 5% CO2. BAL cells were also allowed to adhere to 96-well tissue culture plates (Greiner, UK) for an opsono-phagocytic assay, described below. Non-adherent cells were collected from 24-well tissue culture plates, washed and the pellet re-suspended in 1 ml of complete media in 48-well plates (Greiner, UK) and incubated at 37°C, 5% CO2.
For Intracellular Cytokine Staining (ICS), PBMCs or BAL cells (containing 1–2×105 lymphocytes per well) were stimulated ex vivo for 2 hours with influenza or one of the following pneumococcal antigen preparations: 1.0 µg/ml heat-killed 6B cells (HK-6B), 13 µg/ml (of which 4.2 µg/ml is pneumococcal protein) 6B culture supernatant (6B c/s), 13 µg/ml vegitone broth (‘vehicle’), 0.45 µg/ml of heat-inactivated influenza (Split Virion, Sanofi Pasteur, 2010/11 strains) or left untreated (‘NS’) [28]. After 2 hours, 1 µl of Brefeldin A (BD biosciences) was added and cells incubated for a further 16 hours before harvesting and staining for the presence of intracellular cytokines.
An equal number of BAL cells were also seeded in parallel and at an equal density to that described for ICS. These cells were stimulated for 20 hours with HK-6B, or left untreated as described above. Cells were harvested, pelleted and the supernatant removed and kept at −80°C for cytokine/protein measurements. There were no significant differences in the total number of cells, macrophages (mean [±SD] 8.5±5.5×105 vs 8.4±7.5×105) or lymphocytes (1.8±0.25×105 vs 2.0±0.09×105) per well between non-colonised and colonised groups, respectively.
Cells were harvested, stained and analysed as previously described [28]. We gated on viable, CD3+CD4+CD45RO+ T-cells (hereafter described as CD4+ memory T-cells) and identified individual TNF, IL-17A and/or IFNγ producing cells (or combinations thereof) following stimulation (Figure S1 in the online repository). Alveolar macrophage expression of IL-17RA and RC was determined as described elsewhere [30].
BAL cell culture supernatants (not treated with Brefeldin A) were analysed using a Th-1, Th-2, Th-17 Cytometric Bead Array (CBA) kit (Becton Dickinson, UK). IL-22 and Beta-defensin 2 (BD2) were measured, in duplicate, using an anti-human IL-22 ELISA (R and D Systems, UK) or an anti-human BD2 ELISA (Antigenix America Inc, USA), respectively. For the CBA, bead populations were acquired on a BD LSR 2 and fcs files analysed against the standard curve using FCAP version 1.0.1 (Soft Flow Inc. USA). For ELISAs optical density was measured at 450 nm using a Fluostar microplate reader (BMG Labtech, Germany) and corrected for background at 540 nm (IL-22) or corrected using empty wells (BD2). Standard curves were generated using linear regression fit (IL-22) or 4-parameter fit logistic regression (CBA and BD2) and had an r2 value greater than 0.97.
An OPKA using pneumococci and human alveolar macrophages was performed with minor modifications [31]. Briefly, D39 Pneumococci (serotype 2) were opsonised in a 1∶16 dilution of intravenous immunoglobulin (IVIG, Gamunex, Talecris, USA) in Hanks and incubated at 37°C for 15 mins on a rotating platform. Opsonised D39 (20 µl), complement (10 µl) and either 20 µl of rhIL-17A (rhIL-17A, Biolegend 570502, reconstituted as described below) or vehicle control (HBSS [with Ca2+ Mg2+] containing 10% AB serum [lot 655272]) were added to 1×105 adhered alveolar macrophages (multiplicity of infection of 1 pneumococcus :100 cells) in 30 µl of RPMI+10% FCS to give a total reaction volume of 80 µl in a 96-well flat bottom plate (Greiner, UK). Following 2 hours incubation at 37°C, 10 µl of reaction mixture was tilt plated, in triplicate, onto blood agar (Oxoid) and incubated at 37°C, 5%CO2 overnight. Colony forming units (CFUs) from cell supernatants were counted the following day.
Data with a normal distribution (tested by Shapiro Wilks) were compared with parametric tests. Data not following a normal distribution were compared with non-parametric tests. OPKA counts were assumed to follow a Poisson distribution. Changes in CFUs over the three rhIL-17A doses were examined using Poisson regression, with the corresponding vehicle counts included as covariates and with adjustment for clustering within participants. Flow cytometric data were analysed using FlowJo software version 7.6 (Treestar Oregon, USA). Graph and statistical analysis was performed using GraphPad prism version 5.0 (California, USA). Differences were considered significant if p≤0.05.
We recruited and inoculated 54 healthy young adult volunteers in a dose ranging study, with serotype 6B pneumococcus in which 20 volunteers established carriage (37%). In a 23F dose ranging cohort, 19 healthy adult volunteers were recruited and 2 established carriage (carriage positive 11%). In the 22 volunteers in whom we established carriage, 17 reported no symptoms, 4 reported mild upper respiratory flu-like symptoms and 1 reported abdominal pain and shortness of breath that resolved without therapy.
From the cohort of 22 volunteers with experimentally induced carriage, we were able to recruit 12 volunteers (average age 22.5 years) 36 days later (range 21–56 days) for BAL and blood sampling (6B n = 10; 23F n = 2, Table 1). These 12 carriage positive volunteers had been challenged with a mean dose of 66,617±45,637 CFUs (range 11,166–136,667 CFUs) per naris (Table 1). The proportion CD4+ memory T-cells positive for TNF, IL-17A or IFNγ were measured in BAL and blood and compared to controls without challenge (Table 1).
BAL cells and PBMCs from carriage negative volunteers were stimulated with pneumococcal antigens (HK-6B or 6B c/s) or influenza and cytokine (TNF, IL-17A or IFNγ) producing CD4+ memory T-cells were subsequently detected by ICS and flow cytometry. Pneumococcal-responding CD4+ memory T-cells were identified in BAL (Figure 1A and C and Figure S1 in the online repository) and PBMCs (Figure S2 in the online repository) in the absence of carriage. BAL CD4+ memory T-cells responding to heat-killed pneumococci (HK-6B) in the absence of carriage were TNF+ (0.25±0.19% vs media control 0.1±0.07% p = 0.02, paired T-test) and IL-17A+ (0.12±0.09% vs media control 0.04±0.02%, p = 0.01, paired T-test) but not IFNγ+, consistent with a Th-17 phenotype (Figure 1A). There was a positive correlation between TNF+ CD4+ memory T-cells and IL-17A+ CD4+ memory T-cells (Pearson r = 0.8; p = 0.009) in response to HK-6B. Similar observations were made when cells were stimulated ex vivo with concentrated pneumococcal culture supernatant (6B c/s) and compared to vegetone broth (vehicle) alone. We could detect pneumococcal-responding IL-17A+ (0.12±0.07 vs vehicle 0.06±0.05 p = 0.004) but not TNF+ or IFNγ+ CD4+ memory T-cells, again consistent with a Th-17 phenotype. In contrast, CD4+ memory T-cells responding to influenza stimulation, in the absence of carriage, were detectable in almost all BAL samples (Figure 1B and Figure S1 in the online repository) and these cells were TNF+ (0.33±0.21% vs media control 0.1±0.06%, p = 0.006, paired T-test) and IFNγ+ (0.27±0.22% vs media control 0.07±0.05%, p = 0.03, paired T-test) CD4+ memory T-cells, consistent with a Th-1 phenotype. IL-17A+ CD4+ memory T-cells were not detected in response to influenza (Figure 1B).
To corroborate our flow cytometry findings we stimulated BAL cells from colonised and non-colonised volunteers with HK-6B or left them untreated and measured secreted IL-17A (Figure 4A), TNF (Figure 4B), IL-2, IL-4, IL-6, IL-10, IL-22 and IFNγ (all Figure S4 in the online repository) by ELISA. Large quantities of IL-17A were detected in the culture supernatant of HK-6B stimulated BAL cells from non-colonised volunteers (stimulated mean±SD 14,907±10,843 vs non-stimulated 2,233±3,298 pg/ml, p = 0.06, Figure 4A). BAL cells from colonised volunteers and stimulated with HK-6B elicited significantly greater quantities of IL-17A protein compared to non-stimulated cultures (stimulated 22,393±10,830 vs non-stimulated 2,002±3,738 pg/ml, p = 0.03). HK-6B stimulated BAL cells from colonised volunteers produced 50% more IL-17A protein than HK-6B stimulated BAL cells from non-colonised volunteers but this difference was not statistically significant. IL-17A production did not correlate with the frequency of pneumococcal-responding IL-17A+ CD4+ T-cells in BAL detected by flow cytometry (r = 0.13, p = 0.68). IL-17A production did correlate with the number of alveolar macrophages per well (r = 0.59, p = 0.03) indicative of an alternative source of IL-17A, other than CD4+ memory T-cells, in BAL.
Comparisons between colonised and non-colonised groups, following pneumococcal stimulation, revealed no significant differences with the exception of TNF (198.9±476.8 vs non-colonised 5.8±3, p = 0.05 Mann-Whitney, Figure 4B). When corrected for background (by subtracting data from non-stimulated cells) the significance of this observation increased (TNF 195.7±476.5 vs non-colonised 2.67±2.45, p = 0.026 Mann Whitney).
We then hypothesised that the presence of IL-17A (and TNF) in stimulated culture supernatant would in turn elicit alveolar macrophage secretion of constitutively expressed BD2 protein [32] after 20 hrs but this was below the limit of detection in all samples (data not shown).
Human alveolar macrophages expressed both IL-17 RA (7438±1646 mean channel units, n = 5) and RC (3551±2426 mean channel units n = 6) sub-units consistent with our previous observations [30]. We thus used rhIL-17A and a modified OPKA assay to mimic CD4 T cell action and our hypothesis was that rhIL-17A could enhance the anti-pneumococcal response (independent of serotype) of human alveolar macrophages. To calculate a percentage increase or decrease compared to vehicle treated cells, CFU averages from each rhIL-17A dose and respective control were divided to obtain a ratio (Figure 5 and raw data in Table 3). We showed a dose dependent increase in macrophage uptake of pneumococci using 12.5 ng/ml, 125 ng/ml or 625 ng/ml concentrations of rhIL-17A (Figure 5 and Table 3).
Macrophage uptake of pneumococci was increased 26% in the presence of 125 ng/ml of rhIL-17A, compared to the 12.5 ng/ml dose (12.5 ng/ml Control: 14.2 vs rhIL-17A stimulated 10.5 CFU p = 0.013). Increasing the rhIL-17A dose to 625 ng/ml further increased pneumococcal uptake to 37% (12.5 ng/ml Control: 14.2 vs rhIL-17A stimulated 8.9 CFU p = 0.004) (Figure 5 and Table 3).
We correlated the OPKA data described above with IL-17 receptor RA and RC mean fluorescence intensity on BAL alveolar macrophages from a sub-set of the same volunteers (n = 6) to determine whether this response was mediated by the IL-17 receptor. Our hypothesis was that increased mean receptor expression positively correlates with increased percentage killing compared to vehicle at the 125 ng/ml dose. There were no significant correlations between OPKA data and expression of RA or combined expression of RA and RC. Contrary to our hypothesis, however, mean expression of RC (3551±2426), negatively correlated with killing (Spearman r = −0.9, p = 0.017).
We have shown that pneumococcal-responding IL-17A+ CD4+ memory T-cells are present at very low frequency in the healthy adult lung in the absence of carriage. Further, using a novel experimental human pneumococcal carriage (EHPC) model and post carriage BAL collection an episode of pneumococcal carriage resulted in a 17.4-fold increase and 8-fold increase in the percentage of IL-17A+ CD4+ memory T-cells in BAL and blood, respectively, compared to non-colonised volunteers. Using human alveolar macrophages as effectors we showed that rhIL-17A increased in vitro killing of S. pneumoniae in an opsonophagocytic killing assay. These are the first data of which we are aware to describe the relation of nasal carriage of a pathogenic organism and lung IL-17A responses in humans and together support a role for effector IL-17A+ CD4+ memory T-cell responses in the defence of the lung against pneumococcal infection in adults.
The two major strengths of this study are that we have described human CD4+ T-cell responses in the relevant mucosal site and after a defined period of nasal colonisation. Other investigators have identified and described pneumococcal-responding human CD4+ T-cells in blood [13], [19], [20], [33], [34] and upper respiratory tract mucosal tissue [13], [20], [35] but the initiation and duration of carriage was unknown. The sharp increase in cellular response seen immediately following an episode of carriage in this study, and not seen in similar volunteers challenged with live bacteria but without carriage [28], strongly supports a lung immunising role of carriage in adults. Although a study of subjects at high risk of pneumococcal disease (children, elderly) would be immunologically more relevant, it would clearly be ethically unacceptable in the context of human pathogen challenge.
Our data contrast with decreases in antigen-specific responses observed in blood in pneumococcal carriers in UK children [13] or in endemic areas such as the Gambia [33] and in UK patients with pneumonia [14], probably due to mucosal sequestration. Our data concur, however, with increased IL-17A responses in other studies [17] from an area with a high prevalence of pneumococcal carriage and disease (i.e. Bangladesh) compared to Swedish cohorts. The difference between our study and others may be due to differences in the timing of sample collection relative to exposure.
Human pneumococcal-responding IL-17A responses have been demonstrated previously in peripheral blood [17], [18], [33] and in adenoidal mono-nuclear cells [20], [35] but in our study we have examined the mucosal compartment where pneumonia becomes established – the lung, which has not been examined before. In a healthy adult population we showed, using flow cytometry, that pneumococcal carriage elicits high frequencies of IL-17A+ and TNF+ but not IFNγ+ cells, within 5 weeks of colonisation. We have used an extensive 7-colour panel that includes CD3+ and CD4+ antibodies that together with IL-17+ detection ensures that it is highly likely that the responses we have identified in this manuscript are derived from putative Th-17 cells rather than innate sources. The IL-17A dominant responses in BAL and blood contrast with studies that described higher IFNγ and lower IL-17A responses in blood [19], [33] and tonsil [20] from healthy and HIV affected [19] adults in Malawi and Gambia. There are multiple factors, including the tissue site examined, burden of disease and cellular plasticity [36] that may account for higher IFNγ in these studies and these differences between geographical areas of high and low pneumococcal carriage warrant further attention. It is likely, however, that both IL-17A and IFNγ from T-cell effector cells as well as T-regulatory cell populations play important but different roles in protecting the lung against the pneumococcus and pneunomococcal induced pathology [3], [37]–[40]. Furthermore, we identified increased IL-22 levels in some volunteers that were independent of the IL-17 response suggesting a separate source of IL-22 (possibly Th-22 cells), the diverse functions of which include maintenance of epithelial integrity [41] and remodelling [42]. Murine models of airway inflammation have shown that IL-22 can be pro-inflammatory (and thus pathological) in the presence of IL-17A but in the absence of IL-17A can be anti-inflammatory/tissue protective [43]. Determining the correct “balance” of Th-17, 22 and T-regulatory cells elicited following vaccination may be important for generating adaptive anti-pneumococcal responses that promote resolution and clearance and reduce immunopathology.
We have also measured the cytokine response from lung cells stimulated ex vivo with pneumococcus and shown that pneumococcal stimulated BAL cells (from non-colonised and colonised volunteers) produce IL-17A in quantities far greater than described in other studies using blood [17]–[19] or lymphoid tissue [20], [35]. The response from colonised volunteers was 50% greater compared to the non-colonised group who also had high levels of IL-17A following stimulation that is likely to be derived from non-Th-17 sources. This difference may be of relevance in vivo, however, since TNF [44], which we have shown to be significantly different between colonised and non-colonised groups using flow cytometry (Figure 2A) and ELISA (Figure 4B), and IL-22 [45] can both synergise with IL-17A to enhance epithelial derived CXC chemokine production, important for the recruitment and activation of neutrophils to the airway. It has been shown that murine [46] and human [47] alveolar macrophages can also produce IL-17A utilising a TLR-2 dependent mechanism [46] and this may have contributed to the IL-17A signal detected by our ELISA in both groups. An important role for the alveolar macrophage in the early hours of pneumococcal infection has been highlighted previously in murine models [48] and IL-17A from innate sources are likely to be involved [49].
In this study we showed a significant increase in pneumococcal killing by macrophages when exposed to 125 ng/ml of rhIL-17A, a concentration that is in line with previous publications showing an effect of IL-17 in this dose range [18], [45], [50]–[52]. This is also consistent with the study by Lu et al. [18] who showed that human neutrophils exposed to 100 ng/ml of rhIL-17A showed significantly increased pneumococcal killing. Our data are also in line with those of Higgins et al. [53] who showed that treatment of murine peritoneal macrophages with 2.5–50 ng/ml of rmIL-17A significantly enhanced killing of Bordetella pertussis. IL-17 also acts as a recruitment and survival factor for monocytes and macrophages [54], respectively, thus promoting macrophage-Th-17 interaction in the small volume of airway lining fluid [55].
Both RA and RC subunits are required for human IL-17A signalling with combined surface receptor density of RA and RC determining the magnitude of the response [56] but we did not find any positive correlations between our OPKA data and receptor expression. In contrast to our hypothesis, we observed a negative correlation between IL-17RC (but not RA) expression and macrophage killing activity at the 125 ng/ml dose. The modulation of killing by RC supports our observations of an IL-17-dependent effect in our assay system rather than a contaminant. Furthermore the negative correlation between IL-17RC (but not RA) expression and macrophage killing suggests that killing may be mediated by a different IL-17RA heterodimer other than RA:RC. RC may thus play a regulatory role in this process, separate from its pro-inflammatory role within the RA:RC dimer, fine tuning the phagocytic potential of alveolar macrophages and thus susceptibility to infection. There is evidence that IL-17 receptors play regulatory roles during the inflammatory response [30], [52]. Recent observations have shown that RD expression intensity can differentially regulate p38 mitogen-activated protein kinase and nuclear factor-kappa B pathways and more importantly the control of lung neutrophil recruitment in a CXCL2 dependent manner [52]. Evidence provided here and elsewhere thus suggests that the role of IL-17 receptors is more complex than initially appreciated and may differ depending on the context. The role of IL-17 receptors, other than the classical IL-17RA:RC heterodimer, on alveolar macrophage function in health and disease remains to be clarified and may determine the overall protective effect of Th-17 cells.
Taken as a whole, these results have important implications for vaccine design against pneumonia. First, they show that human nasal carriage can boost innate (alveolar macrophage function) and adaptive (TNF+IL-17A+ CD4+ memory T-cells) cellular lung immunity that may protect the lung from pneumococcal challenge and the establishment of infection in health, without significant recruitment of neutrophils. When these and other protective immunological mechanisms are compromised or the bacterial load overwhelms innate defence mechanisms the responses described in our study may synergise to enhance neutrophil mediated recruitment into the airspace. Second, we have begun to define the phenotypic and kinetic cellular responses elicited by pneumococcal carriage – a natural immunising event, thus providing a bench mark for vaccines that seek to protect against pneumonia.
|
10.1371/journal.pmed.1002142 | Tradeoffs in Introduction Policies for the Anti-Tuberculosis Drug Bedaquiline: A Model-Based Analysis | New drugs for the treatment of tuberculosis (TB) are becoming available for the first time in over 40 y. Optimal strategies for introducing these drugs have not yet been established. The objective of this study was to compare different strategies for introducing the new TB drug bedaquiline based on patients’ resistance patterns.
We created a Markov decision model to follow a hypothetical cohort of multidrug-resistant (MDR) TB patients under different bedaquiline use strategies. The explored strategies included making bedaquiline available to all patients with MDR TB, restricting bedaquiline usage to patients with MDR plus additional resistance and withholding bedaquiline introduction completely. We compared these strategies according to life expectancy, risks of acquired resistance, and the expected number and health outcomes of secondary cases.
For our simulated cohort, the mean (2.5th, 97.5th percentile) life expectancy from time of initiation of MDR TB treatment at age 30 was 36.0 y (33.5, 38.7) assuming all patients with MDR TB received bedaquiline, 35.1 y (34.4, 35.8) assuming patients with pre-extensively drug-resistant (PreXDR) and extensively drug-resistant (XDR) TB received bedaquiline, and 34.9 y (34.6, 35.2) assuming only patients with XDR TB received bedaquiline. Although providing bedaquiline to all MDR patients resulted in the highest life expectancy for our initial cohort averaged across all parameter sets, for parameter sets in which bedaquiline conferred high risks of added mortality and only small reductions in median time to culture conversion, the optimal strategy would be to withhold use even from patients with the most extensive resistance. Across all parameter sets, the most liberal bedaquiline use strategies consistently increased the risk of bedaquiline resistance but decreased the risk of resistance to other MDR drugs. In almost all cases, more liberal bedaquiline use strategies reduced the expected number of secondary cases and resulting life years lost. The generalizability of our results is limited by the lack of available data about drug effects among individuals with HIV co-infection, drug interactions, and other sources of heterogeneity, as well as changing recommendations for MDR TB treatment.
If mortality benefits can be empirically verified, our results provide support for expanding bedaquiline access to all patients with MDR TB. Such expansion could improve patients’ health, protect background MDR TB drugs, and decrease transmission, but would likely result in greater resistance to bedaquiline.
| Bedaquiline is a new tuberculosis (TB) drug approved by the United States Food and Drug Administration in 2012 for patients with multidrug resistant (MDR) TB without other treatment options.
Although the initial clinical trials of bedaquiline in combination with an optimized background regimen for MDR TB showed promising efficacy, one of these studies inexplicably had more deaths in the study group receiving bedaquiline.
The individual and public health benefits of providing bedaquiline to different categories of TB patients are unclear.
We used a mathematical decision model to simulate the potential effects of providing bedaquiline to different categories of TB patients based on their drug resistance patterns.
We found that strategies that conservatively limit bedaquiline access to all but the most resistant patients would minimize risks of resistance to bedaquiline but maximize risks of resistance to important background drugs such as moxifloxacin.
We predict that more liberal bedaquiline use strategies would lower transmission and improve health outcomes among secondary cases.
We found that if bedaquiline safety and efficacy are assumed to be sufficiently high, the optimal strategy in terms of individual patient life expectancy would be to provide bedaquiline to all patients with MDR TB.
Researchers should prioritize collecting additional data to establish a mortality benefit of bedaquiline.
If the safety of bedaquiline is confirmed, expanding bedaquiline access to all patients with MDR TB could improve patients’ health, prevent resistance to background MDR TB drugs, and decrease transmission, but would likely result in greater resistance to bedaquiline.
| Only approximately 50% of the 111,000 people started on treatment for multidrug-resistant tuberculosis (MDR TB) in 2014 are likely to be successfully treated [1]. The remainder will experience high mortality, risk acquisition of extensively drug-resistant (XDR) TB, and may continue to infect others. New antibiotics have the potential to improve both prevention and treatment of highly drug resistant TB. Bedaquiline and delamanid recently became the first new drugs approved for TB treatment in over 40 y [2,3], and other promising drugs such as pretomanid are in development [4]. Effective drug use policies will be necessary to obtain maximal benefit from these new drugs while also managing risks of resistance.
Although clinical management of TB relies on strong multidrug regimens, the initial discovery and development of new TB drugs often occur in isolation. Optimizing multidrug regimens is complicated in both theory (e.g., by the number of drugs, limited data on drug efficacy and interactions, and the prevalence of existing resistance) and practice (e.g., by lack of access to patients’ full drug susceptibility profiles and limited opportunity for controlled trials) [5,6]. Thus, decisions about how best to introduce and combine new TB drugs have relied heavily on expert opinion. Limited guidance exists beyond common-sense strategies, such as never to add a single drug to a failing regimen, and broad considerations, such as the number of drugs and their side-effect profiles [5,7].
Here, we present a Markov decision model to begin formalizing a rational basis for decisions about drug introduction. Using the model, we outline the tradeoffs involved in deciding which patients should receive a new anti-TB drug, based on both their outcomes and those of their immediate contacts. We explore a continuum of policies ranging from most conservative (i.e., restricting the new drug entirely or for use only among the most highly resistant patients) to most liberal (i.e., allowing all patients with MDR TB to receive the new drug). Though the general framework of our analysis is broadly generalizable, we focus this paper specifically on the new TB drug bedaquiline. Bedaquiline was approved by the United States Food and Drug Administration in 2012 for use in MDR TB patients without other treatment options on the basis of its Phase IIb trial culture conversion results. However, concerns about resistance and a mortality imbalance observed in the pivotal Phase IIb trial have generated controversy about the appropriate role of this new drug [8–11]. A formal approach to assessing potential bedaquiline use strategies is therefore especially appropriate.
This modeling study was based on previously published aggregate data and thus did not require ethical approval. To evaluate the impact and potential tradeoffs of different bedaquiline introduction strategies, we created a Markov decision model following a hypothetical cohort of patients initiating MDR TB treatment and their immediate contacts. A model description is provided below, with additional details available in S1 Appendix sections 2, 7, and 8.
Our assumed population was a cohort of European men initiating MDR TB treatment at age 30. All men were assumed to be bedaquiline susceptible at baseline and have either MDR TB without additional resistance (“MDR” from here), MDR TB with additional resistance to either at least one fluoroquinolone or at least one second-line injectable, but not both (“PreXDR”), or MDR TB with additional resistance to at least one fluoroquinolone and at least one second-line injectable (“XDR”). We assumed that 6.7% of patients initially had XDR TB, 26.2% of patients initially had PreXDR TB, and the remaining 67.1% of patients had MDR TB without additional resistance to the fluoroquinolones or second-line injectables, as observed in one published multi-country patient cohort [12].
Fig 1 displays the categories of health states and transitions included in our model. Modeled health states were defined based on TB culture status (positive, negative, or stable cure), treatment regimen (optimized background regimen [OBR]; OBR plus bedaquiline; or no treatment), and resistance pattern (to bedaquiline and background drugs). Transitions between these states included culture conversion, relapse, routine or premature cessation of treatment, treatment re-initiation after cessation, regimen change, resistance acquisition, and death. We assumed that resistance was acquired in a stepwise fashion (i.e., to one drug at a time), and that patients could only relapse after treatment (i.e., culture conversions were only modeled if sustained through the end of treatment). We also assumed that TB-related mortality and acquired resistance rates applied only to patients who were culture-positive, and that some patients self-cured even in the absence of TB treatment.
We considered the following treatment strategies: withholding bedaquiline from all patients, providing bedaquiline to patients with XDR TB only, providing bedaquiline to patients with PreXDR or XDR TB, or providing bedaquiline to all patients with at least MDR TB. We did not allow treatment to differ based on bedaquiline resistance patterns, reflecting the current lack of a validated test with breakpoints defining clinically relevant bedaquiline resistance [5].
For the strategy in which all patients with MDR TB were eligible for bedaquiline, we assumed that all patients received bedaquiline from the beginning of treatment. For the more conservative strategies, we assumed a 13-wk average lag time after acquisition of or treatment initiation with the relevant resistance pattern to account for a delay in obtaining results of second-line drug susceptibility testing (DST). We compared these results to an analysis assuming no lag time, reflecting the potential impact of widespread rapid second-line DST availability.
We considered mortality, resistance, and transmission outcomes. To assess mortality, we compared the average life expectancy from initiation of MDR TB treatment across the different bedaquiline use strategies, and to assess resistance, we recorded the number of patients who acquired particular resistance patterns under each treatment strategy. To assess transmission, we estimated the number of secondary cases as well as life years lost to secondary cases. The methodology used for these estimates is described below.
To assess transmission, we first calculated an approximate number of secondary cases infected by our initial cohort from initiation of MDR TB treatment as follows. We assumed that a single infectious, untreated, drug susceptible individual would infect others at a rate of ten infections per year, and that each infected individual had a 10% chance of progressing to active disease at some point in his or her lifetime [13–15]. We also allowed for varying transmission costs depending on the resistance pattern of the infecting patient, ranging from 0.5 for XDR to 0.7 for MDR in the absence of bedaquiline resistance [16–22]. To estimate the number of secondary cases produced per year by a single untreated, culture-positive case, we multiplied the infection rate by the progression probability and the applicable transmission cost. We reduced this value 5-fold for individuals receiving treatment, assuming that treatment would reduce infectiousness by a value similar to the relative infectiousness of smear-negative as compared to smear-positive TB [13,23,24]. We then converted these values into weekly infection rates and applied them to the individuals in our model based on their culture, resistance, and treatment status at each time step. Greater details and justification for the parameters used are provided in the parameter table in S1 Appendix section 7.
The life expectancy of each secondary case was calculated based on the resistance pattern of the index case at the time of the infection event, with background mortality rates reflecting those used for our initial cohort (i.e., assuming similar demographics to our initial cohort). Secondary cases were subjected to the same treatment strategy as the initial cohort. We assumed secondary cases had similar delays to detection as our initial cohort but were immediately recognized as MDR upon presentation to the health system. Detection of additional resistance was subject to similar delays as for the index patients. To calculate the expected number of life years lost to secondary cases under each treatment scenario, we combined these estimates of the life expectancy among secondary cases with our estimates of the number of secondary cases. These values are intended to be a first approximation of the transmission impact of differing bedaquiline use policies and do not capture the full range of MDR TB transmission dynamics, including within-household transmission and time to disease progression.
Parameters describing TB natural history and outcomes in the absence of bedaquiline were taken from published cohorts, clinical trials, and meta-analyses [25–27]. These parameters were held fixed throughout our analysis. Parameters describing the effect of bedaquiline were derived from the bedaquiline pivotal trials [8,28] and more recent cohorts [3,29,30]. An overview of these studies is included in S1 Appendix section 1. Because only small numbers of patients receiving bedaquiline-containing regimens had completed treatment at the time of this analysis, we explored wide uniform ranges of values for key bedaquiline-associated parameters as described in Table 1. Additional mortality results based on triangular distributions are included in S1 Appendix section 4 and are qualitatively similar to the results from our uniform distributions included in the main text.
All analyses were performed in TreeAge Pro 2015 R2.2. We assumed that transitions occurred on a discrete weekly basis, allowing us to capture potentially rapid changes in infectiousness, prognosis, and resistance patterns. From our bedaquiline-associated parameter ranges, we sampled 5,000 random parameter sets and for each estimated expected values for life expectancy, resistance acquisition patterns, and number and outcomes of secondary cases under each treatment scenario. We then calculated the average outcome for each strategy across all parameter sets, as well as the number of parameter sets for which each strategy was optimal (i.e., produced the maximum or minimum expected outcome across all strategies, as appropriate).
Fig 2 summarizes the optimal bedaquiline use strategies from each simulation for a range of mortality, resistance, and transmission outcomes. An overview of these results and additional analyses for each outcome are provided below.
Providing bedaquiline to all patients with MDR TB maximized the life expectancy of our initial cohort in 76.8% of 5,000 simulations (Fig 2). In nearly all remaining simulations, the optimal strategy was to withhold bedaquiline from all patients, suggesting that the benefits of bedaquiline did not outweigh potential added mortality risks. Intermediate bedaquiline use strategies were optimal in fewer than 1% of simulations. Average life expectancy following the best strategy for each individual parameter set was 36.12 y (after MDR TB treatment initiation at age 30) compared to 34.67 y under the worst strategy for a difference of 1.45 y.
To understand which parameters were most responsible for the variation in life expectancy outcomes associated with each strategy, we first created a tornado plot (Fig 3) showing the impact of varying each parameter to its low and high values while keeping all other variables fixed at their midpoints. As shown in this figure, the most influential parameters are the rates of added mortality and culture conversion associated with bedaquiline. Fig 4 displays the impact of these two parameters on the optimal bedaquiline use strategy more directly, with the remaining parameters set to their midpoints as well as their extreme values that most favored and opposed use of bedaquiline in all patients. Providing bedaquiline to all patients is preferred when bedaquiline strongly reduces median time to culture conversion and has low added mortality risk, whereas withholding bedaquiline from all patients is preferred when bedaquiline has high mortality risks and a low impact on time to culture conversion. For example, when all other parameters are fixed at their midpoints, the “All MDR” strategy is preferred whenever bedaquiline reduces the median time to culture conversion compared to OBR only by at least 35%, regardless of the added mortality risk (within the range explored). Similarly, when all other parameters are fixed at their midpoints, the “All MDR” strategy is always preferred whenever the added mortality risk associated with bedaquiline is less than 0.00025 per week, or 1.3 excess deaths per 100 person-years, regardless of the effect of bedaquiline on time to culture conversion. Of note, the “All MDR” strategy is preferred for a majority of the combinations of culture conversion and added mortality, regardless of the values of all other parameters, highlighting the importance of prioritizing these values for future study.
Table 2 displays the effect of the DST methods available to detect PreXDR and XDR TB on life expectancy under the different bedaquiline use strategies. The rapid DST method, which shortens the lag time for eligible individuals to receive bedaquiline, increased the average life expectancy for both the “XDR only” and “PreXDR+XDR” strategies. Notably, the availability of rapid second-line DSTs also changes the proportion of times each strategy would be optimal in terms of life expectancy, with the “all MDR” scenario providing optimal life expectancy for 69.3% of scenarios, compared with 16.9% for “PreXDR+XDR” and 13.7% for “None.” The “XDR only” strategy was chosen only once out of 5,000 runs. These results suggest that widespread availability of rapid second-line DSTs could alter decisions about optimal bedaquiline use. However, the “all MDR” scenario was still the most frequently chosen, and its average life expectancy still exceeded those for the “PreXDR+XDR” as well as “XDR only” strategies, suggesting that the potential benefits of making bedaquiline available for all patients with MDR TB extend beyond simply shortening the time to bedaquiline initiation for patients with more extensive resistance. Additional information on the parameter space in which each strategy would be preferred if rapid second-line DSTs were available is provided in S1 Appendix section 5.
Fig 2 and Table 3 show the impact of different drug use strategies on acquired resistance to the new and existing drugs in our initial cohort. The best strategy to avoid resistance to bedaquiline was to strictly constrain bedaquiline availability. The simulation mean percentage of people acquiring resistance to bedaquiline was 5.88% (2.5th percentile 2.18%, 97.5th percentile 9.45%) in the scenario providing bedaquiline to all patients with MDR TB, compared with 3.50% (1.30%, 5.62%) when restricting bedaquiline for patients with XDR TB only. However, expanding bedaquiline availability is predicted to reduce the rate of acquired XDR TB by providing additional protection to the existing drugs. The percentage of people acquiring XDR TB was 2.56% (1.09%, 7.68%) in the scenario providing bedaquiline to all patients with MDR TB, compared with 9.82% (no variability, as non-bedaquiline parameters are assumed fixed) when restricting bedaquiline for patients with XDR TB only.
When we only consider scenarios in which at least some patients are eligible for bedaquiline, complete resistance to the new and existing drugs (XDR+bedaquiline resistance [BDQR]) was minimized most often by the intermediate strategy of providing bedaquiline to patients with PreXDR and XDR TB only. However, the “XDR only” strategy is preferred in 10.8% of the 5,000 simulation runs and the “all MDR” strategy in 3.6% of runs, indicating that the optimal decision for this outcome is parameter-dependent. This pattern reflects the differential effects of the bedaquiline use strategies on patients with different initial resistance patterns (see S1 Appendix section 6). For many (though not all) parameter sets, providing bedaquiline to all patients with MDR TB minimized the number of cases of acquired XDR+BDQR among patients with initial MDR or PreXDR TB, but maximized the number of cases of acquired XDR+BDQR among patients with initial XDR TB. However, the absolute differences in the number of cases of acquired XDR+BDQR across scenarios are small when bedaquiline is provided to at least some categories of patients, indicating that the costs of making a suboptimal decision with respect to this variable may be limited. The results are sensitive, however, to assumptions about time to treatment initiation; if we assume rapid second-line DSTs are available, providing bedaquiline to all patients with MDR TB is the most frequently preferred strategy (see S1 Appendix section 6).
As shown in Table 4, the total number of secondary cases produced from the time of MDR TB treatment initiation was less than one per person across all treatment strategies, indicating non-sustainable transmission in the population from the point of appropriate treatment initiation. This number was higher but remained below one if we assumed individuals were initially untreated, reflecting the high mortality rate and lack of diagnostic delay in our model. Making bedaquiline available to all patients with MDR TB was the preferred strategy to minimize the number of secondary cases for all 5,000 simulation parameter sets and the years of life lost amongst secondary cases for all but one (Fig 2).
New anti-TB drugs such as bedaquiline hold much promise to reduce morbidity and mortality associated with drug resistance. In this paper, we performed a decision analysis to explore the potential impact of different bedaquiline use strategies on a range of individual and public health outcomes. Different strategies may be preferred based on the outcome of primary interest (e.g., minimize resistance, minimize years of life lost), illustrating the tradeoffs involved in decision-making for the introduction of new antibiotics.
Drugs for which the risk of mortality due to adverse events exceeds expected reductions in mortality should not be used regardless of their potential public health benefits. Our model most often preferred providing bedaquiline to all patients with MDR TB for parameter sets in which bedaquiline introduced only a small added risk of mortality and substantially decreased the median time to culture conversion, and favored withholding bedaquiline from all patients when it was associated with high mortality and small declines in time to culture conversion. The frequency with which each of these parameters was preferred reflects the shape of our assumed uncertainty distributions, which we chose to be conservative estimates of the impact of bedaquiline; modifying these assumptions changes the quantitative results, but not the finding that in some cases withholding bedaquiline from all patients would be the preferred strategy (see S1 Appendix section 4). These results demonstrate the vital importance of continued research into bedaquiline safety and efficacy and support prioritizing additional safety data over secondary concerns such as the risk of acquired resistance. Although a model such as this can support such research prioritization, decisions about safety of bedaquiline must ultimately be based on data from real patients. Thus far, interim cohort analyses of patients receiving bedaquiline outside of trial settings have not identified excess bedaquiline-associated mortality [29,30]; however, continued data from compassionate use programs and, in particular, phase III trial results are needed to verify that the unexplained mortality imbalance of the pivotal phase IIb trial was not drug related.
Antibiotic introduction strategies may affect rates of acquired resistance to the new drug, existing drugs, or both. In general, we would expect more expansive access to a new drug to promote resistance to the new drug while preventing resistance to existing drugs. These expectations are reflected in our results. Acquired bedaquiline resistance occurred most often under the most liberal bedaquiline use policy (providing bedaquiline to all patients with MDR TB); however, this same policy was most effective at preventing new cases of PreXDR and XDR TB. The effects of expanding access to a new drug on composite resistance to new and existing drugs are less clear-cut. When considering only strategies providing bedaquiline to at least some categories of patients, the majority of our simulations predicted an intermediate strategy targeting bedaquiline to patients with PreXDR and XDR TB only to minimize the combination of XDR plus BDQR. However, both the “All MDR” and “XDR only” strategies were preferred for some combinations of parameter values, and differences in the proportions of people acquiring XDR+BDQR across different strategies were small. Because tuberculosis antibiotic resistance cannot be horizontally transferred, the spread of bedaquiline resistance to other patient cohorts is restricted to patients directly infected with bedaquiline-resistant bacteria. As such, although future spread of bedaquiline resistance will limit its benefits in terms of both patient health and protection for other drugs, we do not expect bedaquiline resistance to appear except in the context of background resistance patterns to which it is already being applied.
For this paper, we limited our assessment of future transmission of TB and drug resistance to the second generation of infected patients. We found that, for all but one of the 5,000 parameter sets tested, making bedaquiline available to all patients with MDR TB would minimize the total number of and expected number of life years lost to secondary cases. This relationship can be explained by the correlation between severe and highly infectious disease within our model. For diseases and treatments for which this assumption does not hold, associations may appear in the opposite direction [37]. Future drug development and policy changes may also affect the relationship between new drug use strategies and outcomes among potential secondary cases. Bedaquiline use strategies chosen now could alter the effectiveness of potential future TB regimens incorporating both bedaquiline (e.g., the NC-005 trial of bedaquiline, pyrazinamide, and pretomanid) and background drugs such as pyrazinamide and the fluoroquinolones (e.g., the STAND trial of pretomanid, moxifloxacin, and pyrazinamide) [3]. Of course, the desire to be prepared for the range of outcomes that could result from these trials must be weighed against the need to provide the best available care to patients presenting today. A full modeling analysis of these costs and benefits would require a transmission dynamic structure not included here.
This study has several limitations. We have not explored the full range of potential bedaquiline use strategies, e.g., as an early drug substitution method to prevent hearing loss during MDR TB treatment. For simplicity, we held the natural history and treatment parameters unrelated to bedaquiline fixed throughout our analysis, which does not reflect the potential uncertainty and heterogeneity in these parameters. Many of these estimates were based on large meta-analyses with data from multiple countries, allowing us to average over but not fully address the variability expected, e.g., in settings with standardized versus individualized treatment regimes. We assumed that our initial cohort was comprised of 30-year-old European men, which may differ from the target population of bedaquiline in many settings; however, as this assumption was used only in defining background mortality rates, it is most likely to affect the magnitude rather than the direction of the observed effects. Similarly, the effects of our particular background distribution of resistance are likely mitigated by the range of explored scenarios, which incrementally account for expanded access of bedaquiline to patients with XDR, then PreXDR+XDR, and finally all MDR. Changing the HIV status of this cohort could have greater effects if bedaquiline is found to have differential impact on HIV-positive and HIV-negative individuals. Similarly, we may see differential effects of bedaquiline if the background regimen varies substantially from the data on which our model was based, as in the STREAM II trial of shorter MDR regimens [3]. This limitation is especially relevant given the new World Health Organization guidelines that support the use of shortened MDR regimens for patients without anticipated second-line resistance, though we expect the differences between our PreXDR, XDR, and No Bedaquiline strategies to remain unchanged under this new policy [38]. Finally, we assumed that the efficacy of the background regimen did not differ depending on bedaquiline use; however, this may not be the case if it is necessary to modify the background regimen to avoid providing bedaquiline in combination with other QT-interval prolonging drugs.
Our results support the prioritization of verifying a mortality benefit of bedaquiline for patients with MDR TB. If such a benefit can be verified, they may provide support for expanded access of bedaquiline beyond the strict qualifications of compassionate use programs to all patients with MDR TB, particularly in settings where rapid second-line drug susceptibility testing is unavailable. Policymakers considering such expanded use should weigh the benefits of extending access to bedaquiline for all MDR TB patients seen in this analysis (including lower proportions of people acquiring resistance to background drugs and decreased onward transmission) against its potential drawbacks (including increased resistance to bedaquiline, as explored here, and the need to change the background regimen to avoid combining multiple QT-prolonging drugs, which we have not addressed).
|
10.1371/journal.pntd.0004088 | Snakebite is Under Appreciated: Appraisal of Burden from West Africa | Snakebite envenoming (SBE) is a major problem in rural areas of West Africa (WA). Compared to other Neglected Tropical Diseases (NTD), the public health burden of SBE has not been well characterized. We estimated the impact of snakebite mortality and morbidity using the Disability Adjusted Life Years (DALYs) metrics for 16 countries in WA.
We used the reported annual number of SB deaths and mean age at time of SB and converted these into years of life lost (YLL). Similarly, the years of life lived with disability (YLD) were estimated by multiplying the number of amputations by the respective disability weight of 0.13.
In WA, the annual cases of SB mortality and amputations ranged from 24 (95% Confidence Interval: 19–29) and 28 (17–48) respectively in Guinea-Bissau with the highest estimates of 1927 (1529–2333) and 2368 (1506–4043) respectively in Nigeria. We calculated that the annual DALYs associated with a SB death ranged from 1550 DALYs (95%CI: 1227–1873 DALYs) in Guinea Bissau to 124,484 DALYs (95%CI: 98,773–150,712 DALYs) in Nigeria. The annual DALYs associated with amputation for the two countries were 149 DALYs (95%CI: 91–256 DALYs) and 12,621 DALYs (95%CI: 8027–21,549 DALYs) respectively. The total burden of SBE was estimated at 319,874 DALYs (95% CI: 248,357–402,654 DALYs) in the 16 countries in WA. These estimates are similar, and in some instances even higher, than for other NTDs encountered in WA (e.g., Buruli ulcer, Echinococcosis, Intestinal Nematode Infections, Leishmaniasis, Onchocerchiasis, Trachoma and Trypanosomiasis) as reported in the Global Burden of Diseases 2010 (GBD).
The public health burden of SBE in WA is very substantial and similar to other more widely recognized NTDs. Efforts and funding commensurate with its burden should be made available for the control of snakebite in the sub-region.
| Snakebite envenoming (SBE) is a major problem in rural West Africa (WA). However, despite the high incidence of SBE in this region, government funding for the prevention or treatment of SBE is generally limited. In this analysis, we attempted to estimate how the public health burden of SBE compares to other more widely recognized Neglected Tropical Diseases (NTD). To this end, we estimated the impact of SBE mortality and morbidity based on the methodology outlined in the global burden of disease and reported our results in Disability Adjusted Life Years (DALYs) for 16 countries in WA. We calculated the total burden of SBE in WA at 320,000 DALYs (95% CI: 248,000–403,000 DALYs) per year with the least and highest burdens in Guinea-Bissau and Nigeria accounting for 0.5% and 43%, respectively. The vast majority of the public health burden (91%) is attributed to early mortality. We conclude that the public health burden of SBE in WA is substantial and similar to, and in some cases even exceeds, other more widely recognized NTDs such as Buruli ulcer, Echinococcosis, Intestinal Nematode Infections, Leishmaniasis, Onchocerchiasis, Trachoma and Trypanosomiasis. Efforts and funding commensurate with its public health burden should be made available for the control of snakebite.
| Snakebite envenoming (SBE) is a major public health problem among communities of the savanna region of West Africa, notably in Benin, Burkina-Faso, Cameroon, Chad, Ghana, Nigeria, Senegal and Togo [1,2,3,4]. The precise incidence of snakebite is difficult to determine and is often grossly underestimated. An early estimate in northeastern Nigeria reported a bite incidence of 500 per 100,000 population per year with a 12–20% natural mortality, with carpet vipers (Echis ocellatus) accounting for at least 66% [5]. However, this study probably exaggerated the overall incidence by extrapolating from data in selected areas notorious for their incidence of snakebites. Up to 10% of hospital beds may be occupied by SBE patients in certain areas of the country. A recent global reappraisal estimated 10,001 to 100,000 snakebite envenomings with an incidence of 8.9–93.3/100,000 persons per year with an estimated 1,001 to 10,000 deaths and a mortality rate of 0.5–5.9/100,000 persons per year occurring in the West African sub-region [6]. A more recent study estimated over 314, 000 bites, 7300 deaths and nearly 6000 amputations occurring annually in sub-Saharan Africa (SSA) [4]. As a condition affecting poor vulnerable rural dwellers, it is not only a major health problem but also a major impediment to economic prosperity from loss of income following initial incapacitation, hospitalization, long-term disabilities and premature deaths [7]. It is preventable and treatable with antivenom which has been shown to be cost effective [8]. In this analysis, we estimated the impact of snakebite mortality and morbidity using the Disability Adjusted Life Years (DALYs) metrics for 16 countries in Western Africa (WA). This will allow for comparison to other diseases as well as guide prioritization of resource allocation.
From the most recent reliable literature available, projected annual burden of SBE in Sub-Saharan Africa was derived using a meta-analytic approach which has been described in detail elsewhere [4]. In summary, SBE data was obtained using a meta-analytic approach based on indexed, non-indexed or grey literature and conference proceedings over the past 40 years. Studies included in the analysis were categorized based on type of survey (national, household and hospital studies) and location whether conducted in urban or rural areas; with the latter representing 95% of envenoming. The pooled incidence rates, amputation rates and mortality rates were obtained and applied to the population size to derive the mortality and amputation estimates (see S1 Annex) [4]. For each country in Western Africa the annual number of snakebite deaths and mean age at time of envenoming was obtained from the analysis and from the literature respectively. The corresponding Years of Life Lost (YLL) was derived for each of the countries, following the methodology outlined in the latest global burden of diseases report [9], which applies a standard loss function specifying the years of life lost due to death at a specific age. The standard loss function is based on projected frontier period life expectancy at birth for Japan and South Korea in the year 2050 estimated at 91.9 years and is not discounted [9]. Thus, we defined the YLL due to SBE in each country as 91.9 years minus the mean age at the time of envenoming. The mean ages of SBE were not available for all countries included in our analysis, but were reported for Chad at 25.2 years, Niger at 29 years, Nigeria at 26 years and Mali at 28 years [10,11,12,13]. So, since SBE consistently occurs in victims at a mean age in the late twenties, we made the simplifying assumption that SBE occurs in the 25–29 year age bracket and applied the standard loss function that corresponds to the this age bracket for all countries in our analysis, which is 64.6 years [9]. We then multiplied the number of SBE-related deaths in each country (Table 1, column 2) by 64.6 years to calculate the YLL. Similarly, the Years of Life Lived with Disability (YLD) were estimated by multiplying the number of amputations (Table 1, column 3) by the respective disability weight of 0.13 and applying this disability weight for the remainder of undiscounted local life expectancy [9,14]. In this age group, the remaining local life expectancies for the 16 countries ranged from 37 years in Sierra-Leone to 45 years in Ghana and Senegal (Table 1, column 4). The sum of YLL and YLD then defined the total DALY burden for each country.
Data on mortality and amputations obtained from Chippaux, 2011 [4] are presented in Table 1. The resulting burden was subsequently derived as described above and compared to other NTDs reported in the Global Burden of Diseases (GBD). In sum, SBE is associated with 319,874 DALYs annually (95% Confidence Interval: 248,357–402,654 DALYs) in the 16 West African countries included in the analysis. Most of the public health burden is due to early mortality, with YLL accounting for 290,275 DALYs (95%CI: 229,911–352,135 DALYs) and YLD accounting for 29,599 DALYs (18,446–50,519 DALYs). The highest local public health burden associated with SBE is estimated for Nigeria, at 137,105 DALYs or 43% of the total burden, followed by Ghana 22,243 DALYs, Burkina Faso 21,283 DALYs, Niger 18,833 DALYs and Cameroun 18,690 DALYs. The lowest public health burden is estimated for Guinea Bissau at 1,699 DALYs or 0.5% of the total burden [Table 1].
Using sub-regional level alternative data that reported low and high estimates of 1504 and 18654 annual snakebite deaths for WA by Kasturiratne et al 2008 [6] yielded burden of YLL from SBE deaths of 97,158 DALYs and 1,205,048 DALYs for low and high estimates respectively. The derived high estimate is 3.77 times higher than that obtained using data from Chippaux 2011 [4].
Similarly, using recent alternative estimates of annual snakebite deaths reported in a WHO document for Benin Republic 650, Burkina Faso 200 and Togo 199 yielded alternative YLL values of 41,990 DALYs, 12,920 DALYs and 12,855 DALYs for those countries respectively [15,16].
In the current reappraisal of data from WA, SBE accounted for 320,000 DALYs although using higher mortality estimates the YLL could be as high as 1.2 million DALYs [6]. Our estimate of 0.32 million DALYs is higher than the worldwide burden estimated for Buruli ulcer, Echinococcosis, Leprosy, Trachoma, Yaws and Yellow Fever. The estimate is also higher than the burden of African Trypanosomiasis, Leishmaniasis and Onchocerciasis within the 16 countries in WA region [17]. It is also higher than that of Podoconiosis the only other non-communicable disease in the expanded WHO NTD list. Compared to NTDs reported in the Global Health Estimates (GHE) for 2012, SBE has the fourth highest burden in the 16 WA countries, ranking below Schistosomiasis, Lymphatic Filariasis and Rabies [17] (Fig 1). Despite these estimates, SBE remains under-recognized. The resources allocated are not commensurate with its burden. In a study that evaluated funding for developing world health from 42 major donors (comprising industrialized countries 23, international financial institutions 5, multinational pharmaceutical companies 6 and philanthropic foundations 8), annual donor dollar direct funding for 8 of the ten NTDs (Fig 1) ranged from $3.30 per DALY for Intestinal Nematode Infections to $146.96 per DALY for Onchocerchiasis [18]. There is no evidence that any amount was provided for SBE by these donors during the period of the survey.
The difference in burden estimates from the two studies might have arisen from their methodologic approaches [4,6]. The SBE data reported by Chippaux 2011 [4] was obtained using a meta-analytic approach as described above. In contrast the study of Kasturiratne et al 2008 [6] modeled data from electronic databases, indexed and grey literature from 1985. They provided lowest and highest SBE estimates and rates when more than one source was available from a country. This led to a very wide range and imprecise estimates. Furthermore, country-level estimates were not provided and data from Chad, Ghana, Guinea-Bissau, Liberia, Nigeria and Sierra-Leone were not used to derive the mortality estimates for the WA sub-region. They extrapolated from data in adjacent countries, ignoring the geographical variations in snake-bite incidence. Both studies have common limitations. The authors did not choose the survey sites, studied variables and analytical strategies of the data. Most of the collected data were incomplete and spotty, resulting in a questionable representativeness. Nevertheless, some extrapolations were corroborated by national health statistics of some countries, such as in Benin [19].
Disability Adjusted Life Years (DALYs) are the sum of two components: years of life lost (YLLs) and years lived with disability (YLDs). The DALY represents one of the few metrics available that could estimate acute and chronic effects and allow for comparison of significance of burden of several conditions. With a few exceptions, notably Rabies with nearly 100% mortality, most of the NTDs currently listed by the World Health Organization (WHO) and those on the expanded list are disablers rather than killers. In contrast SBE is both a disabler and killer with DALYs accruing from both components. In WA, SBE is an important killer and most of the DALYs (over 90%) accrued from early deaths. These deaths are partially driven by envenoming from saw-scaled or carpet vipers (Genus Echis) which cause a high mortality of about 12–20% without antivenom therapy. However, our analysis has been conservative given only amputation was used as the main disability. Several important but rare sequelae (e.g., blindness, malignant ulcers, fetal loss, cognitive and pyschological impairment) were not considered. In WA the frequency of venom ophthalmia and blindness from cobra spits is <0.01% although blindness rarely may result from carpet viper induced ocular bleeding [20,21]. We have also observed 1 case of fetal loss out of 1800 SBE cases or <0.1% but no reports of cognitive/psychological impairment have been made from WA in contrast to Asia [22,23]. This underestimates the total burden and the contribution of DALYs accrued from YLD.
Globally, the public health significance of SBE is generally neglected and underappreciated. It is not among the WHO’s 17 major NTDs although it is mentioned among the ‘other neglected conditions’ (http://www.who.int/neglected_diseases/diseases/en/. However, there is no official WHO program for its prevention or treatment. For the first time, the GBD 2010 provided disease burden estimates for these ‘other NTDs’, i.e., amoebiasis, cryptosporidiosis, trichomoniasis, scabies, fungal skin infections, and venomous animal contact including snakebite, although they are not listed under the NTD and Malaria category. Out of the approximately 48 million DALYs ascribed to both groups of NTDs, venomous animal contact was projected to account for 2.72 million DALYs in the GBD 2010 [24,25]. Interestingly, the annual deaths from SBE in India alone was estimated at 45,900 in the rigorously conducted Million Death Study [26]. While there may be certain minor differences between WA and India, using the approach in this study will translate to 2.97 million DALYs from SBE related YLL in India alone. Thus, estimates from WA and India when combined with the burden from Latin America, Papua New Guinea, the rest of Africa and Asia would be very substantial and much more than the current gross underestimation. Indeed, global burden, using reported high mortality estimates of 93,945 annual deaths worldwide by Kasturiratne et al [6], would result in 6.07 million DALYs.
Effective antivenom therapy has been shown to prevent death from SBE by at least 75% and is a very cost-effective intervention with an incremental cost-effectiveness ratio of $100/DALY averted [8,27]. With expanded access to appropriate and affordable antivenom therapy, the burden of SBE will be considerably curtailed. About $33.61 million (95% Confidence Intervals: $25.85-$43.03 million) will be required annually to control SBE in WA.
This analysis is subject to a number of limitations, including data scarcity, variability and inherent difficulties in accurately estimating the number of incident cases reported in SBE studies. We used the approach adopted by WHO in 2012 and the GBD 2010 in computing DALYs, i.e., with a time discount rate of 0% and no age-weighting [9,24,25]. This is now the standard way to assess disease burden but compared to the previous method it leads to a substantial increase in the absolute number of DALYs lost and a relative increase in the share of DALYs at the extremes of life.
In conclusion, SBE is a major public health problem with a burden higher than that of most other NTDs in the WA sub-region. Commensurate efforts and funding compared to its burden should be made available for control globally and in the sub-region.
|
10.1371/journal.pcbi.1005909 | Molecular recognition and packing frustration in a helical protein | Biomolecular recognition entails attractive forces for the functional native states and discrimination against potential nonnative interactions that favor alternate stable configurations. The challenge posed by the competition of nonnative stabilization against native-centric forces is conceptualized as frustration. Experiment indicates that frustration is often minimal in evolved biological systems although nonnative possibilities are intuitively abundant. Much of the physical basis of minimal frustration in protein folding thus remains to be elucidated. Here we make progress by studying the colicin immunity protein Im9. To assess the energetic favorability of nonnative versus native interactions, we compute free energies of association of various combinations of the four helices in Im9 (referred to as H1, H2, H3, and H4) by extensive explicit-water molecular dynamics simulations (total simulated time > 300 μs), focusing primarily on the pairs with the largest native contact surfaces, H1-H2 and H1-H4. Frustration is detected in H1-H2 packing in that a nonnative packing orientation is significantly stabilized relative to native, whereas such a prominent nonnative effect is not observed for H1-H4 packing. However, in contrast to the favored nonnative H1-H2 packing in isolation, the native H1-H2 packing orientation is stabilized by H3 and loop residues surrounding H4. Taken together, these results showcase the contextual nature of molecular recognition, and suggest further that nonnative effects in H1-H2 packing may be largely avoided by the experimentally inferred Im9 folding transition state with native packing most developed at the H1-H4 rather than the H1-H2 interface.
| Biomolecules need to recognize one another with high specificity: promoting “native” functional intermolecular binding events while avoiding detrimental “nonnative” bound configurations; i.e., “frustration”—the tendency for nonnative interactions—has to be minimized. Folding of globular proteins entails a similar discrimination. To gain physical insight, we computed the binding affinities of helical structures of the protein Im9 in various native or nonnative configurations by atomic simulations, discovering that partial packing of the Im9 core is frustrated. This frustration is overcome when the entire core of the protein is assembled, consistent with experiment indicating no significant kinetic trapping in Im9 folding. Our systematic analysis thus reveals a subtle, contextual aspect of biomolecular recognition and provides a general approach to characterize folding frustration.
| Molecular recognition is the basis of biological function. For different parts of the same molecule or different molecules to recognize one another, a target set of interactions need to be favored while other potential interactions are disfavored. Biomolecules accomplish these simultaneous tasks via the heterogeneous interactions encoded by their sequences. For proteins, such energetic heterogeneity is enabled but also constrained by a finite alphabet of twenty amino acids. Thus the degree to which non-target interactions can be avoided through evolutionary optimization is limited [1, 2]. Conflicting favorable interactions, referred to as frustration, are often present in biological systems. From a physical standpoint, it is almost certain that some of the frustration is a manifestation of the fundamental molecular constraint on adaptation, although under certain circumstances frustration can be exploited to serve biological function [3, 4].
Protein folding entails intra-molecular recognition. Early simulations suggested that nonnative contacts can be common during folding [5]. This predicted behavior applies particularly to models embodying a simple notion of hydrophobicity as the main driving force [6, 7]. Experimentally, however, protein folding is thermodynamically cooperative [7, 8]. Folding of many single-domain proteins does not encounter much frustration from nonnative interactions in the form of kinetic traps [9]. Celebrated by the consistency principle [10] and the principle of minimal frustration [11], these empirical trends have inspired Gō-like modeling, wherein native-centric interactions are used in lieu of a physics-based transferable potential [12–14]. Extensions of this approach allow nonnative interactions to be treated as perturbations in a largely native-centric framework [15–17]. The success of these models poses a fundamental challenge to our physical understanding as to why, rather non-intuitively, natural proteins are so apt at avoiding nonnative interactions. Solvation effects must be an important part of the answer [18], as has been evident from the fact that coarse-grained protein models incorporating rudimentary desolvation barriers exhibit less frustration and higher folding cooperativity than models lacking desolvation barriers [7, 19, 20]. More recently, and most notably, folding of several small proteins has been achieved in molecular dynamics studies with explicit water [21, 22]. Nonnative contacts are not significantly populated within sections of the simulated trajectories identified as folding transition paths [23] though they do impede conformational diffusion [24]. These advances suggest that certain important aspects of protein physics are captured by current atomic force fields, although they still need to be improved to reproduce the high degrees of folding cooperativity observed experimentally [22, 25–28].
In this context, it is instructive to ascertain how atomic force fields, as they stand, disfavor nonnative interactions, so as to help decipher molecular recognition mechanisms in real proteins. We take a step toward this goal by comparing the stabilities of native and nonnative configurations of fully formed helices from a natural protein. By construction, this approach covers only a fraction of all possible nonnative configurations and therefore only provides, albeit not unimportantly, a lower bound on the full extent of frustration. Nonetheless, because of its focus on tractable systems, we obtain a wealth of reliable simulation data from which physical insights are gleaned. We do so by applying explicit-water molecular dynamics simulations to compute potentials of mean force (PMFs) between various helices [29] of the E. coli colicin immunity protein Im9 [30]. Im9 is a small single-domain protein that undergoes two-state-like folding [31, 32] to a native structure with four helices packed around a hydrophobic core [33]. Its folding mechanism and that of its homolog Im7 have been extensively characterized experimentally [30–40] and theoretically [41–46]. Of particular relevance to our study are experimental Φ-value analyses suggesting that the Im9 folding transition state has a partially formed hydrophobic core stabilized by interactions between helix 1 (H1) and helix 4 (H4), whereas helix 3 (H3) adopts its native conformation only after the rate-limiting step of folding [32]. These experimental inferences have since been rationalized by simulations showing that H1 and H4 are formed whereas about one half of helix 2 (H2) remains unstructured in the Im9 transition state [41], and that, unlike Im7, there is no significant kinetic trap along the Im9 folding pathway [45, 46].
Building on these advances, our systematic PMF analysis provides a hitherto unknown perspective on these hallmarks of Im9 folding. Notably, we found significant packing frustration between H1 and H2, viz., a nonnative packing orientation can achieve a lower free energy than that afforded by the native packing of these two helices in isolation. Superficially, this simulation result seems at odds with experiments indicating little frustration in Im9 folding. On closer examination, however, our discovery provides an unexpected rationalization for experiments indicating that folding is initiated by the more stabilizing H1-H4 interactions rather than by H1-H2 packing. Because the H1-H2 packing frustration can be circumvented by following such a kinetic order, our finding suggests that the Im9 folding pathway might have evolved to avoid a potential H1-H2 kinetic trap. This example underscores that the inner workings of molecular recognition can be rather subtle and deserves further exploration, as will be elaborated below.
With the above rationale in mind, we apply the technique described in Methods and S1 Text for extensive molecular dynamics simulations to study the 86-residue helical protein Im9 [47], focusing primarily on the interactions among various sets of fragment(s) comprising one or more helices. For terminological simplicity, each fragment set in an interacting pair—including a single helix—is referred to as a bundle below. PMFs of nine pairs of bundles (Fig 1 and Table 1) are computed to ascertain whether native or nonnative associations are preferred. Although intra-bundle conformational variations are restricted in most of our model systems (Methods), the studied configurations are all physically realizable. It follows logically that the observation of favorable nonnative packing in our simulations is sufficient to demonstrate, at least for the atomic force field used here, that favorable nonnative interactions do exist in Im9.
We begin by investigating the free energy landscape for the association of H1 with H2, a packing interaction that accounts for the largest two-helix interface in the native state of Im9, burying 5.3 nm2 or 17% of the total surface area of H1 and H2. Throughout this study, surface areas of helical bundles are computed as the solvent-accessible surface areas of the given bundles in isolation, irrespective of the solvent exposure of the configurations in the complete Im9 folded structure. Using an enhanced sampling technique known as umbrella sampling with virtual replica exchange (US-VREX, see Methods) for restrained helical configurations at systematically varied target packing angles, we compute PMFs for H1-H2 association in the absence of their intervening loop (the H1→H2 system in Fig 1B and Table 1). The PMFs are determined for the native orientation as well as for nonnative orientations and nonnative crossing angles entailed by the imposed rotational preferences (Methods and S1 Text). Our technique allows these simulations to converge rapidly (S1 Fig). Each PMF is then integrated over a free-energy basin to provide a binding free energy, ΔGbind, for a specific inter-helix geometry.
Unexpectedly, H1-H2 association is favored by a 20–30° positive rotation of H1 against H2. Binding in this nonnative orientation is 10–12 kJ/mol more stable than that in the native orientation (black circles in Fig 2A and Table 2), a free energy difference equivalent to a ~50-fold increase in bound population (S1 Text). In contrast, the binding free energy profiles for rotating H2 against H1 (Fig 2A, red squares) or changing the H1-H2 crossing angle (Fig 2A, blue triangles) indicate that the state corresponding to native packing (0° angle in Fig 2A) is situated well within the basin of lowest free energy with respect to these degrees of freedom, although a ≤50° positive change in H1-H2 crossing or a ≤20° negative rotation of H2 against H1 would leave the system approximately iso-energetic with the native packing (Fig 2A). As mentioned, these binding energies are computed from PMFs such as those in Fig 2B and S2 Fig.
A broader view of the orientation-dependent H1-H2 packing free energy landscape can be seen in Fig 2C. Instead of fixing either H1 or H2 in its native orientation (as in Fig 2A), Fig 2C provides the relative favorability of packing orientations resulting from simultaneous rotations of H1 and H2. This two-dimensional PMF is generated by combining sampling data for H1 and H2 rotations under harmonic biasing potentials (S1 Text). It is clear from this two-dimensional landscape that native packing [(H1, H2) rotations equal (0°, 0°)] is less favored than the free energy minimum at (+19°, +4°). Indeed, this minimum is situated in a rather broad basin encompassing many nonnative orientations with simultaneous H1 rotation from approximately +5° to +25° and H2 rotation from approximately ‒3° to +15° that are energetically more favorable than the native H1-H2 orientation (0°, 0°). Fig 2C reveals further that there exists another basin of favorable nonnative H1-H2 packing for which both helices rotate by approximately ‒20°. In short, our systematic analysis in Fig 2 demonstrates unequivocally that packing frustration exists in Im9, in that when H1 and H2 are considered in isolation, nonnative packing is favored over native packing.
To assess the prospect that intervening loop residues may provide additional guidance for native packing of H1 against H2, we also simulate this helix-loop-helix as a single chain (H1LH2 system; Table 1). Because the covalent connection of H1 to H2 is incompatible with the large helical separations used in our importance sampling, we study the H1LH2 system without inter-helical distance bias in simulations initiated in either the native state or one of 20 different nonnative orientations in which H1 or H2 is rotated by ±10–50°. [Because the actual rotations sampled during simulations are close to those targeted by the restraining potentials (S4 Fig), we do not distinguish between target and actual rotations hereafter]. Although these simulations do not converge to a single conformational distribution, they show broad sampling of H1 rotation with a stable or metastable state near +20° rotation of H1, even when simulation is initiated at the native packing angle (S6 Fig).
To explore how the H1-H2 packing frustration might be overcome in Im9 folding, we next investigate the impact of the rest of the protein on the packing between H1 and H2 by computing binding free energies for the association of H1 and H2 not in isolation but in the presence of additional protein fragments involving the other two helices H3 and H4 as well as loop and terminal residues. The conformations of the loop and terminal residues in our simulations are restrained to those in the Im9 PDB structure.
We first consider the association of H1 with a bundle comprising helices 2, 3, and 4 connected by their intervening loops and extending to the protein's C-terminus (H1→H2LH3LH4C; Table 1). Interestingly, for this system, native packing is found to be 13 ± 3 kJ/mol more favorable than the nonnative packing resulting from a +30° rotation of H1 (Table 2). The very fact that a nonnative rotation of H1 is substantially favored in H1→H2 (Fig 2A and Table 2) but disfavored in H1→H2LH3LH4C (Table 2) demonstrates clearly that some components of the H2LH3LH4C bundle besides H2 are crucial for overcoming the H1-H2 packing frustration and guiding H1 to pack natively. Furthermore, because native packing is favored in H1→H2LH3LH4C despite the residues N-terminal to H1 (including a short 3–10 helix) being excluded in this model system, these N-terminal residues are likely not necessary for ensuring native packing of H1 against the rest of the Im9 protein.
We now dissect the H2LH3LH4C bundle to ascertain the contributions from different parts of this bundle to native H1 packing. To this end, binding free energies for the association of H1 with a variety of subsets of H2LH3LH4C are computed. We first consider a bundle comprising helices 2 and 4 (H1→H2/H4; Table 1). Somewhat surprisingly, native packing in the H1→H2/H4 system is disfavored by as much as 22 ± 1 kJ/mol when compared against nonnative packing with H1 rotated by +30°, even more than the corresponding nonnative preference of 10 ± 1 kJ/mol for H1→H2 (Table 2). This observation implies that H4 by itself is not promoting H1-H2 native packing and therefore H3, loops, and/or the C-terminus must be responsible for driving native packing of H1 with H2LH3LH4C. Indeed, when compared against H2/H4, the presence of these other elements in H2LH3LH4C results in a 26 ± 1 kJ/mol preference for native H1 packing and a 9 ± 3 kJ/mol discrimination against nonnative H1 packing with a +30° rotation (Table 3).
To better pinpoint the role of H3 in this intra-molecular recognition process, we compute binding free energies for the association of H1 and a bundle comprising helices 2, 3 and 4 but without the intervening loops and the C-terminus (H1→H2/H3/H4; Fig 1D and Table 1). For this model system, native packing is less favorable than +30° rotation of H1 by 11 ± 3 kJ/mol (Table 2). Nonetheless, in comparison to H1→H2/H4, the inclusion of H3 favors native packing more than it favors nonnative packing with a +30° rotation of H1 (Table 2). This observation indicates that H3 is capable of correcting part of the nonnative tendencies of H1 imparted by its interactions with a bundle comprising only of H2 and H4; but H3 is insufficient to ensure native packing in the absence of the connecting loops and/or the C-terminus.
To explore whether inclusion of residues neighboring H4 may alter its effect on H1-H2 packing, we consider three residues immediately N-terminal to H4 (Asp62, Ser63, and Pro64). These residues are chosen because they are known to associate directly with H1 in the NMR structure [47] and thus they may contribute positively to native intra-molecular recognition. Consistent with this expectation, once these three residues are included, the H1-binding free energies in the resulting H1→H2/NH4 system (Table 1) for native packing and nonnative +30° rotation of H1 become essentially energetically equivalent (ΔΔGbind = 2 ± 6 kJ/mol; Table 2). Inasmuch as promoting native H1-binding is concerned, this represents a significant improvement over H1→H2/H4 that favors the +30°-rotated nonnative packing by 22 ± 1 kJ/mol (Table 2). Indeed, in the context of H1→H2/H4, addition of these N-terminal flanking residues assists native packing by 31 ± 5 kJ/mol, much more than the 7 ± 3 kJ/mol increase in stability they also impart on the nonnative packing of H1 with a +30° rotation (Table 3). These numbers underscore the important role of Asp62, Ser63, and Pro64 in discriminating against nonnative packing of H1.
Another set of helix-flanking residues that may assist native packing in Im9 is its C-terminus. Such an effect is expected because a +30° rotation of H1 would likely place its constituent residue Phe15 into a steric clash with the C-terminal residue Phe83 (S7 Fig) and thus existence of the C-terminus should discriminate against such a rotation of H1. To evaluate this hypothesis, we compute H1-binding free energies with a bundle comprising H2 and H4 as well as the protein's C-terminus (H1→H2/H4C; Table 1). Similar to the addition of Asp62, Ser63, and Pro64 N-terminal to H4 in H2/NH4 bundle, inclusion of the C-terminus in H2/H4C eliminates the strong nonnative bias in H1→H2/H4, resulting in essentially no discrimination between the native orientation and a +30° rotation of H1 (ΔΔGbind = 1 ± 3 kJ/mol; Table 2). Relative to H1→H2/H4, addition of the C-terminus not only favors native packing by 6 ± 2 kJ/mol but also directly disfavors +30° rotation of H1 by 17 ± 2 kJ/mol (Table 3). The latter penalization of nonnative packing (which does not occur in H1→H2/NH4) is consistent with the aforementioned steric consideration (S7 Fig).
Interestingly, the native-promoting effects of N- and C-terminal extensions to H4 are essentially additive. When both extensions are added to H4, the H2/NH4C system (Table 1) is sufficient to favor native packing of H1 by 14 ± 6 kJ/mol over the nonnative packing with +30° rotation of H1 (Table 2).
After analyzing systems involving H2, we now turn to the intra-molecular recognition between H1 and H4 without involving H2. Native H1-H4 packing constitutes the second largest two-helix interface in the Im9 folded structure, burying 3.7 nm2 which amounts to 13% of the sum of individual surface areas of H1 and H4. PMFs for helices 1 and 4 in isolation (H1→H4; Fig 1C and Table 1) are computed in the native orientation as well as nonnative orientations resulting from rotations of H1 or H4. When H1 is rotated while H4 is fixed, native packing is favored (Fig 3A, black circles); however, when H4 is rotated with H1 fixed, a +30° nonnative rotation of H4 leads to 5 ± 1 kJ/mol stabilization (decrease in ΔGbind) relative to native (red squares in Fig 3A and Table 2). Distance-dependent PMFs for the native orientation and ±30° rotations of H4 are shown in Fig 3B, indicating that the favored nonnative packing at +30° is attained at an H1-H4 separation slightly larger than native by about 0.1 nm. The two-dimensional PMF (Fig 3C) as a function of H1 and H4 rotation angles shows further that native H1-H4 packing (0°, 0°) is situated at the periphery of a broad basin of favored orientations centered roughly around (+10°, +10°). The same two-dimensional landscape suggests that H1 rotations of ≥ +50° or ≤ ‒50° can also be favored with little or no H4 rotation.
We noted earlier that a 3-residue N-terminal extension to H4 directly contacts H1 in the native state and that the inclusion of these residues assisted the native packing of H1 against a bundle comprising helices H2 and H4. Consistent with that observation, these three residues—Asp62, Ser63, and Pro64—likewise assist the native packing of H1 against H4, viz., their inclusion in the H1→NH4 system (Table 1) makes native packing (ΔGbind = ‒44 ± 1 kJ/mol) significantly more favorable than the nonnative packing with a +30° rotation of H4 (ΔGbind = ‒21 ± 2 kJ/mol) while still favoring native orientation of H1 (Table 2). We conclude from these results that helices H1 and H4 are nearly capable of associating in native-like conformations by themselves in isolation; and that they can certainly achieve native packing with the assistance from the 3-residue N-terminal extension to H4. These results suggest that Im9 residues 12–23 and 62–78 may serve as major components of a native-like folding nucleus.
To better understand the driving force for nonnative H1-H2 packing, the potential energies between specific pairs of amino acid residues on the H1-H2 interface in the native orientation are compared against those in the nonnative orientation with a +30° H1 rotation. We make this comparison for helix-helix center of mass distance di0 = 1.10 nm in both the native and non-native configurations, wherefore each pair of helices in question is in close spatial contact (Fig 4). The analysis indicates a prominent role by the more favorable Lennard-Jones interactions between interfacial residue pairs Glu14-Met43, Leu18-Phe40, and Ile22-Phe40 in favoring the nonnative packing, whereas electrostatic interactions between these residue pairs are of similar strengths for the native and nonnative packing orientations. In contrast, the interaction between Ile22 and Leu33 favors native packing, but its effect is more than compensated by the aforementioned multiple residue-residue interactions that drive nonnative packing such that a +30° rotation of H1 is favored over the native orientation for H1-H2 packing in isolation. It is noteworthy, however, that while these residue-residue energetic effects can be significant individually (Fig 4) and collectively (Table 2), they are not accompanied by obvious, drastic structural changes at the level of residue-residue contacts. When contacts between residues on different helices at a helix-packing interface are identified by a commonly used proximity threshold, contact probabilities between the helices are seen to remain essentially unchanged upon a +30° H1 native-to-nonnative rotation in both the H1→H2 and H1→ H2LH3LH4C systems (S8 Fig).
Seeking physical reasons for favoring native packing in H1→ H2LH3LH4C but not in H1→H2, we compare the potential energies of these systems in the native and the +30° H1-rotated nonnative configurations (Fig 5). When potential energies are analyzed by the molecular species involved in the interactions, for H1→H2, solvent-protein (solvent-helix) interactions are more unfavorable with nonnative rotation of H1 by +30°, but this effect is overwhelmed by larger, favorable changes in solvent-solvent and intra- and inter-helix interactions (Fig 5A). More specifically, this nonnative H1 rotation favors inter-helix Lennard-Jones interactions (as exemplified by the three residue pairs circled in red in Fig 4A) as well as intra-helix and solvent-solvent electrostatic interactions (Fig 5A), netting an overall favorable (more negative) potential energy for the nonnative orientation (Fig 5A, “sum”). In contrast, the corresponding analysis for H1→H2LH3LH4C yields a set of average potential energies that favors the native state overall (Fig 5B, native “sum” more negative than nonnative). This potential energy (enthalpic) trend is consistent with the above PMF/binding free energy prediction that the native orientation is favored for H1→H2LH3LH4C (Table 2), though entropic effects may make additional contribution to the stability of native packing of H1 against H2LH3LH4C (see below). Because nonnative +30° H1 rotation has opposite effects on intra-H2 (Fig 5A) versus intra-H2LH3LH4C (Fig 5B) Coulomb energies, one of the reasons for disfavoring nonnative +30° H1 rotation in H1→H2LH3LH4C is that this rotation of H1 induces energetic strain within H2LH3LH4C, resulting in a destabilizing increase in intra-H2LH3LH4C Coulomb energy collectively, whereas the same +30° H1 rotation leads to an overall stabilizing decrease in intra-H2 Coulomb energy. The atomic basis of this difference remains to be analyzed.
To gain further insight into the differential effects of H2 and H2LH3LH4C on the favorability of the native orientation upon H1 binding, we resolve the distance-dependent H1→H2 and H1→H2LH3LH4C PMFs (Fig 6A and 6B, respectively) into their enthalpic (Fig 6C and 6D) and entropic (Fig 6E and 6F) components. Since the backbones of the helical elements in our simulation systems are restrained to be essentially rigid, the entropic contributions computed here originate almost exclusively from the water solvent and sidechain degrees of freedom, whereas contributions from mainchain conformational entropy are negligible in comparison. Despite sampling uncertainties, several likely trends can be quite clearly discerned: For H1→H2, the lower PMF (ΔG) minimum for the nonnative orientation (Fig 6A) is driven by enthalpy (lower ΔH minimum for +30° H1 rotation than for native in Fig 6C). This effect is partially, but not completely, compensated by the entropic component of the free energy, ‒TΔG. The latter is seen favoring native packing in Fig 6E (red curve below blue curve at distance marked by vertical blue dashed line), although the differences are largely within error bars. Entropy has a similar effect on H1→H2LH3LH4C in stabilizing native packing (Fig 6F). In this case however, unlike H1→H2, enthalpy is also favorable (though only slightly) to the native state (Fig 6D, see also Fig 5B), thus the entropic and enthalpic effects reinforce each other, yielding a ΔG favorable to native packing for H1→H2LH3LH4C (Fig 6B). It should be noted that the trends of entropic stabilization seen here in Fig 6 are similar to those exhibited by a pair of poly-alanine or poly-leucine helices [29]. In both cases, the entropic trends are likely manifestations of the well-recognized solvent-entropic origin of hydrophobic interactions at ambient temperatures.
Every helix-helix association in Fig 6 entails an enthalpic barrier at separation ≈1.5 nm (Fig 6C and 6D). As implied by the absence of PMF barriers at these positions (Fig 6A and 6B), the enthalpic barriers here are compensated by a larger decrease in entropic free energy at the same positions (Fig 6E and 6F). Further examples of enthalpic barriers and entropic compensations are provided in S2 Fig. These results are consistent with burial of hydrophobic surfaces being concomitant with increase in solvent (water) entropy at room temperature and the idea that enthalpic barriers to protein folding [20, 29, 49, 50] may arise largely from steric dewetting [29]. Because steric dewetting creates voids (between the approaching helices in the present cases; S9 Fig), it leads to volume barriers [29] such as those seen in Fig 6G and 6H. As has been discussed, such volume barriers probably amount to part of the activation volume of protein folding [51, 52]. For the systems studied in Fig 6, it is not surprising that the enthalpic and volume barriers are higher for H1→H2LH3LH4C than for H1→H2 because the former binding process buries a significantly larger protein surface area. Therefore, we expect a larger transient void volume between the approaching helices before close packing is achieved for H1→H2LH3LH4C than for H1→H2. It is interesting to note that, perhaps because void volumes are largely a consequence of geometry and less of energetics, the volume barrier heights in Fig 6G and 6H are essentially insensitive to the difference between native and nonnative packing.
To recapitulate, we have conducted a systematic analysis of the relative stability of native versus nonnative packing of helices in the Im9 protein as a means to address the physical basis of biomolecular recognition. These results are summarized schematically in Fig 7: Relative to native packing, three nonnative configurations (H1→H2, H1→H2/H4, and H1→H2/H3/H4, each with H1 rotated) are significantly stabilized whereas one other nonnative packing orientation (H1→H4 with H4 rotated) is mildly stabilized. Other Im9 systems that we have simulated either favor the native configuration or essentially do not discriminate between native and nonnative packing. As emphasized at the outset, our method is designed to characterize packing frustration of constrained, locally native protein substructures by varying the orientation between interacting substructures that are rigid by construction, viz., the secondary structure (main-chain conformation) of each of the helices is essentially fixed. It follows that while our substantial computational effort has succeeded in gaining structurally and energetically high-resolution information about frustration that is novel and complementary to that obtained from our previous coarse-grained chain model study of Im7 and Im9 [45], the present investigation—unlike our coarse-grained modeling [45]—cannot by itself address certain general questions regarding folding pathways such as the viability of nucleation-condensation mechanisms [53] because backbone conformational freedom is not treated. For the same reason, the present method does not tackle frustration involving disordered, flexible main-chain segments that may adopt locally nonnative conformations. A notable example in this regard is the second helix of Im7. Among the four respective helices in Im7 and Im9, the amino acid sequence of the second helix varies the most between the homologs [30]. The second Im7 helix has been identified as a part of the protein which is disordered and participates in nonnative interactions that stabilize a kinetically trapped folding intermediate during the process of non-two-state folding of Im7 [45]. However, revealingly, the significant role of a disordered H2 in frustrating Im7 folding is not reflected by its behavior as an ordered helix: Unlike the H1→H2 system of Im9, the H1→H2 system of Im7 exhibits no favorable nonnative packing (S10 Fig). This finding underscores the importance of disordered conformations to frustration in globular protein folding, an effect that the present analysis has not addressed. From a broader perspective, such effects have to be even more critical for molecular recognition among intrinsically disordered proteins [54, 55].
Notwithstanding aforementioned limitations of the present approach, several important lessons can already be learned from our extensive computational investigation. First, a majority of the helical systems that we consider favor native packing, indicating that the Im9 amino acid sequence encodes a sufficiently strong native bias such that the native structure can be recognized by the folding protein. Second, frustration exists, manifested most notably by—but not necessarily limited to—the significantly stabilized nonnative H1-H2 packing. Although the conformational space accessible to an 86-residue polypeptide is vast compared to what is accessible via contemporary simulation and thus our ability to identify all possible sources of frustration is limited, the systematic approach taken in the present study does pinpoint one class of frustrated configurations. Third, the native fold is favored overall despite frustration, at least within the class of configurations we tested, because nonnative H1-H2 packing is destabilized when other parts of the protein, especially H4 and its flanking residues, are involved in the interaction.
A logical inference from our results is that favorable nonnative interactions can be largely suppressed during Im9 folding by favoring trajectories that assemble H1 and H2 not in isolation but only in the presence of H4 plus flanking residues. Such preference would help avoid kinetic traps to facilitate known two-state folding behaviors of Im9 [32, 36]. This expectation is consistent with the Im9 folding mechanism deduced from experimental phi-values (ΦF) by Radford and coworkers, who determined that residues in H2 have the lowest ΦF-values among H1, H2, and H4; but ΦF-values are higher for the hydrophobic residues in H1 and H4. This and other findings led them to conclude that the H1-H4 interface “is the most structured region in the transition state ensemble”, and that the native configuration of H1, H2, and H4 is partially formed in the transition state whereas H3 is formed after the rate-limiting step [32]. Since our simulation results also suggest that H1-H2 interactions should be weaker than those between H1-H4 to minimize kinetic trapping, our data offer a physical rationale as to why the Im9 folding pathways might have evolved.
A general theoretical formalism due to Wolynes and coworkers provides quantitative estimates of local frustration [3, 42, 56]. Of relevance here is their configurational frustration index, which quantifies the likelihood of a pair of residues that are in contact in a protein’s native structure to be engaged in favorable nonnative interactions in alternate conformations. Their web-based “Protein Frustratometer” algorithm [56] predicts a high configurational frustration region in Im9 encompassing residues 25–38, which overlaps substantially with H2 (residues 30–44, Table 1). In contrast, H1 and H4 are predicted to be situated in lower configurational frustration regions on average (S11 Fig). These predictions are consistent with, and therefore lend further support to the aforementioned perspective emerging from our simulation results. It is noteworthy, however, that the Frustratometer-computed configurational frustration Fc of Im7 is not noticeably higher on average than that of Im9 (S11 Fig), notwithstanding the fact that folding is significantly more frustrated for Im7 than for Im9 experimentally [30–40]. In particular, while the predicted frustration of H4 is higher for Im7 than for Im9 (which is consistent with H4’s involvement in nonnative interactions with H2 in Im7 folding), the predicted configurational frustration of H2 of Im7 is similar to, or even slightly lower than that of Im9. It would be instructive to investigate whether this apparent inability of the algorithm to clearly delineate the key experimental difference in Im7 and Im9 folding kinetics is because the decoy inter-residue contact distances used to compute configurational frustration Fc [56] are insufficient to fully capture the conformational possibilities of a disordered H2 that make strong nonnative interactions in Im7 possible [45]. Intuitively, this limitation might be similar or even related to the impossibility of discerning Im7 frustration from the packing of fully formed H1 and H2 alone (S10 Fig) despite the fact that many of the favorable nonnative interactions in Im7 folding are between residues in H1 and H2. This question deserves further attention.
Owing to the high computational cost of the present approach, applications have been confined to the commonly used OPLS-AA/L force field. While useful insights are gained as reported above, it should be noted that current molecular dynamics force fields can be limited in their ability to accurately model disordered protein states (reviewed in [57, 58]) and to capture subtle effects such as conformational switches [59]. It is important, and would be instructive, to assess how discrimination against nonnative interactions is affected by ongoing efforts to improve current force fields [57, 58]. Much work remains to be done before the physical basis of biomolecular recognition can be fully deciphered.
We use molecular dynamics (MD) simulations to systematically study helix packing in Im9 (PDB ID: 1IMQ [47]) by constructing pairs of various Im9 fragments (bundles) comprising one or more helices (Fig 1 and Table 1) and computing their PMFs of association (Figs 2–4). A more limited set of Im7 bundles is also studied for comparison. Helical residues (Table 1) are defined by DSSP [48]. Helical rotations with positive and negative angles indicate clockwise and counter-clockwise angular displacements, respectively, around the helix’s long axis in the N- to C-terminal direction (Fig 1) relative to the native orientation. Positive and negative changes in helix-helix H1-H2 crossing angles are, respectively, rigid rotations of H1 in the clockwise and counter-clockwise directions with respect to the vector directed from the center of mass of H1 to the center of mass of H2, the angular changes being relative to the native H1-H2 crossing angle.
Umbrella sampling (US) [60] simulations are employed to quantify the extent to which the residues in two pre-folded regions of the protein are sufficient to drive native-like association. Specifically, we compute orientation-specific free energies for the binding of H1 to a systematic selection of helices from other parts of the protein with and without connecting loops. To enhance computational tractability, the latter helical bundles are prevented from unfolding or changing their relative orientations by imposing harmonic restraints on the positions of all Cα atoms with force constants of 1000 kJ/mol/nm2. Unfolding of H1 is disallowed by Cα position restraints that are enforced only in the Cartesian y and z dimensions, using the same force constant. The US order parameter is the magnitude of the Cartesian x component of the vector connecting the centers of mass of Cα atoms in the two bundles. This linear displacement, d, is harmonically restrained at a specified target value, di0, in each umbrella i, with a force constant of 2000 kJ/mol/nm2. For each system, 39 umbrellas span 0.7 nm ≤ di0 ≤ 2.6 nm in 0.05 nm increments. To further enhance the rate of convergence in these US simulations, we allow equilibrium exchange of umbrellas using the virtual replica exchange (VREX) approach [61, 62]. Further details of the US-VREX approach are provided in S1 Text. US-VREX simulations are conducted for the H1→H2, H1→H4, H1→NH4, H1→H2/H4, H1→H2/NH4, H1→H2/H4C, H1→H2/NH4C, H1→H2/H3/H4, and H1→H2LH3LH4C systems of Im9 (Table 1), where the arrow separates the two interacting fragments (bundles) under consideration. The H1→H2 system of Im7 is also simulated using the same method. The two bundles in any given system are on equal footing because their association with each other is mutual. The arrow in our notation serves merely to indicate their spatial association without regard to the arrow’s direction. Each system is simulated for 100 ns/umbrella, except for H1→H2 and H1→H2LH3LH4C in the native orientation and with nonnative H1 rotation by +30°, which are simulated for 500 ns/umbrella. In total, these US-VREX simulations comprise >300 μs of simulated time. Despite the application of position restraints to prevent the helices from unfolding or changing their relative orientations during PMF computations, the rotation angle of helices varies within ±10° of the target packing angle. We identify the simulated systems by the angles to which they are targeted.
The H1LH2 system comprising H1, H2, and their connecting loop is simulated from the native [47] and twenty different nonnative initial conformations generated by removing inter-helical loop residues 24–29, rotating H1 or H2 about its long axis by ±10°, ±20°, ±30°, ±40° and ±50°, and then modeling loop residues using the prediction program Loopy [63]. Secondary structure is maintained while allowing changes in helical rotation and separation by applying intra-helical distance restraints on all backbone atom pairs with force constants of 1000 kJ/mol/nm2. Each simulation covers 1 μs, with the first 125 ns discarded in subsequent analysis.
MD simulations are conducted with version 4.5.5 of the GROMACS simulation package [64]. The water model is TIP3P [65]. Protein is modeled by the OPLS-AA/L parameters [66, 67]. Simulation systems are neutralized and excess NaCl is added at 0.4 M, mimicking experimental conditions [31, 68]. Water molecules are rigidified with SETTLE [69] and protein bond-lengths are constrained with P-LINCS [70]. Lennard-Jones interactions are evaluated using a group-based cutoff and truncated at 1 nm without a smoothing function. Coulomb interactions are calculated using the smooth particle-mesh Ewald method [71, 72] with a Fourier grid spacing of 0.12 nm. Simulations are in NPT ensembles by isotropic coupling to a Berendsen barostat [73] at 1 bar with a coupling constant of 4 ps and temperature-coupling the simulation system using velocity Langevin dynamics [74] at 300 K with a coupling constant of 1 ps. The integration time step is 2 fs. The nonbonded pair-list is updated every 20 fs. Further details are provided in S1 Text and S1 Table.
|
10.1371/journal.pcbi.1005710 | Funneled potential and flux landscapes dictate the stabilities of both the states and the flow: Fission yeast cell cycle | Using fission yeast cell cycle as an example, we uncovered that the non-equilibrium network dynamics and global properties are determined by two essential features: the potential landscape and the flux landscape. These two landscapes can be quantified through the decomposition of the dynamics into the detailed balance preserving part and detailed balance breaking non-equilibrium part. While the funneled potential landscape is often crucial for the stability of the single attractor networks, we have uncovered that the funneled flux landscape is crucial for the emergence and maintenance of the stable limit cycle oscillation flow. This provides a new interpretation of the origin for the limit cycle oscillations: There are many cycles and loops existed flowing through the state space and forming the flux landscapes, each cycle with a probability flux going through the loop. The limit cycle emerges when a loop stands out and carries significantly more probability flux than other loops. We explore how robustness ratio (RR) as the gap or steepness versus averaged variations or roughness of the landscape, quantifying the degrees of the funneling of the underlying potential and flux landscapes. We state that these two landscapes complement each other with one crucial for stabilities of states on the cycle and the other crucial for the stability of the flow along the cycle. The flux is directly related to the speed of the cell cycle. This allows us to identify the key factors and structure elements of the networks in determining the stability, speed and robustness of the fission yeast cell cycle oscillations. We see that the non-equilibriumness characterized by the degree of detailed balance breaking from the energy pump quantified by the flux is the cause of the energy dissipation for initiating and sustaining the replications essential for the origin and evolution of life. Regulating the cell cycle speed is crucial for designing the prevention and curing strategy of cancer.
| We have uncovered that the non-equilibrium network dynamics and global properties are determined by two essential features: the potential landscape and the flux landscape. We have found that the funneled potential landscape is crucial for the stability of the states on the cell cycle, however, the stabilities of the oscillation states cannot guarantee the stable directional flows. We have uncovered that the funneled flux landscape is important for the emergence and maintenance of the stable limit cycle oscillation flow. This work will allow us to identify the key factors and structure elements of the networks in determining the stability, speed and robustness of the fission yeast cell cycle oscillations. We see that the non-equilibriumness characterized by the degree of detailed balance breaking from the energy pump quantified by the flux is the cause of the energy dissipation for initiating and sustaining the replications essential for the origin and evolution of life. Regulating the cell cycle speed is crucial for designing the prevention and curing strategy of cancer.
| The global stability and robustness are crucial for maintaining the function. They are also important for uncovering underlying mechanisms of the networks. [1–7] However, it is difficult to quantify them for dynamic systems and networks. This presents a challenge for the dynamical systems and the field of systems biology. [8–23]
In equilibrium systems, the global nature of the system is characterized by the underlying equilibrium potential landscape U which is directly linked to the equilibrium probability through the Boltzmann distribution law P ∼ exp(−βU). The local dynamics is determined by the gradient of the equilibrium potential landscape. However, most dynamical systems do not typically have a gradient potential as in the equilibrium case. They are open systems usually not in isolations. Global natures of such systems are hard to address. In addition, for mesoscopic systems, the intrinsic fluctuations can also be significant. Under stochastic fluctuations, instead of following the dynamical trajectories which are stochastic and unpredictable, the evolution of the probabilistic distributions should be followed, which is inherently global as well as predictable due to its intrinsic linearity. The probabilistic evolution is governed by the master equations for discrete state space (more general) and Fokker-Planck equations for continuous state space.
It turns out the steady state distribution of the probability evolution in long time limit can give a global quantification of the dynamical systems [5–23]. This defines a probability or weight landscape for characterizing the system states. On the other hand, the dynamics of the systems can be decomposed to gradient of the potential landscape related to the steady state probability distribution and a curl probability flux. The existence of a non-zero curl flux directly reflects the degree of the breakdown of the detailed balance. This quantifies the degree of the non-equilibrium. While this decomposition is shown explicitly in continuous space through Fokker-Planck equation description of the stochastic dynamics [10–23], the corresponding decomposition and associated statistics of stochastic dynamics in discrete space from the master equation still needs further explorations [22–31].
In this work, we study the more general stochastic dynamics in discrete space of the non-equilibrium networks (Markov chains) governed by the probabilistic master equation. We found the network dynamics and global properties are determined by two features: the potential landscape and the probability flux landscape. While potential landscape quantifies the probabilities of different states forming hills and valleys, the probability flux landscape is composed of many flux loops flowing in state space. Therefore, statistics of the flux loops becomes important. These two landscapes can be quantitatively constructed through the decomposition of the dynamics into the detailed balance part and non-detailed balance part.
We found that while funneled landscape is crucial for the stability of the single attractor networks and stability of oscillation states. The funneled flux landscape is crucial for maintaining the stable limit cycle oscillation flow. The stability and the robustness of the networks can be quantified through a dimensionless ratio of the gap or steepness versus the averaged variations or roughness of the landscape (which measures the degree of funnel, we termed as robustness ratio RR), and explored under the changes of the network topologies and stochastic fluctuations.
This flux landscape picture provides a new interpretation of the origin of the limit cycle oscillations. The global oscillation only emerges when one specific loop stands out and carries much more probability flux, and therefore becomes more probable than the rest of the others.
We specifically studied the fission yeast cell cycle as an example to illustrate the idea. We found the flux landscape of the fission yeast cell cycle oscillations is funneled, which guarantees its stability and the robustness of the oscillation flows. The global stability is quantified by the robustness ratio RR of the funneled flux landscape and the robustness is quantified by RR against the changes in topology of the network (wirings) and stochastic fluctuations. The flux is directly related to the speed of the cell cycle. The landscape analysis here allows us to identify the key factors and structure elements of the networks in determining the global stability, speed, and robustness of the fission yeast cell cycle oscillations. We see that the non-equilibriumness characterized by the degree of detailed balance breaking from the energy pump quantified by the flux is the cause of the energy dissipation for initiating and sustaining the replications essential for the origin and evolution of life. The cell cycle speed is a hallmark of cancer. Regulating the cell cycle speed thus provides a possible strategy for preventing and curing strategy against cancer.
We follow a boolean network model mainly built for fission yeast in [32]. As illustrated in the Table 1, the cell cycle of fission yeast is divided into several phases: START phase → G1 phase → S phase → G2 phase → M phase → G1 phase. The network wiring diagram shown in Fig 1 consists of one check point and 9 gene nodes, that is CS (Cell Size), SK, Cdc2/Cdc13, Ste9, Rum1, Slp1, Cdc2/Cdc13*, Wee1/Mik1, Cdc25, and PP. The check point of this cell cycle network is named as Cell Size (CS) as it mainly works as an indicator of mass of the cell [32, 33]. In the global state space, there are 210 states. Each state is the combination of the “on” (si = 1) and “off” (si = 0) states, which can be represented by a state vector S = {s1, s2, s3, …, s10}. With this representation, it can form a state space of a complex gene regulation interaction networks [34]. In general, one can use the boolean dynamics model to explore the coarse grained dynamics with the information of the wiring topology of the networks [33], and one can find that there are robust biological pathway and a global robust G1 state inside the state space.
However, from the view of cell cycle, the stationary state G1 which converges the biological pathway is a temporal fixed point. To perform the function of cell cycle, one needs a positive feedback to push the G1 state to pass through the checkpoint of cell size to activate the node SK, and then go through the dynamic oscillatory trajectory, and finally get back to the G1 state. Therefore we add a kinetic excitation once the global G1 state is reached. For biological meaning, if there is no nutrition supply constantly pumping into the system, the cell cycle will stay in the stationary G1 state (so-called G0 state). While given enough nutrition supply, the cell staying at G1 gains a large transition probability to reach the cell size checkpoint and then activate the node SK to drive the cycle. Such a pumping or driving force has been explicitly discussed in a model system for limit cycle oscillation as the chemical energy supply [35]. There, the chemical energy supply, which may be generated from ATP or GTP, acted as a chemical pump or battery for initiating and maintaining the cycle. In dynamics, the chemical energy supply is in terms of flux driving the cycle flow.
This cycling conception can be emerged well in a stochastic Boolean network model. We have developed an approach to probabilistically describe the network dynamics as follow:
Firstly, we calculate transition probabilities T(si(t′)|S(t)) of gene node i jumping from state si(t) to state si(t′), where t and t′ are two close neighborhood time moments. Based on the Markovian process theory, all the cases of transition probability T(si(t′)|S(t)) are shown in the black box in Table 2. [24, 28, 36–40].
In Table 2, I = ∑ j = 1 10 a i j s j ( t ) is defined as the total input of the ith gene node at time t, which is the summation of each of the interaction strength aij from jth gene node to ith gene node.
Just considering the simplifications of the interactions between two nodes (activated (+1) or repressed (-1)), one can obtain that: when the total input I > 0, gene node i has high probability to stay at state si(t′) = 1; when the total input I < 0, gene node i tends to be repressed at state si(t′) = 0. It is similar with the behavior of switching function, that is why we define the transition probability as T(si(t′)|S(t)) = 1/2 ± 1/2tanh(μI). In this way, we can clearly see that the transition probability is mainly determined by the input. While in the case of I = 0, the ith gene node mainly stays at present state, just with a little probability of c to change the state due to the background production and self-degradation. There is only one exception for the case I = 0, that is when the cell cycle stays at the G1/G0 state, where we add a cycling activation strength of γ to activate the check point of CS, which corresponds to the condition with enough chemical supply as mentioned above.
The biological meaning of the three parameters μ, c and γ in Table 2 can be illustrated as follows: μ can be considered as a mean transition strength from the input to output of a gene or protein node, which is also related to the inverse of the fluctuation or noise strength.
c is a parameter to quantify the effect of perturbation from the background production or degradation with not input (I = 0). For example, gene node i without any input interactions has a probability of c to change the present state, with 0 → 1 due to the background production, while 1 → 0 due to the background self-degradation (direct degradation and growth dilution).
γ represents the probability of the positive feedback or stimulation from the stationary G1 state to the check point of cell size, that is from G1 phase to START phase. The cell cycle stays at G1/G0 at present implies the fact that there is no input interaction (I = 0) for all the node (so-called global stable steady state). When the nutrition supply is enough, the chemical energy pump will push the stationary G1/G0 state to activate the check point node CS (i = 1) and reach the state of START phase, with a probability of γ (with a probability of (1 − γ) to stay at G0).
Secondly, we figure out that the transition probability between two neighbor states in the Markov time series chain can be written as the product of each node transition [23, 27]
T ( S ( t ′ ) = { s 1 ( t ′ ) , s 2 ( t ′ ) , . . . , s 10 ( t ′ ) } | S ( t ) ) = ∏ i = 1 10 T ( s i ( t ′ ) | S ( t ) ) , (1)
Finally, we obtain the evolution equation to guide the probabilistic dynamics, which is so called master equation [23, 41, 42]:
d P i d t = - ∑ j T i j P i + ∑ j T j i P j , (2)
where Pi represents the probability of state i, and the transition probability Tij represents the transition probability from state i to state j. Here, we use Tij as a discrete transition probability to fit the master equation, whose meaning is equal to the continuous transition probability expression of T(S(t′)|S(t)). The physical meaning of the master equation is the conservation law of probability: the local change of the probability of a particular state i in time is equal to the probability flow (flux) from the other states to this state i given by ∑j Tji Pj subtracting the probability flow (flux) from the state i to other states ∑j Tij Pi. By solving the 210 = 1024 master equations numerically, we obtain both the time-dependent evolution and the steady-state probability of each state in the global state space.
For steady state, we set the left term of the master eq (2) to zero, that is d P i d t = 0, then we obtain the numerical steady state solution P i s s, which is the long time limit. Given the steady solution, we can define the steady state flux between state i and j as: F i j s s = - T i j P i s s + T j i P j s s, If for any i, j pair, Fijss = 0, this Markov chain is detail-balance preserved, and the steady state of the system becomes the equilibrium state (without net local flux), since dPi/dt = ∑j Fijss = 0. However, in general the steady state probability can be obtained, but it does not have to satisfy the detailed balance condition(Fijss ≠ 0). In other words, the net flux does not have to be zero. The system is then in non-equilibrium steady state. Although the steady-state distribution is fixed and does not change in time, there can be an internal probability flow among states.
In order to study the non-equilibrium steady states and characterize the global properties, one can separate the dynamical process into two parts, a detailed balance part and a pure irreversible non-detailed balance flux part by decomposing the transition probability matrix M [29]. The master equation can be rewritten as d P / d t = M T P, where P is the vector of probability of all the discrete states, M is the transition probability matrix (or rate matrix) with Mij = Tij, i ≠ j and Mii = (−1)∑j Tij. We define a matrix C such that the ith row and jth column of it is given as C i j = m a x { T i j P i s s - T j i P j s s , 0 } / P i s s , i ≠ j and Cii = (−1)∑j Cij, and matrix D whose ith row and jth column is given as D i j = m i n { T i j P i s s , T j i P j s s } / P i s s , i ≠ j and Dii = (−1)∑j Dij. It follows that M = C + D and D T P s s = 0. Since M T P s s = ( C + D ) T P s s = 0, C T P s s = 0. By separating the transition probability matrix this way, two Markov processes are obtained [29]. The probability transition matrix (or rate matrix) M for characterizing the dynamics can be decomposed into two terms: C and D. Both C and D have the same steady-state(stationary) probability distribution, and one of the processes D satisfies detailed balance(Dij Piss = Dji Pjss), while the other C is non-detailed balanced and irreversible (if Cij Piss > 0, Cji Pjss = 0). In this way, the dynamics is decomposed to detailed balance preserving part and detailed balanced breaking part.
The non-equilibrium irreversible part can be termed as the circulation or flux part, since it can be further decomposed into flux circles or loops with a flux value on each cycle [29]. The prove of the circulation also provides a way to obtain all the circles and their corresponding flux values for the dynamic part. By definition of flux, we have F i j = - C i j P i s s + C j i P j s s = - F j i. Now define J i j = C i j P i s s , i ≠ j ; J i i = 0, we have ∑i Jij = ∑i Jji. Since Jii = 0, suppose J k 0 k 1 > 0, from the summation equation just mentioned, we can find a k2 ≠ k0, k1 such that J k 1 k 2 > 0. We can keep on doing this, until a repeat is found: kn ∈ {k0, k1, …, kn−2}. Suppose k n = k n 0, let i 1 = k n 0 , i 2 = k n 0 + 1 , … , i n 1 = k n − 1 and i n 1 + 1 = i 1, we now construct a cycle or closed loop with i 1 , … , i n 1. Let r 1 = m i n k = 1 , 2 , … , n 1 { J i k i k + 1 }, define r1 as the flux value of this cycle. Then subtract their flux value from the matrix J, thus
J i j ( 1 ) = { J i j - r 1 , i ∈ { i 1 , . . . , i n 1 } , J i j , o t h e r w i s e .
If J(1) ≠ 0, repeat what we did above to find another cycle as well as its flux value, then subtract those fluxes from J(1) to get J(2). Since the number of non-zero elements in J(i) is at least one less than that in J(i−1), there exits an integer N such that J(N+1) = 0 (all the elements of the matrix J are zero). Therefore for the non-detailed balanced part of the dynamics, the flux can be decomposed to finite number of circles or closed loops, each with a flux value, J = ∑ i = 1 i = N J ( i ) [29]. Therefore, the decomposition and associated flux statistics can be directly carried out at the master equation level.
On the other hand, one can also find another way for decomposition and associated statistics of the fluxes from the stochastic boolean trajectories. One can follow the trajectory from one state S(t) at moment t to another state S(t′) at the next moment t′. If one finds the same state at different moment, that is S(t″) = S(t), then all the states between these two same states can be defined as one loop. Then one should remove this loop from the trajectory and repeat the steps above to obtain all the loops. By making statistical analysis on all the flux loops, one can calculate the flux landscape using the formula below, Uflux = −lnPflux, where P f l u x = F l u x l o o p = l i m N → ∞ N J ( i ) N J. [26–29, 43]
Therefore we have two quantitative features to characterize the system, one is the steady state probability and the other is the non-zero steady state probability flux which can be further decomposed into loops. The steady state probability obeys the evolution equation of the transition probability matrix (or rate matrix) at long time limit characterizing only the detailed balance part. The detailed balance condition allows one to identify the path independent probability measures [30]. This naturally leads to the potential. We can see how both potential and flux landscape influence the dynamics and stability of the system through an example on fission yeast cell cycle.
As known, when an open system is under long time evolution, it can reach non-equilibrium steady state (NESS) [6, 7, 19–23, 31, 44]. The local steady state flux F i j s s = - T i j P i s s + T j i P j s s is not necessarily equal to zero (no detailed balance). In this condition we can define a generalized force referring to the generalized chemical potential (from j to i) A j i = l n ( T j i P j T i j P i) [8, 10, 19–23, 31, 44]. There is a mapping between the cellular networks and electric circuits. The flux Fij corresponds to current I and chemical potential Aij corresponds to voltage V. The non-equilibrium cell network dissipates energy just as the electric circuits.
In the steady state, the heat loss rate is related to the entropy production rate. The entropy production or dissipation characterizes “time irreversibility” and provides a lower bound for the actual heat loss in Boolean network [8, 10, 19–23, 44]. The total entropy change is equal to the part from the system or source plus the part from the bath or sink (dissipation). Since in steady state the entropy change of the system is equal to zero, thus the total entropy change (source) is equal to the entropy change of the sink (dissipation). The total entropy change (source) = ∑Fij Aij is the entropy production and the sink term is dissipation. Therefore in steady state, knowing the entropy production, we know the dissipation quantitatively. The entropy S from the system part is defined as S = −∑i PilnPi and entropy production rate d S t o t d t is given by: d S t o t d t = ∑ F j i A j i = ∑ i j T j i P j l n ( T j i P j T i j P i ).
Entropy production is correlated with flux. When the steady state flux is zero, the entropy production or dissipation at steady state is zero. When the flux increases, the entropy production typically increases. Therefore, the entropy production or dissipation can also serve as a quantitative measure of how far away the system is from the equilibrium, or in other words, the degree of the detailed balance breaking.
We have uncovered that both the potential landscape and the flux landscape are crucial for the stability and the robustness of the limit cycle oscillation of the fission yeast cell cycle. As a practical application, we will perform global sensitivity analysis based on the two landscapes, to explore a backbone subnetwork to carry out the biological functions. We firstly perform perturbations through adding, deleting or repressing arrows between nodes in the wiring diagram in Fig 1, or replacing an active arrow with an inactive arrow, or deleting an individual node. And then we try to analyze the variation of the important characteristic of the two landscapes, such as RR, PG1, PCircle, RR of flux and so on. Finally, we try to work out which key links or nodes are responsible for the stability, speed (function), and robustness of the cell cycle.
Through the global stability analysis for the key wirings of the networks upon perturbations of links and nodes, one can identify the key network structure elements or motifs responsible for the stability and biological function. To further identify the stable and functional backbone subnetwork, we choose to delete the link one by one and find out which will make the network more unstable. Based on the discussion in the text above, we select three essential elements to measure the importance of each network edge, that is RR, PG1, and RRflux (RR of flux landscape), which can represent not only the global stability of G1 state but also the robustness of the biological path.
To measure the robustness of each edge, we performed the orders as follow: First of all, for each edge, we respectively calculate the difference of the three essential elements between mutated network and original network. Then we rank the difference of the 27 edges and give them score from 1 to 27. The larger is the difference, the larger the score is. Then we rank the summation of all the three scores as a total evaluation score(TES), and obtain the rank of robustness of all the 27 edges of the fission yeast network. (One can see the result in Table 3)
Based on the edge robustness, we attempt to reconstruct a minimal but most robust or stable backbone network of the fission yeast. However, nature does not necessarily use the robust edges to build the network, as the aim of a network is to perform certain biological function. Therefore, we need to reconstruct a biological meaningful network based on the major functional biological path for cell cycle. There are several strategies to do so.
In an approach [46], the backbone network is obtained directly from the criterion of emergence of major biological path for cell cycle. However, in our approach, we chose the backbone subnetwork based on the global sensitivity and robustness from potential and flux landscape for quantitatively guarantee a stable biological path. We have labeled each edge of this subnetwork both in the Table 3 with italic type and the original network wiring in Fig 18 with red color.
Notice that our backbone subnetwork has large overlaps with the one obtained by Zeng et al. [46] through biological path requirement. This shows that the backbone network chosen from biological path is also likely the stable one as we have quantitatively shown here for this case of fission yeast cell cycle.
This result seems not so intuitive. As the backbone subnetwork directs to the minimal network to perform the biological function, while the robust network means that it has strong capability to persist in the mutation or fluctuation environments. Therefore, the minimal network should not always lead to the most robustness one.
In our study, we explore parameters listed in the Table 3 of both potential and flux landscapes. The three elements which contribute the TES are the key features representing the potential and flux landscape topography. Therefore, the rank of the TES is calculated in a quasi-quantitative way to gain insights on both potential stability and period persistence for each edge. It suggests that due to the consideration of flux landscape in the periodical dynamics, the minimal backbone subnetwork with highest rank of TES tends to be a global robust one to perform the cell cycle function. Therefore, we state that this study gives a physical principle and basis in terms of the potential and flux landscape for the backbone finding.
Furthermore, based on the global sensitivity analysis, we can identify key links that change significantly the occupation probabilities of the states on the cell cycle path and robustness ratio separating the dominant cycle loop from the rest compared to the wild type. These links and nodes are responsible for biological function of the cell cycle.
As we have stated, for this case of oscillation network, the flux landscape have a large impact on the dynamical behavior of the network. In Table 3, if we delete the index of flux landscape, i.e. RRflux, there are several edges in the original network acutely changing their orders of importance. The edges such as SK to Rum1, and Cdc25 to Cdc2/Cdc13* will lose their high ranks due to the missing of flux index, while the remaining edges such as Cdc2/Cdc13 to Rum1 and to Ste9 and others tend to squeeze into the subnetwork. This leads to the conclusion that the former edges tend to perform the biological cycle function while the remaining edges tend to keep the robustness of G1 state. It is in this perspective these results provide a strong support for the potential and flux landscape theory in the study of cell cycle.
Cell cycle is a hallmark of cancer. Cancer cells has a much faster speed of cell cycle than normal cells. Therefore, regulating the cell cycle speed is crucial for preventing and curing the cancer. From above global sensitivity analysis, we can identify the key nodes and links in the fission yeast cell cycle network for regulating the cell cycle speed. These key links and nodes form the backbone network of the cell cycle. Therefore, we can based on this to do network design and network medicine discovery targeting the cancer.
We explore the global natures of the networks. We found the network dynamics and global properties are determined by two essential features: the potential landscape and the flux landscape. While potential landscape quantifies the probabilities of different states forming hills and valleys, the flux landscape quantifies the probability fluxes of different loops flowing through states. These two landscapes can be quantified through the decomposition of the dynamics into the detailed balance preserving part and detailed balance breaking part. While funneled landscape is crucial for the stability of the single attractor networks, the argument can be extended to the stabilities of the states on the oscillation paths by including them in the same (line) basin of attraction. Importantly, we have uncovered that the funneled flux landscape is crucial for the stable and robust oscillation flow.
This provides a new interpretation of the origin of the limit cycle oscillations: There are always many cycles and loops forming the flux landscapes, each with a probability flux going through the loop. The oscillation only emerges when one specific loop stands out and carries much more probability flux than the rest of the others.
We studied the fission yeast cell cycle as an example to illustrate the idea. We found both the potential landscape and the flux landscape of the fission yeast cell cycle oscillations are funneled, which guarantees the global stability. While the funneled potential landscape guarantees the stabilities of the states on the oscillating path, the funneled flux landscape guarantees the directional flow of the oscillations which breaks the detailed balance and time reversal symmetry, leading to the stand out of the dominant flux loop against others. The stability and robustness of the oscillations are quantified through a dimensionless ratio of the steepness or gap versus the averaged variations or roughness of the landscape (measuring funnelness as we termed as robustness ratio RR).
We explore how RR changes with respect to the stimulations, self degradations, state switching rate or fluctuations, and changes in topology of the network (wirings). This allows us to identify the key factors and structure elements of the networks in determining the stability, speed and robustness of the fission yeast cell cycle oscillations.
Based on the global sensitivity analysis, we obtain that our most robust subnetwork is nearly the same as the minimal biological functional network, and by setting the cell cycle period as the evolution goal, we suggest the fission yeast should follow this evolution goal to form a 27-link network with faster period but not using minimal backbone network. We see that the non-equilibriumness characterized by the degree of detailed balance breaking from the energy pump quantified by the flux is the cause of the energy dissipation for initiating and sustaining the replications essential for the origin and evolution of life. Finally we are looking forward to the good future by controlling the speed of the cell cycle as an important in designing targeting drugs for preventing and curing the cancer.
|
10.1371/journal.pgen.1003351 | Spreading of a Prion Domain from Cell-to-Cell by Vesicular Transport in Caenorhabditis elegans | Prion proteins can adopt self-propagating alternative conformations that account for the infectious nature of transmissible spongiform encephalopathies (TSEs) and the epigenetic inheritance of certain traits in yeast. Recent evidence suggests a similar propagation of misfolded proteins in the spreading of pathology of neurodegenerative diseases including Alzheimer's or Parkinson's disease. Currently there is only a limited number of animal model systems available to study the mechanisms that underlie the cell-to-cell transmission of aggregation-prone proteins. Here, we have established a new metazoan model in Caenorhabditis elegans expressing the prion domain NM of the cytosolic yeast prion protein Sup35, in which aggregation and toxicity are dependent upon the length of oligopeptide repeats in the glutamine/asparagine (Q/N)-rich N-terminus. NM forms multiple classes of highly toxic aggregate species and co-localizes to autophagy-related vesicles that transport the prion domain from the site of expression to adjacent tissues. This is associated with a profound cell autonomous and cell non-autonomous disruption of mitochondrial integrity, embryonic and larval arrest, developmental delay, widespread tissue defects, and loss of organismal proteostasis. Our results reveal that the Sup35 prion domain exhibits prion-like properties when expressed in the multicellular organism C. elegans and adapts to different requirements for propagation that involve the autophagy-lysosome pathway to transmit cytosolic aggregation-prone proteins between tissues.
| Alzheimer's, Parkinson's, Huntington's, frontotemporal lobar degeneration (FTLD), amyotrophic lateral sclerosis (ALS), and prion diseases are all age-related, fatal neurodegenerative disorders. Hallmarks of these diseases include the expression of toxic protein species. The ability to spread and infect naive cells was thought to be limited to prions but has recently been observed for other disease-linked protein aggregates in tissue culture cells and transgenic mice. The underlying cellular pathways of this cell-to-cell transmission, however, remain elusive. We have developed a new prion model in the roundworm Caenorhabditis elegans and show that the appearance of aggregate species is associated with cellular toxicity, not only in the expressing cell but as well as in adjacent tissues. We monitored in real time the spreading of prion domains by autophagy-derived lysosomal vesicles from cell-to-cell. Given that autophagy and lysosomal degradation have a role in several neurodegenerative diseases, this cellular pathway might be the basis of amyloid infectivity in general.
| Transmissible spongiform encephalopathies (TSEs) or prion diseases are fatal, age-related, and infectious neurodegenerative disorders that affect humans (e.g., Creutzfeldt-Jakob disease) and animals (e.g., scrapie in sheep and bovine spongiform encephalopathy in cattle) [1]. At the molecular level, prions propagate by recruitment and conversion of the soluble α-helix-rich cellular PrPC into toxic aggregates of the pathological β-sheet-rich PrPSc isoform, via a mechanism described as seeded or nucleated polymerization [2]–[5]. The TSE agent is also infectious at the cellular level, where it transmits from cell-to-cell and infects naïve cells, both from within and outside the central nervous system [6], [7].
In yeast, prions can function as heritable epigenetic factors [8]–[12] upon forming an alternative self-propagating β-sheet-rich state from a soluble α-helix-rich fold. Yeast and mammalian prion determinants, however, do not share similarities in amino acid sequence, function, or subcellular localization. Yeast prions are naturally propagated within the cytosol from mother to daughter cells during cell division and require the disaggregase activity of the molecular chaperone Hsp104 to generate seeds and ensure dissemination [13]. In contrast, cell-surface localized mammalian prions are transmitted from cell-to-cell in terminally differentiated non-dividing cells. Sup35, like the majority of yeast prion proteins, contains a glutamine/asparagine (Q/N)-rich domain that confers the prion phenotype [14]. Although the mammalian prion protein PrP lacks this domain, other neurodegenerative disease proteins such as FUS (Fused in Sarcoma) and TDP-43 (TAR DNA-binding protein 43) have been shown to contain Q/N-rich prion-like domains [15]–[17].
There is increasing evidence that proteins closely associated with the neurodegenerative diseases Alzheimer's, Parkinson's, Huntington's, frontotemporal lobar degeneration (FTLD) and amyotrophic lateral sclerosis (ALS), exhibit prion-like properties [18]–[20]. Amyloid fibril assembly in general follows a nucleated polymerization reaction in vitro [9], and the addition of preformed fibrils or pathological brain extract seeds the polymerization of the corresponding monomeric protein in cell culture models or following injection into healthy mouse brains [24]–[31]. Furthermore, many proteins that form aggregates and fibrils exhibit cell non-autonomous effects and might spread among tissues within an organism [32]–[37]. The cellular processes and mechanisms that underlie cell-to-cell transmission of toxic protein species remain elusive in the current animal models that employ tissue culture cells and mice to investigate prion biology.
The nematode Caenorhabditis elegans has been widely used as a model system to investigate the biology of protein misfolding and toxicity [38]–[43], and has the advantage of transparent tissue types including muscle, intestinal, and neuronal tissue. We aimed to establish a new prion model system in this metazoan to examine the mechanisms of propagation of protein misfolding across tissues in a living organism. Since there are no known prions in C. elegans, and we wanted to avoid potential complications of infectious mammalian prions, we used the well-characterized cytosolic yeast prion protein Sup35 [8]. Here, we show that a cytosolic prion domain, NM, is highly toxic and can spread among tissues within the animal. The cell non-autonomous organismal toxicity of Sup35NM was associated with the accumulation of autophagy-derived vesicles, disruption of mitochondrial integrity, and the dynamic movement of the prion domain protein between tissues via autolysosomal vesicles.
Three versions of Sup35NM, corresponding to the full-length wild-type domain, NM, a deletion of the oligorepeat region (RΔ2-5), and an expansion of the oligorepeats (R2E2), were fused to YFP (yellow fluorescent protein) (Figure 1A) and expressed under the control of the body wall muscle (BWM) cell specific (m) promoter unc-54p. These NM constructs were selected based on previous observations that deletion of four of the five oligorepeats of the prion domain (RΔ2-5) leads to a strong decrease in prion induction, while expansion of this region (R2E2) significantly increases spontaneous prion formation [44].
In C. elegans lines expressing approximately similar levels of the transgenes (Figure 1C), NMm::YFP aggregates appeared in early embryonic stages of development and persisted through all larval stages into adulthood (Figure 1B). The appearance of aggregates was strictly related to the length of the oligorepeats such that R2E2 formed aggregates more rapidly and to higher levels than NM, while deletion of the oligorepeats in RΔ2-5 resulted in expression of a mostly soluble and diffuse protein (Figure 1B, 1D). The fluorescent foci in NMm::YFP and R2E2m::YFP coincided with higher levels of detergent insoluble protein relative to RΔ2-5m::YFP (Figure 1D).
To further characterize the biochemical and biophysical properties of the NM aggregates, we employed the dynamic imaging method of fluorescence recovery after photobleaching (FRAP). The diffuse fluorescence observed in RΔ2-5m::YFP expressing animals was shown to correspond to highly mobile protein species by FRAP analysis, in addition to the infrequent appearance of foci that were too small to be assessed by FRAP (Figure 2A). In contrast, examination of NMm::YFP and R2E2m::YFP foci at high magnification revealed highly diverse shapes and sizes that can be described as long fibril-like species (∼10 µm), large (∼2 µm) round spherical structures, and small (∼0.1 µm) foci (Figure 2B, 2C). These foci did not exhibit any obvious patterns among the BWM cells and were randomly distributed. Moreover, each progeny descending from a single hermaphrodite exhibited a unique pattern of R2E2m::YFP foci (Figure 2C) suggesting an influence of the individual cellular environment on aggregate phenotypes.
FRAP analysis on animals expressing NM and R2E2 (Figure 2A) revealed foci ranging from immobile aggregates that exhibited no FRAP recovery to foci that rapidly recovered fluorescence and thus were comprised of mixed populations of slowly mobile protein species. These two biophysical states of prion domain aggregates were closely aligned with the distinct visual morphologies, in that every fibril-like aggregate tested was comprised of immobile species, and that spherical aggregates detected in both R2E2 and NM animals corresponded to mobile aggregates that showed recovery following photobleaching (Figure 2A).
R2E2m::YFP animals exhibited a severe reduction in motility relative to wild-type N2 or RΔ2-5m::YFP animals (Figure 3A; Video S1, S2, S3) that was associated with a disruption of muscle ultrastructure revealed by rhodamine-phalloidin staining of myofilaments (Figure 3B). Moreover, nearly all of the R2E2m::YFP and NMm::YFP adults exhibited developmental delay and reduced fecundity, with R2E2m::RFP adults being often sterile (Ste) (Figure 3C, 3D). Whereas adult N2 wild-type and RΔ2-5 animals lay approximately 16 eggs within a 2.5 hour period, NM and R2E2 animals laid 8.5 and 4 eggs, respectively (Figure 3C). Furthermore, only 8% of R2E2m::YFP embryos and 1% of NMm::YFP embryos attained adulthood over a three day period at 20°C, relative to >93% achieving adulthood for wild-type N2 or RΔ2-5m::YFP embryos (Figure 3C). The slightly higher fraction of adult R2E2 animals detected after 72 hours is due to a more severe egg laying defect (Egl) of these animals. R2E2 animals often retained their eggs due to dysfunction of the vulva muscle leading to embryos being laid at later time points of development. Consequently, eggs laid by R2E2 animals tended to be older than corresponding NM, RΔ2-5, and wild-type N2 embryos that are deposited at the same time. NM animals exhibited a more severe embryonic lethal phenotype (Emb) than R2E2, while the latter animals exhibiting increased sterility (Ste) and producing fewer total progeny (Figure 3C). Animals that reached the L4 state of development after 72 hours became adult animals on the next day, whereas younger larvae were more likely to arrest in development (Figure 3D, Video S1, S2, S3, data not shown). In summary, while the populations of NM and R2E2 animals differed in their distribution among developmental states after 72 hours, expression of the highly aggregation-prone R2E2 resulted in a more severe toxic phenotype than NM (Figure 3C, 3D; Video S1, S2, S3).
Another characteristic of R2E2m::YFP expressing animals was a plethora of morphological defects that included reduced size (Sma), vacuolation (Vac), defective molting (Mlt), clear appearance (Clr), and disrupted gonadal and intestinal morphology (Figure 3E, and data not shown). Such defects affecting other tissues were observed to a lesser extent in NMm::YFP animals, and not detected in RΔ2-5m::YFP lines or in C. elegans lines expressing other aggregation-prone proteins [38], [39], [41]–[43] (data not shown).
The NM-dependent cellular defects were examined in more detail using transmission electron microscopy (TEM). Compared to wild-type N2 animals, the muscle cells of R2E2 expressing animals exhibited disrupted sarcomeres, fragmented mitochondria containing a drastically reduced number of cristae, and a complex array of double and single membrane bound organelles (Figure 4A). These vesicular structures are a hallmark of autophagy. Surprisingly, the cellular pathology observed in R2E2 expressing animals was not restricted to BWM cells and was also observed in other tissues in which R2E2 was not expressed. For example, intestinal cells that did not express R2E2 exhibited mitochondrial fragmentation with loss of cristae, and reduced levels of yolk and lipid droplets (Figure 4B).
These studies show that the number of oligorepeats in the prion domain directs the toxicity that results in multiple organismal phenotypes that extend beyond the primary tissue of NM expression.
To examine whether the induction of autophagy is a secondary cellular response due to damage of essential components like mitochondria, or if the prion domain is directly targeted by the autophagy-lysosome pathway (ALP), we employed C. elegans lines expressing markers of specific membraneous organelles. As the available markers for C. elegans are tagged with green fluorescent protein (GFP), we generated a C. elegans line expressing R2E2 tagged with red fluorescent protein (RFP) under the control of the myo-3 promoter for BWM cell specific expression (R2E2m::RFP). LGG-1::GFP transgenic animals that express the ortholog of the autophagosome marker LC3 (in mammals) or ATG8 (in yeast) were used to monitor autophagic vesicles [45].
In R2E2; LGG-1 transgenic lines, we observed co-localization of a subset of R2E2m::RFP foci with autophagosomes (Figure 5A). We also detected co-localization of R2E2m::RFP foci with RAB-7 positive late endosomes and specifically with LMP-1 positive lysosomes (Figure 5C, 5D). The majority of these lysosomes exhibited an unusual tubular shape (Figure 5D). Co-localization was not observed with RAB-5 (early endosomes) (Figure 5B), indicating that the R2E2-containing vesicles were derived from the autophagy pathway rather than from endocytosis. Our studies do not distinguish whether the R2E2m::RFP that co-localizes with vesicular structures corresponds to specific classes of aggregate species described before, as these vesicles have been excluded from FRAP analysis due to their small size (see materials and methods for more details). These data, together with the TEM analysis, suggest that the prion domain is a target of quality-control autophagy and is transferred from autophagosomes to RAB-7 positive amphisomes and LMP-1 positive autolysosomes, respectively.
Another striking characteristic of the tubular vesicles containing R2E2m::RFP was their dynamic movement within and between muscle cells, monitored by live-cell time-lapse imaging (Video S4, S5; Figure 6B). In particular, the over-expression of RAB-5 enhanced (and facilitated by visualizing single muscle cells) the detection of cell-to-cell transmission of RFP-positive vesicles between BWM cell quadrants (Video S5, Figure 6B). These observations are consistent with previous findings that RAB-5 over-expression increases autophagy [46].
The intercellular transport of R2E2-containing vesicles was unexpected as C. elegans body wall muscles are composed of individual mononucleated cells that are connected through gap junctions to allow electrical coupling for coordinated body movement [47] (Figure 6A). No dye coupling has been observed between single muscle cells, implying that there is no unregulated transfer of cytosolic proteins under normal physiological conditions [47]. This leads us to propose that R2E2 is actively transported by tubular vesicles from cell-to-cell. As mentioned before, these vesicles are different from the foci described in Figure 2. Neither the spherical (mobile in FRAP analysis) nor the fibril-like (immobile in FRAP analysis) aggregates are moving within or between cells. Only the small tubular vesicles are getting transmitted and we do not know the conformational state of R2E2 protein within these vesicles. Nevertheless, misfolding and aggregation is central to the toxicity phenotype, as RΔ2-5m::YFP, which does not form these aggregates, exhibits neither a cell autonomous nor cell non-autonomous toxicity.
The moving, tubular-shaped vesicles were only detected in R2E2m::RFP animals, but not observed with the corresponding proteins tagged with YFP. In contrast, the diverse aggregate species and other small vesicular structures (neither tubular nor moving) were visible with both YFP and RFP (compare Figure S1A and Figure 1A, Figure 2B and 2C). In transgenic animals expressing only the RFP fluorescent tag in BWM cells (RFPm), no movement of RFP between cells was observed (Figure 6C, Figure S1B, data not shown). This apparent discrepancy with the fluorescent tags can be explained by RFP being more stable in acidic environments whereas YFP is pH sensitive [48], indicating that these vesicles might exhibit a low lumenal pH that could explain the lack of similar fluorescent structures in R2E2m::YFP expressing animals. This speculation is supported by our results that the moving tubular vesicles co-localize with LMP-1::GFP, but not with LGG-1::GFP (compare Figure 5A and 5D). Indeed, staining of R2E2m::YFP animals with an anti-GFP antibody by indirect immunofluorescence revealed tubular structures in addition to foci visible with YFP fluorescence (Figure S2). This further supports our conclusion that acidified lysosomal vesicles containing the prion domain are transported from cell-to-cell.
Muscle cell-expressed R2E2 was also detected in vesicles of coelomocytes and the intestine (Figure 6D, Figure S3). Both, the intestine and coelomocytes, have been shown to endocytose molecules from the body cavity (pseudocoelom) [49], [50], suggesting that the tubular vesicles containing R2E2 could be released from BWM cells into the pseudocoelom and taken up by endocytosis from surrounding coelomocytes or intestinal cells. While the uptake of proteins from the pseudocoelom into coelomocytes and the intestine is not specific [49], [50], the amount of R2E2m::RFP that accumulates in these tissues is much more pronounced than for RFPm (compare Figure 6D and 6E). These results suggest that R2E2m::RFP is actively released from muscle cells into the pseudocoelom.
To examine the specificity of tissue movement of R2E2, we expressed RFP-tagged R2E2 in intestinal cells and monitored the dynamics of R2E2i::RFP-containing vesicles (Figure S4). Movement of R2E2 was observed by real-time imaging within intestinal cells (Video S6), and from the intestine into adjacent non-expressing muscle cells (Figure 6F, Figure S5, Video S7), thus confirming the spreading of the aggregation-prone prion domain across tissues.
Taken together, these observations reveal that R2E2m::RFP accumulates in tubular vesicles of autolysosomal origin that spread from expressing cells to non-expressing tissues in C. elegans. Furthermore, R2E2 seems to spread by two different pathways, either by a direct cell-to-cell transport of lysosomes, or through release into and endocytic uptake from the pseudocoelom (Figure 6B, 6D, 6F, Video S5 and S7).
We next examined whether the prion domain induces aggregation of its soluble isoform in C. elegans. Such a self-templating or seeding activity forms the basis of amyloid infectivity [9], [22]. To address this, we took advantage of the different aggregation properties of the prion domain constructs and used the non-aggregating RΔ2-5m::YFP as a folding sensor.
The seeding model posits a direct interaction of newly forming with preexisting aggregates, which in part is sequence-specific [51]. To examine this, we introduced in vitro generated recombinant fibrils by microinjection (Figure S7), into intestinal cells expressing RΔ2-5, as muscle cells were too small for microinjection. These studies were based on previous in vitro and cell culture observations that addition of preformed fibrils induces aggregation of the corresponding soluble NM in a sequence-specific manner [9], [44], [52]. We monitored the aggregation state of the RΔ2-5 folding sensor expressed under an intestine-specific (i) promoter (vha-6p). RΔ2-5 and NM constructs exhibited similar patterns of aggregation in the intestine as in muscle cells (Figure S6). Analogous to the biophysical properties exhibited in BWM cells, RΔ2-5i::GFP is soluble in intestinal cells (Figure S6; Figure 7A, 7E). Introduction of recombinant Sup35NM fibrils into intestinal cells resulted in the conversion of RΔ2-5 from a soluble to an aggregated state (Figure 7A, 7B, 7C, 7D) that did not co-localize with the injected Sup35NM fibrils. To address the sequence specificity of these effects, RΔ2-5 animals were also injected with recombinant fibrils of the asparagine-rich yeast prion protein Ure2p with high alpha-helical content [53], [54], or β-sheet rich amyloid Aβ1-42, respectively (Figure 7E). No aggregation of RΔ2-5 was observed upon injection of either protein.
To test whether cross-seeding occurs when both proteins are co-expressed in C. elegans tissues, we crossed RΔ2-5m::YFP with R2E2m::RFP animals. Despite being impaired for spontaneous aggregation, RΔ2-5m::YFP readily formed aggregate species that exhibited slow exchange in FRAP when co-expressed with R2E2m::RFP in BWM cells (Figure 8A, 8B, 8C). RΔ2-5m::YFP aggregates, however, only rarely co-localized with R2E2m::RFP foci (Figure 8B), consistent with observations from the injection experiments (Figure 7B, 7C). The RΔ2-5 sensor was further employed to test whether protein misfolding spreads from R2E2-expressing muscle cells to the intestine. Indeed, aggregation of RΔ2-5i::GFP increased when R2E2m::RFP and RΔ2-5i::GFP were co-expressed (Figure 8E, 8F). The absence of co-localization of RΔ2-5 and R2E2 foci, even when co-expressed (Figure 8B), indicates that aggregation of RΔ2-5 could be due to the global disruption of the folding environment, as seen in tissues co-expressing aggregates of polyglutamine and temperature sensitive mutant proteins [55], rather than from cross-seeding, which would imply co-aggregation of both proteins into heterologous aggregates [51]. Indeed, expression of R2E2 in muscle cells accelerated the age-dependent aggregation of an intestinal polyglutamine (polyQ) folding sensor (Q44i::YFP) [56] (Figure S8).
Taken together, these results show that R2E2m::RFP aggregates have multiple effects by seeding homologous soluble proteins in a sequence-specific manner and causing an imbalance of organismal proteostasis.
We have developed a metazoan prion model and examined the properties of a Q/N-rich prion domain in non-dividing terminally differentiated cells using C. elegans. A summary model describing the different aggregate species, vesicular structures and phenotypes observed in the C. elegans prion model, is shown in Figure 9. As the mechanism of prion propagation differs between unicellular eukaryotes and metazoans, it was unclear whether the prion propensities of Q/N-rich domains are universal and can adapt to different biological systems of cell-to-cell transmission. Spreading of the prion domain from an initial site of expression via autolysosomal vesicles occurs through actively regulated cellular processes, involving a direct transport from cell-to-cell and the release and endocytic uptake of these vesicles from the body cavity. This differs substantially from the propagation of [PSI+] in yeast that involves transfer of cytosolic NM propagons from mother to daughter cells during cell divison, that neither requires uptake into membraneous compartments nor active transport. Rather, the transmission of NM between cells and tissues in C. elegans is reminiscent of mammalian PrPSc propagation between post-mitotic neurons. Exosomes [57] and tunneling nanotubes [58] have been suggested as possible routes for cell-to-cell transmission of PrPSc. Either way, cytosolic content also gets transmitted, suggesting that cytosolic and membrane localized prion-like proteins might share some mechanistic aspects of transmission [59]. Indeed, there is now growing evidence that other disease-associated cytosolic protein aggregates can spread from cell-to-cell [19], [20], [26], [34], [35]. The spreading of the prion domain described here in C. elegans will allow us to compare the relative potential of tissue transmission with other aggregation-prone amyloidogenic proteins in our model system. It remains to be established if all major disease-associated proteins can spread from cell-to-cell themselves in a similar fashion like NM. Alternatively, prion-like domains might have implications in the spread of pathology throughout the nervous system by allowing a subset of modifiers like FUS and TDP-43 to transmit between cells, which then cause the subsequent aggregation of other disease-linked proteins.
Although motility defects are often associated with the expression of protein aggregates in C. elegans muscle cells [38], [39], [41]–[43], the expression of NM was unusually toxic compared to the expression of other disease-associated aggregation-prone proteins [38]–[43]. Aggregation and toxicity of NM were dependent on the oligopeptide repeats. Likewise, in yeast and mammals, the oligorepeats affect spontaneous prion induction and disease prevalence, respectively [44], [60], [61]. In yeast and infected tissue culture cells, prions often elicit no toxicity, suggesting that only non-toxic rapidly replicating variants are selected upon infection in these systems [62]. The unc-54 promoter used to express NM becomes active post-mitotically in 81 of 95 body wall muscle cells [63], [64]. The toxicity in C. elegans could therefore reflect the vulnerability of terminally differentiated non-dividing cells.
Autophagy is important for protein quality control and homeostasis in non-dividing neuronal cells [65], [66], consequently, autophagic failure has been implicated in prion diseases and other neurodegenerative disorders [67]–[69]. While activation of autophagy is beneficial to promote the clearance of disease-associated proteins [70]–[75], the chronic induction of autophagy could have deleterious consequences and may be insufficient to suppress toxicity [76]–[78], in particular if lysosomal function is already compromised during aging or by the chronic expression of mutant proteins [77]–[79]. In line with this, our preliminary results revealed that blocking autophagy by RNAi, to inhibit prion transmission, has only marginal effects to ameliorate NM toxicity in BWM cells (as measured by motility assays, data not shown), indicating that the autophagy-lysosomal pathway has a dual role and also reduces the load of misfolded proteins. Future studies using genome-wide RNAi screens will identify the cellular pathways that improve fitness in these animals.
One of the most striking consequences attributed to expression of the prion domain in C. elegans was mitochondrial fragmentation and loss of cristae. An equilibrium of steady fission and fusion events is critical for mitochondrial structure and function, and disruption of this homeostasis has been observed in disease and aging [80]. Intriguingly, a collapse of mitochondrial function was also observed in lysosomal storage disorders associated with impaired lysosomal degradation [81]–[83], and has been proposed to be a common secondary and final mediator of cell death in several diseases associated with autophagic failure and lysosomal dysfunction [81]–[83]. It remains to be determined whether related mechanisms are associated with the disruption of mitochondrial ultrastructure observed here for the C. elegans prion model.
There is accumulating evidence that lysosomes have additional roles to their conventional function as digestive organelles. Lysosomes constitute the exosomes of nonsecretory cells [84], are exocytosed during plasma membrane repair [85], and were shown to be transferred via tunneling nanotubes from endothelial progenitor cells to rescue prematurely senescent endothelia [86]. Our results reveal the involvement of lysosomes in the cell-to-cell transmission of cytosolic aggregation-prone proteins. Of note, the exocytosis or transfer of lysosomes may represent a specific cellular response to the prion domain as a cargo, because non mitotic aging tissues fail to secrete indigestible lysosomal content, which leads to the characteristic accumulation of lipofuscin [87]. It is tempting to speculate that proteins with prion domains might trigger a specific cellular response that leads to the release of LMP-1 positive vesicles.
Aggregation of NM in C. elegans occurs spontaneously upon its over-expression, in contrast to observations in bacterial and mammalian models [18], [88], [89]. In yeast, the induction of [PSI+] is dependent on the co-existence of [PIN+] or other compatible aggregation-prone proteins [90]–[92]. Perhaps similar factors such as endogenous Q/N-rich proteins are expressed in C. elegans that can act as [PIN+] [93].
The injection of preformed fibrils or co-expression of aggregation-prone variants seeds the polymerization of the corresponding monomeric protein [12], [22], [24], [28], [29] by a reaction known as nucleated or seeded polymerization [9], [22]. Only Sup35NM fibrils were able to cross-seed RΔ2-5 to form aggregates, whereas injection of fibrillar Abeta1-42 or Ure2p failed to do so, which suggests that seeding of RΔ2-5 is sequence-specific. However, Sup35NM fibrils or R2E2 aggregates did not co-localize with RΔ2-5 foci. The absence of co-aggregation might be due to conformational variations resulting from sequence differences within the NM oligorepeats [94], [95]. While the different prion domain variants might initially form heterologous seeds below the resolution of our imaging approaches, the preferred coalescence of molecules that have the same conformation might lead to distinct aggregates [51]. Alternatively, the ability to induce polyQ aggregation in a cell non-autonomous manner, suggests that expression of the aggregation-prone prion domain causes a global disruption of cellular proteostasis, and subsequent misfolding of unrelated metastable proteins, perhaps by titrating chaperones and other anti-aggregation factors [55], [90]. Most likely, misfolding of RΔ2-5 upon co-expression of R2E2 in the same or neighboring tissue results from a combinatory effect of sequence-specific cross-seeding together with an overload of the cellular folding capacity. Under these chronic proteotoxic stress conditions, one misfolded protein can accelerate aggregate formation of another aggregation-prone protein independent of protein sequence homology [41], [55].
In summary, this study provides new insights on the intrinsic properties of Q/N-rich prion domains in metazoans. Although the yeast prion domain NM is not a disease relevant peptide, this novel genetic C. elegans prion model can elucidate cellular pathways underlying the prion-like propagation of conformational changes in proteins between cells and tissues of multicellular organisms in health and disease.
Sup35NM constructs were amplified from the yeast expression vector p316CUP1-3SGFPSG [44] containing either the full-length NM, NM with a deletion of oligorepeats # 2-5 (aa 56-93) (RΔ2-5), or NM with a two-times extended oligorepeat # 2 (QGGYQQYNP) (R2E2) [44], by PCR standard methods. Insertion of appropriate restriction sites allowed cloning of the PCR amplicons into pPD30.38 [39]. This vector contains the promoter and enhancer elements from the unc-54 gene [96], as well as EYFP from the vector pEYFP-N1 (Clontech) [39]. For constructing myo-3p::sup35(r2e2)::rfp, myo-3p::rfp, vha-6p::sup35(rΔ2-5)::gfp, vha-6p::sup35(nm)::gfp, vha-6p::sup35(r2e5)::rfp, unc-54p::cfp::rab-5, and unc-54p::lmp-1::gfp expression vectors, the MultiSite Gateway Three-Fragment Vector Construction Kit (Invitrogen) was used. NM constructs were amplified from the pPD30.38 vectors using appropriate oligonucleotides for gateway cloning and inserted into the pDONR 221 entry vector by recombination. Likewise, the lmp-1 coding sequence was amplified from a N2 cDNA sample and inserted into the pDONR 221 entry vector. The plasmid pCZGY#3 ( = pDONR 201_rab-5) was a kind gift from Dr. Yishi Jin. Entry vectors pDONR P4-P1R containing myo-3 (approx. 2.4 kb upstream of the myo-3 gene), vha-6 (approx. 1.2 kb upstream of the vha-6 gene), or unc-54 (approx. 1 kb upstream of the unc-54 gene) promoter region and pDONR P2R-P3 coding for the C-terminal monomeric RFP or GFP tag, were generated accordingly. The N-terminal CFP was cloned between the unc-54 promoter and rab-5 using appropriate restriction sites. For co-localization, CFP::Rab-5 was false-colored green. All pDONR P2R-P3 entry vectors also contained the unc-54 3′UTR. Promoters, genes of interest and fluorescent tags were finally inserted into the destination vector pDEST R4-R3 in an in vitro recombination reaction.
Wild-type (N2, Bristol) and transgenic worms were handled using standard methods [97]. If not otherwise indicated, nematodes were grown on NGM plates seeded with the E. coli strain OP50 at 20°C. The strains NP1129 cdIs131[cc1p::gfp::rab-5+unc-119(+)+myo-2::gfp], NP871 cdIs66[cc1p::gfp::rab-7+unc-119(+)+myo-2::gfp], NP744 cdIs39[cc1p::gfp::rme-1(271alpha1)+unc-119(+)+myo-2::gfp], RT258 pwIs50[lmp-1p::lmp-1::gfp+Cb-unc-119(+)], and DA2123 adIs2122[lgg-1::GFP + rol-6(su1006)] were ordered from the Caenorhabditis Gene Center (CGC). The strain FY777 lin-15(n765ts); grEx170[Pmyo-3::gfp::rab-7; lin-15(+)] was a kind gift of Dr. Bruce Bamber. The following strains were generated for this study using germline transformation by microinjection:
AM801 rmIs319[unc-54p::sup35(rΔ2-5)::yfp], AM803 rmIs321[unc-54p::sup35(nm)::yfp], AM806 rmIs324[unc-54p::sup35(r2e2)::yfp], AM815 rmIs323[myo-3p::sup35(r2e2)::rfp], AM809 rmEx319[vha-6p::sup35(rΔ2-5)::gfp+myo-2p::mcherry], AM823 rmEx326[vha-6p::sup35(rΔ2-5)::gfp], AMf814 rmIs326[vha-6p::sup35(nm)::gfp+myo-2p::mcherry], AM883 rmEx338[myo-3p::rfp::rfp], AM887 rmEx339[unc-54p::cfp::rab-5], AM890 rmEx340[unc-54p::lmp-1::gfp].
Transgenic lines carrying extrachromosomal arrays were generated by microinjection of the above-mentioned plasmids into N2 wild-type animals. Integrations were obtained by gamma irradiation of animals carrying the respective extrachromosomal array. Integrated lines were backcrossed at least five times. Importantly, due to the high toxicity of some of the transgenes, the lines had to be backcrossed into N2 wild-type background regularly to avoid the occurrence of mutations that improve the health or suppress the NM aggregation phenotype of the transgenic lines. For the same reason, assays were performed on freshly backcrossed or crossed animals.
Nematodes were synchronized by transferring adult animals on a fresh plate and were allowed to lay eggs for 2.5 hours before removing. The amount of eggs laid was determined. Embryos were grown at 20°C for 72 hours before assessment of their developmental stage. Data obtained in at least three independent experiments were analyzed (≥200 worms total). In parallel, 20 L1 larvae were picked on fresh plates and grown for four days at 20°C before pictures and movies were taken.
L4 larvae from N2 and transgenic lines were transferred on fresh plates. Movement of crawling animals was recorded 24 hours later (with young adult worms) using a Leica M205 FA microscope with a Hamamatsu digital camera C10600-10B (Orca-R2, Leica Microsystems, Switzerland), and the Hamamatsu Simple PCI Imaging software. Videos were imported into ImageJ and speed (measured as body length per second) was analyzed using the wrMTrck plugin for ImageJ. Each sample containing 20–30 worms was recorded three times and the average speed of these movies was considered one biological sample. At least three biological replicates were obtained for each strain tested.
For rhodamine-phalloidin staining, transgenic lines were fixed (4% formaldehyde solution), permeabilized (130 mM Tris, pH 6.8; 700 mM ß-mercaptoethanol; 1% Triton X-100) and stained with rhodamine-phalloidin (Molecular Probes). For indirect antibody staining of R2E2, R2E2m::YFP animals were washed in M9, transferred onto Poly-L-Lysine coated microscope slides (Electron Microscopy Sciences), covered with coverslips and frozen on a metal block chilled to about −70°C on dry ice. The coverslips were snapped off and the slides were fixed in −20°C methanol, washed twice (1x PBS), blocked (1x PBS; 4% BSA; 0.1% Triton X-100), and incubated with anti-GFP antibody (ab6556 from Abcan) in blocking solution at 4°C over night. The next day, slides were washed 4x (1x PBS), incubated with secondary antibody (Alexa-456 conjugated goat anti-rabbit IgG, Invitrogen) for 1 h at room temperature, before being washed again, mounted (PermaFluor Aqueous Mounting Medium, Thermo Scientific) and sealed.
For light and fluorescence microscopy, animals were mounted on 2% agarose pads and immobilized in 2 mM levamisole. DIC (Nomarski) images were obtained using a Zeiss Axiovert 200 microscope with a Hamamatsu Orca 100 cooled CCD camera. Fluorescence microscopy and FRAP analysis were performed on a Zeiss LSM 510 confocal microscope with a 488 nm, 514 nm, or 563 nm line for excitation.
FRAP was performed by using the 63 x objective lens at 5 x zoom, with the 514 or 488 nm line for excitation of YFP or GFP, respectively. An area of 0.623 µm2 was bleached for 50 iterations at 100% transmission, after which time an image was collected every 123.35 ms. The relative fluorescence intensity (RFI) was determined by using RFI = (Tt/Ct)/(T0/C0), where T0 denotes the fluorescence intensity of the bleached region and C0 the control unbleached region, prior to bleaching, and Tt and Ct represent the fluorescence intensity at time t after photobleaching for the bleached and control region, respectively. Results show an average of at least 20 independent measurements for each strain. Foci that got rapidly and evenly bleached allover beyond the outline of the bleached region of interest (ROI) were excluded from these analysis, as they likely constitute vesicles containing soluble protein. In addition, foci, that had the same or a smaller size than the bleached ROI of 0.623 µm2 were not taken into account, as the motility of the protein within the same focus could not be assessed and therefore, vesicles containing soluble protein could not be distinguished from aggregates. FRAP analysis of RΔ2-5 foci in RΔ2-5m::YFP;R2E2m::RFP animals, RΔ2-5i::GFP animals 24 h after injected with rec. fibrils, and RΔ2-5i::GFP;R2E2m::RFP animals were performed on young adult (day 1 and 2 of adulthood) worms. Where indicated, a Leica SP5 II LSCM equipped with HyD detectors was used, especially for time-lapse imaging.
Nematodes were collected from a densely populated not starved 6 cm or 10 cm plate, washed in M9 buffer, and resuspended in lysis buffer (20 mM Tris, pH 7.5; 10 mM ß-mercaptoethanol; 0.5% Triton X-100; supplemented with complete protease inhibitor (Roche)) before shock freezing in liquid nitrogen. After three freeze-thaw cycles, the worm pellet was grinded with a motorized pestle, and lysed on ice, in the presence of 0.025 U/ml benzonase (Sigma). The lysate was centrifuged at 1000 rpm for 1 min in a tabletop centrifuge to pellet the carcasses. Protein concentration was determined using Bradford assay (Bio-Rad).
For the solubility assay, 200 µg of total protein was mixed with 2% N-Lauroylsarcosine, before ultracentrifugation at 100.000 g for 1 hour at 4°C. Supernatant and pellet fractions were separated and subjected to SDS-PAGE and subsequent immunoblotting. For transgene detection, the mouse anti-GFP monoclonal antibody (MMS-118R, Covance) was used, together with ECL plus (GE Healthcare). As a loading control, α-tubulin was detected by anti-α-tubulin antibody (T-5168, Sigma).
Samples were high pressure frozen (HPF) using a Leica EM PACT2, and maintained in LN2 until processed at low temperature in an automated freeze substitution unit (Leica EM AFS2). The freeze substitution solution (2% OsO4, 0.5% uranyl acetat, 3% H20 in Acetone) was cooled to −90°C before adding the HPF sample. Low temperature processing was performed over 3 days where the temperature was gradually increased to room temperature, followed by a gradual infiltration with EMBed 812 resin and polymerization. 90 nm thin sections were collected on Formvar-coated slot grids and stained with 2% uranyl acetate and Reynold's lead citrate. Samples were imaged at 80 kV in a JEOL 1230 transmission electron microscope. 4 different R2E2 expressing animals and 2 N2 animals were imaged.
Animals were synchronized via bleaching as described earlier [39]. Synchronized L1 larvae were transferred on fresh OP50-seeded plates ( = day 1). Animals were observed at 30–40x magnification with a stereomicroscope equipped for epifluorescence. The number of animals containing intestinal polyQ aggregates was determined on day 1 to 5 after synchronization. At least 300 animals (total) were assessed in 3 independent experiments.
Sup35NM, Aß (1–42), and full-length Ure2p expression, purification and assembly were performed as described [98]–[100]. Sup35NM, Aß (1–42) and Ure2p fibrils were spun at 15.000 g for 15 min at 4°C. The fibrils were resuspended in 50 mM HEPES, pH 7.5. Labeling was achieved by addition of 2 molar excess of the aminoreactive fluorescent dye Alexa Fluor 555 carboxylic acid, succinimidyl ester (Invitrogen) following the manufacturer's recommendations. Unreacted dye was removed by 3 cycles of sedimentation and resuspension of the fibrils in HEPES buffer. The amount and quality of recombinant fibrils was determined by solubility assay (16.000 g, 30 min, 4°C) and TEM.
The fibrillar nature of the generated assemblies was assessed using a JEOL 1400 transmission electron microscope following adsorption of the samples onto carbon-coated 200-mesh grids and negative staining with 1% uranyl acetate. The images were recorded with a Gatan Orius CCD camera.
Immediately before injection, recombinant fibrils were diluted into 50 mM HEPES buffer, pH 7.5, to a final concentration of 100 µM and sonicated (1510 Branson water sonicator) for 30 min. 50 mM HEPES buffer only was used as a control. Young adult worms were microinjected according to standard methods into the cytosol of intestinal cells. After 24 h nematodes were imaged using a Zeiss LSM 510 confocal microscope. At least 5 animals were injected with each fibril preparation and analyzed per experiment and experiments were repeated 2–3 times. Our microinjection setup does not allow controlling for the injection of the exact same amount of protein fibrils into each animal. Therefore, we assessed different concentrations of the fibril preparations in their ability to induce RΔ2-5 aggregation. The fibril concentration had only an impact on the quantity (how much protein was seeded), but not on the quality (if there was seeding) of aggregation induction.
|
10.1371/journal.pgen.1003419 | Genome-Wide Testing of Putative Functional Exonic Variants in Relationship with Breast and Prostate Cancer Risk in a Multiethnic Population | Rare variation in protein coding sequence is poorly captured by GWAS arrays and has been hypothesized to contribute to disease heritability. Using the Illumina HumanExome SNP array, we successfully genotyped 191,032 common and rare non-synonymous, splice site, or nonsense variants in a multiethnic sample of 2,984 breast cancer cases, 4,376 prostate cancer cases, and 7,545 controls. In breast cancer, the strongest associations included either SNPs in or gene burden scores for genes LDLRAD1, SLC19A1, FGFBP3, CASP5, MMAB, SLC16A6, and INS-IGF2. In prostate cancer, one of the most associated SNPs was in the gene GPRC6A (rs2274911, Pro91Ser, OR = 0.88, P = 1.3×10−5) near to a known risk locus for prostate cancer; other suggestive associations were noted in genes such as F13A1, ANXA4, MANSC1, and GP6. For both breast and prostate cancer, several of the most significant associations involving SNPs or gene burden scores (sum of minor alleles) were noted in genes previously reported to be associated with a cancer-related phenotype. However, only one of the associations (rs145889899 in LDLRAD1, p = 2.5×10−7 only seen in African Americans) for overall breast or prostate cancer risk was statistically significant after correcting for multiple comparisons. In addition to breast and prostate cancer, other cancer-related traits were examined (body mass index, PSA level, and alcohol drinking) with a number of known and potentially novel associations described. In general, these findings do not support there being many protein coding variants of moderate to high risk for breast and prostate cancer with odds ratios over a range that is probably required for protein coding variation to play a truly outstanding role in risk heritability. Very large sample sizes will be required to better define the role of rare and less penetrant coding variation in prostate and breast cancer disease genetics.
| For breast and prostate cancer, GWAS have revealed many risk variants (>70 for each cancer as of this report). All together the common variants in these regions explain only a minority of familial risk of these cancers. Using the Illumina HumanExome SNP array, we explored the hypothesis of rare coding variation contributing to breast and prostate cancer risk in a sample of African American, Latino, Japanese, Native Hawaiian, and European American breast and prostate cancer cases and controls from the Multiethnic Cohort study. While only one association exceeded significance thresholds after correcting for multiple comparisons, a number of suggestive associations involving genes previously reported to be associated with a cancer-related phenotype were noted. Our results do not generally support a major role of protein-coding variants with odds ratios over a range that is probably required for protein coding variation to play a truly outstanding role in risk heritability. If very rare and/or less penetrant coding variants underlie disease heritability of these cancers, then very large sample sizes (i.e. consortia) will be required for their discovery.
| For most common diseases and traits the genetic basis underlying susceptibility has yet to be completely revealed. While genome-wide association studies (GWAS) have been remarkably successful in identifying common genetic variants associated with risk, the effect sizes of the risk alleles have been modest (relative risk, RR of 1.1–1.4) and in most cases, even in sum, they can explain only a fraction of familial risk or disease heritability. GWAS have relied almost exclusively on Illumina and Affymetrix SNP arrays, with SNP content selected primarily from HapMap to capture a large fraction of common variation in coding and non-coding regions in populations of European ancestry. The vast majority of alleles with frequencies <5%, and in particularly those with frequencies ≤1%, have not been tested. This low allele frequency spectrum of genetic variation represents a very large fraction of all variation in the human genome. Thus, to date, a large fraction of genetic variation has yet to be explored with respect to disease etiology.
It is possible that the majority of less common (1–5%) and rare variants (<1%) will have weak effects, like the GWAS-identified common variants, and if this is the case then very large studies will be required for their discovery. An alternative hypothesis is that less common and rare variants convey larger relative risks than common variants, and indeed this assumption is required in order that rare variants contribute meaningfully to the understanding of inherited susceptibility. Such enhancement of effect sizes for rarer alleles may be especially relevant to rare coding variants given their dominant role in the etiology of “Mendelian” disorders (e.g. the OMIM database [1]). Support for the hypothesis that rare coding variation also profoundly affects risk of certain “complex” diseases is growing and there are now a number of such examples including rare missense variants in CHEK2, ATM, NBS1, RAD50, BRIP1, and PALB2 in breast cancer [2], rare coding mutations in RAD51D and BRIP1 in ovarian cancer [3], [4], as well as rare coding variants in genes implicated in hyperglyceridemina [5] and colorectal cancer adenomas [6]. More recently, whole-genome and candidate gene sequencing studies have revealed rare coding variants in ALDH16A1 for gout [7] and a number of genes (NOD2, IL23R, CARD9, IL18RAP, CUL2, C1orf106, PTPN22 and MEC19) involved in inflammatory bowel disease [8]. Studies in prostate cancer have reported rare gene coding mutations in BRCA2 (found in 2% of cases <55 years) to be associated with greater risk of prostate cancer (RR>4.5) and more aggressive disease [9], [10]. For many of these examples, in addition to single SNP association testing, burden of rare variation analyses have been applied to increase the number of observations in the comparison groups (and thus the statistical power), and to provide statistical support for the involvement of the gene which is not achieved when examining large number of SNPs in any given gene.
To date, a lack of technology to survey the genome and accurately enumerate and test the variants in large numbers of samples has limited the exploration of less common and rare alleles. In the past year the Illumina Infinium HumanExome array (or “exome chip”) has been developed in collaboration with investigators who combined whole-exome sequencing conducted in >12,000 individuals of primarily European ancestry as well as in small numbers of other racial/ethnic minorities including African Americans, Hispanics, and Asians; the content on the array includes >200,000 putative functional exonic variants and is aimed to provide comprehensive testing on all non-synonymous variants above 0.1% frequency in Europeans. In the present study, we have utilized this array to test the hypothesis that there are less common and rare functional variants in the coding regions of genes that convey risk for breast and prostate cancer of greater magnitude than the common variants revealed through GWAS. We tested both single markers as well as gene summaries of the burden of rare alleles in multiethnic studies of invasive incident breast cancer and prostate cancer in the Multiethnic Cohort study (MEC: 3,141 breast cancer cases, 4,675 incident prostate cancer cases and 8,021 controls). In addition we conducted exploratory analyses of rare variants in relationship with several breast and prostate cancer-related traits ascertained at baseline in the entire MEC sample (n = 15,837).
The analysis included 217,601 putative functional variants (of 247,870 total markers listed on the array), predicted to alter the protein coding sequence, and which passed quality control procedures (see Methods). Of the 15,837 samples, 14,905 were included in the analysis (3,315 European Americans, 3,854 African Americans, 3,106 Latinos, 3,843 Japanese Americans and 787 Native Hawaiians; see Methods for exclusion criteria). A few mitochondrial SNPs were included on the array (n = 165 SNPs passing quality control) but are not discussed here (no associations with them were seen in the top ranked 1,000 associations for either breast or prostate cancer). The number of breast and prostate cancer cases and controls are shown in Table 1. In this multiethnic sample, 191,032 (88%) putative functional variants were found to be polymorphic in at least one population, with 26,569 (12%) being monomorphic in all five populations (Figure 1). The percentage of monomorphic SNPs ranged from 34.1% in African Americans, 39.6% in European Americans and 43.3% in Latinos to 66.8% in Native Hawaiians and 74.2% in Japanese Americans (Figure S1). Of the polymorphic SNPs, 178,776 (93.4%) were nonsynonymous (NS) variants, 8,308 (4.4%) splice site (SP) variants, and 3,948 (2.1%) nonsense variants which either lead to a gain or loss of a stop codon. Of the polymorphic SNPs, 34,834 (18.2%) were polymorphic in all four of the largest populations (excluding Native Hawaiians), with 81,713 SNPs (42.7%) being polymorphic in African Americans, Latinos and European Americans (Figure 2). African Americans had the largest number of unique polymorphic SNPs (21,908, 11.4%), followed by European Americans (16,653, 8.7%), Japanese Americans (6,776, 3.5%) and Latinos (5,134, 2.7%).
In the pooled sample, 190,662 putative functional (NS, SP, or stop) SNPs had a minor allele frequency (MAF) <1% (56,759<0.01%; 85,897 between 0.01% and 0.1%, and 48,006 between 0.1% and 1%) (Figure 1, Figure S1). The minor allele frequency distributions were similar across three of the five populations with African Americans, European Americans and Latinos having roughly the same number of SNPs with frequencies greater than 0 and less than 1% (100–110 thousand); However there were only 37,979 SNPs with a frequency above zero and less than 1% in Japanese Americans and 52,985 in Native Hawaiians. The number of SNPs with a frequency >1% ranged from approximately 18–35 thousand between sampled populations.
Inspection of the distribution of the chi-square (score) tests from models for overall breast or prostate cancer showed evidence of over-dispersion of test statistics (genomic control lambda estimate to be approximately 1.15 for breast and 1.20 for prostate) however when very rare SNPs were removed (MAF<0.1% overall) then the Wald statistics appeared to be sampled from an overall central chi-square distribution (genomic control lambda = 1.00 for breast cancer and lambda = 1.05 for prostate cancer). In the gene burden analyses, the distribution of observed score tests showed mild evidence of over-dispersion (lambda = 1.04 for breast cancer and lambda = 1.06 for prostate cancer). When the single SNP analysis was restricted to estrogen receptor-negative (ER-) breast or advanced prostate cancer, where there were many more controls than cases included in each model, then the behavior of the score test for the single SNP associations was problematic for rare SNPs. For such SNPs we followed up any apparently globally significant associations with exact logistic regression analysis, in order to reduce what appeared to be a proliferation of false positive signals.
The total number of genes having at least one polymorphic functional variant genotyped and passed quality control varied slightly between breast (17,168 genes) and prostate cancer (17,203 genes) due to sampling (i.e. some variants were polymorphic only for breast cancer cases and so were not included in the prostate cancer analyses and vice versa).
In the ethnic-pooled breast cancer analyses (2,984 cases and 7,545 controls), the most significant predicted protein-altering variant was a rare SP variant rs145889899 at the splice donor site in the second intron of the gene LDLRAD1 (OR = 3.74, p = 2.5×10−7), which was almost exclusively seen in African Americans, this variant was statistically significant at our exome-wide level (nominal p<3.9×10−7, see Methods). Of the top 10 ranked associations, the remaining 9 involved NS variants (p-values ≥1.3×10−6, Table 2, Table S1). None of the other associations met the Bonferroni adjustment for multiple comparison testing. All of the 10 most associated variants, were quite rare and present mainly or exclusively in one or two ethnic groups. The genes containing the most significant SNPs for breast cancer ranged widely in apparent function (see Table 2) with GWAS associations reported with SNPs in CFB (complement factor B) for age-related macular degeneration [11], BAZ2A for platelet counts [12] and ACADS for metabolic traits [13].Table S1 gives information for the 100 most significant associations for breast cancer, both overall and by ethnic group when including all SNPs passing quality control (not just the non-synonymous, splice site and nonsense variants described here).
For ER- breast cancer (n = 441 cases) many associations (358) with very rare SNPs were nominally significant using the score test but the p-values failed to stand up to further investigation using exact logistic regression (the exact p-values ranged from 3×10−5 to 0.21). The many small p-values apparently reflected overly liberal behavior of the score test when alleles are rare and when there are many more cases than controls. In order to reduce discussion of a large number of likely false positive tests we consider in the subtype analyses only SNPs with at least 10 minor alleles seen over all cases and controls. With this restriction we found a total of ten globally significant SNPs (using the score test). However, p-values from exact logistic regression for these SNPs were again far less striking (ranging from 3×10−5 to 1.5×10−3).
When restricted to estrogen receptor-positive (ER+) cases (n = 1,688) (and screening out SNPs with less than 10 minor alleles seen) the most significant coding SNP was a rare NS variant in UMODL1 (exm1573155, Ala542Thr, OR = 7.28, p = 9.8×10−7) (Table 2, Table S2). This SNP had a frequency of just over 0.2% in African Americans controls and 0 in the other groups. No associations are reported for this gene in the GWAS catalog. Neither this SNP nor any others were significant after correction for multiple testing.
In ethnic-specific analyses of overall breast cancer only one additional SNP (in FANCI) met our criteria (p<3.9×10−7) of global significance. This NS variant (rs62020347, Pro55Leu) was common in European Americans, African Americans, and Latinos (3–8% frequency) but was only associated with risk among European Americans (MAF 8%, OR = 0.47, p = 1.8×10−7) and was weakly associated with risk overall (p = 0.02) (Table S1).
Table 3 summarizes the most significant findings from the gene burden (sum of coding variants) analysis based on all common and rare (≤1%) functional SNPs in each gene. Further details are given in Table S5. For overall breast cancer no gene burden sum passed the Bonferroni criteria (3×10−6) for global significance for testing approximately 17,200 genes (see Methods). The strongest associations were seen for MMAB (p = 5.0×10−5), SLC16A6 (p = 5.0×10−5) and INS-IGF2 (p = 1.2×10−4). The MMAB gene is close to non-exonic SNPs that have been associated with HDL cholesterol [14] and one of those GWAS SNPs (the intronic variant rs7134594) was among our top 100 single SNP associations with breast cancer (Table S1). INS-IGF2 contains an intronic SNP that has been associated with type 1 diabetes [15]. Restricting the gene burden analysis to only SNPs with overall frequency ≤1% gave non-significant associations as well (p>8×10−6) and none of the top five genes in these analyses have globally significant GWAS associations reported. For ER+ breast cancer, the burden of rare SNPs in gene FGFBP3 was nominally globally associated (p = 6×10−7) although follow-up using exact logistic regression gave a larger p-value (1.0×10−4). This gene included five rare SNPs and no reports of any GWAS associations for SNPs near this gene are found in the GWAS catalog. When examining ER- breast cancer, the burden of variants in MMAB remained one of the strongest associations (p = 2.0×10−5). The burden of coding SNPs (all of which were rare) in EGR2 was the leading association in the ER- analysis with a p-value from the score test of 1.2×10−11. A variant upstream of EGR2 has been associated in a GWAS of Ewings sarcoma [16]. Rare variant burdens also met our criteria for global significance for CNR1 (p = 1.7×10−10), FKSG83 (p = 1.5×10−8), GATM (p = 4.8×10−7), and ACSBG1 (p = 5.3×10−7). Again as for the single SNP results for ER- disease, these p-values were found to be overly liberal compared to an exact test (the smallest exact logistic regression p-value was 2.8×10−5 for ACSBG1)
For overall prostate cancer (4,376 cases and 7,545 controls) none of the single SNP associations with prostate cancer met the Bonferroni adjustment for multiple comparison testing (nominal p<3.9×10−7). The top two associations found for prostate cancer were for rare NS variants in F13A1 (rs140712764, Val170Ile, OR = 28.0, p = 9.1×10−7) and ANXA4 (rs146778617, Val315Phe,OR = 4.52, p = 6.0×10−6), Table 4, see also Table S2. Gene F13A1 is a coagulant factor gene not obviously related to prostate cancer etiology. ANXA4 encodes a protein that has been discussed as a possible marker for gastric cancer [17]. Of note, the third most significant association was for a common NS variant in GPRC6A (rs2274911, Pro91Ser, OR = 0.88, P = 1.3×10−5). This gene is nearby to RFX6, which harbors an intronic variant (rs339331) that has been reported in a GWAS of prostate cancer in Japanese men [18]. The SNP rs2274911 is common in all populations (MAFs of 24–43%) (Table 4) and the protective effect of the minor allele was generally consistent in each group (OR = 0.78 to 0.95, over the five groups). This NS variant is correlated with the known intronic variant (rs339331, which is included on the Illumina HumanExome array) in all populations (r2 between 0.74 and 0.98) ; in conditional analyses neither of these two SNPs remained significant after the other was forced into the model (P>0.2); thus these two variants are probably capturing the same signal, with the NS SNP in GPRC6A a potentially plausible susceptibility variant. The top 10 ranked associations (Table 4) were all NS variants and 4 were common with a MAF>10% in all ethnic groups.
When restricted to advanced cases (n = 499), similarly as for ER- breast cancer, many associations with very rare SNPs were nominally significant using the score test (69 total for SNPs with less than 10 minor alleles observed) but the p-values failed to stand up to further investigation using exact logistic regression (with p-values all <3×10−5). In order to reduce discussion of a large number of likely false positive tests we considered in subtype (advanced/nonadvanced) analyses only SNPs with at least 10 minor alleles seen over all cases and controls used in the analysis. Of the remaining SNPs we found that four NS SNPs with at least 10 minor alleles present were nominally significant using the score test criteria (Table 4, Table S4). These included NS variants in KLHL30 (exm280349, Arg108His, OR = 13.9, p = 1.7×10−9), PPP1R15A (rs45533432, Arg65Gly: OR = 4.67, p = 1.2×10−8), MUC12 (rs143984295, Ala101Thr , OR = 14.4, p = 1.5×10−8) and RP1 (rs114797722, Ala1326Pro, OR = 13.4 p = 2×10−8). These SNPs were all quite rare in the four largest populations (0.1%–1%). P-values from exact logistic regression for these SNPs were again less significant with p-values between 1.4×10−6 and 4.6×10−4).
For non-advanced disease (n = 3,666 cases), the strongest associations were with the same SNPs as overall prostate cancer (rs140712764 in F13A1, rs146778617 in ANXA4, rs2274911 in GPRC6A) and also with rs61746620 in ZKSCAN2 (Ala574Val, OR = 13.4, p = 1.3×10−5), although none of these were significant at our Bonferroni criteria.
None of the gene burden analyses were significant for overall prostate cancer after correcting for multiple comparisons (p<3×10−6) either when including common coding variants or when restricting the results to SNPs with frequency ≤1% (Table 5, Table S5). When the analysis was restricted to advanced prostate cancer, four gene burdens (for SAMD1, FOXF2, NOL4 and CPA3) were significant using the score test but not by exact logistic regression (p = 2.5×10−3 , 3.3×10−3 , 5.0×10−3 and 3.4×10−6 respectively). No notable findings were observed when only localized prostate cancer was assessed.
We also examined additional cancer-related traits: body mass index (BMI), alcohol intake, as well as circulating PSA levels (Table S8). A number of NS variants have already been strongly associated with many of these traits, such as rs671 (Glu504Lys) in ALDH2 with alcohol intake [23], rs17632542 [Ile179Thr] in KLK3 and circulating PSA levels [24], [25] and rs198977 [Arg250Trp] in KLK2 and the ratio of free to total PSA [26]. For each trait, the 10 most associated variants on the array (including non-functional SNPs, i.e. GWAS SNPs) are provided in Table S9. We also observed a number of suggestive associations at p<3.9×10−7 with rare coding variants in some genes that are biologically plausible for each trait. Three variants were strongly associated with blood PSA levels (chr19: Hg19 position: 4552446, Thr326Met, SEMA6B, 0.1% MAF in African Americans and monomorphic in all of the other populations, beta = 3.8, p = 3.8×10−9; rs17632542, Ile136Thr, beta = −0.4588, p = 1.0×10−8 MAF.06 in European Americans; rs148595483, Asn322Lys, CCDC78, 0–0.1% MAF across populations, beta = −2.9, p = 2.4×10−8). We also found a number of significant associations with very rare NS variants that were observed in 2–7 individuals and BMI (rs146199292, Asn31Lys, OSBPL11, beta = 19.9 p = 1.2×10−10; rs149954327, Leu458Val, STON1-GTF2A1L, beta = 15.2 p = 1.5×10−9; rs146922831, Lys608Asn, LRGUK, beta = 9.2, p = 3.0×10−8). The variants were very rare in African Americans with frequencies <0.09% and monomorphic in all of the other populations except for rs146199292 in Latinos (0.02%). Variations in these genes have been reported in association with conditions related with BMI, including cardiovascular risk factors, type 2 diabetes and polycystic ovarian syndrome [27], [28], [29]. The carriers of these rare alleles were clustered at the extreme high end of the BMI distribution. All these potentially novel associations will need further follow-up.
This paper presents an initial investigation of the role of coding variation in the genetics of breast and prostate cancer. Our initial analysis fails to find strong evidence for the hypothesis that relatively rare coding variation is highly determinative of breast or prostate cancer risk either overall or by subtype. Our sample sizes in each racial/ethnic group were each relatively small (roughly 1,000 cases and 2,000 controls in the largest groups) however these sample sizes are large enough to detect risk alleles with moderate to large effects (odds ratios of 3–13) appearing in quite low frequency (0.1–1%) and to examine whether such coding variation underlie (by so-called synthetic association [30]) many GWAS associations. While caution is advised in interpreting our results, especially for other than European racial/ethnic groups (since the array utilized was predominantly based upon sequence information for Europeans and is not expected to cover other groups equally well), it appears that future studies to understand the relationship between rare coding variation and breast and prostate cancer risk will likely require the very large sample sizes needed to target much less penetrant alleles.
Our analyses consisted of both single variant analysis and simple gene burden analyses. The gene burden analyses consisted of summing the minor alleles of coding variants including either all coding variants regardless of their frequency, or only those variants with MAF <1% in our overall sample. While this gene-burden test assumes implicitly that all coding variants have the same direction of effect, this is reasonable given that the power of detecting rare protective alleles in a case-control study such as this one (where controls can be regarded as representative of the population) is much less than the power to detect rare risk alleles. The rare variant sum therefore is not very sensitive to the presence of rare protective alleles in a gene.
One association for breast cancer, a single SNP in LDLRAD1, appeared to pass our established level of global significance (p<3.9×10−7) when all cases were examined. No associations (either single SNPs or gene burdens) were globally significant for overall prostate cancer. Subset analyses, by ER status for breast cancer or advanced/non-advanced for prostate cancer generally failed to show believable associations. While the score test gave many “globally significant” associations these apparently reflected excess type I error of this test when both the number of cases is small compared to the number of controls and when the SNPs were rare. This breakdown in reliability is similar to that seen for the uncorrected Pearson chi-square test (a special case of the score test when no covariates are present), which is well-known to have poor control of type I error when the expected number of cases is very small for a cell. Following-up such associations with exact logistic regression implemented in SAS (Cary, NC) provided larger p-values not globally significant using our criteria.
Nevertheless a number of suggestive findings were observed that are worthy of further attempts at replication: The splice site variant rs145889899 in LDLRAD1 (our top finding for overall breast cancer) is found in low frequency (<1%) in African American controls (higher of course in cases since this is nominally a risk variant), and only seen among cases in the other groups. No associations with any disease or phenotype have to date been reported for this gene. Among the other genes highlighted in Table 2 or Table 3, associations have been reported for SNPs in SLC19A1 and CASP5 for renal cancer [31], [32]; BAZ2A has been reported to be up-regulated in CLL patients [33]. Also notable is a strong link between SNPs in EGR2 (ER- association) and risk of Ewing's sarcoma [16].
For prostate cancer (all cases) the third strongest association result was for a common NS coding variant (rs2274911) in GPRC6A that is in very high LD with the known intronic GWAS variant rs339331. In our data the NS variant was slightly more associated (Table S3) with prostate cancer risk (p = 1.3×10−5) than was rs339331 (p = 2.1×10−5). The coding SNP is arguably a more likely causal variant than the intronic SNP since expression of GRPC6A is substantially increased in prostate cancer cell lines, and mice deficient in GRPC6A show retarded prostate cancer progression [34]. In addition, GRPC6A deficiency in mice also attenuates the rapid signaling responses to testosterone, an androgen that is critical for initiation and progression of prostate cancer [35].
Other suggestive findings for prostate cancer include SNPs in a variety of genes such as F13A1 expression of which has been associated with bone metastasis in prostate cancer [36], ANXA4 which is up-regulated in gastric and other cancers [17], NSD1 where cryptic translocations may be involved in AML occurrence [37] and MUC12, expression of which has been reported to be a prognostic marker in colon cancer [38]. The burden of rare SNPs in FGFBP6 (one of the stronger association seen for breast cancer) was also among the top associations for overall prostate cancer (Table 5, p = 1.5×10−4).
We evaluated also associations in regions surrounding known (GWAS) risk alleles as a partial fine-mapping exercise; we specifically focused upon (1) coding alleles reported to be in high LD (in Europeans using 1000 Genomes data) with the index marker, and (2) other (generally less common) coding alleles within 500 kb of the GWAS alleles, that might show associations that could underlie (by synthetic association [30]) GWAS associations. A number of GWAS risk alleles are in reasonable LD (r2>0.3) with coding SNPs on the array and several of the latter show nominal associations (p<0.05) with breast cancer risk including SNPs in STXBP4, ZNF45, and ZNF404 which are all worth evaluating as candidate loci potentially explaining the index GWAS associations. For prostate cancer, a similar observation is made most notably for GPRC6A but also for MLPH (GWAS index = rs7584330, chromosome 2, p = 0.003), PDLIM5 (rs12500426, chromosome 4, p = 0.019), RNMTL1 (rs684232, chromosome 17, p = 0.024), KLK3 (rs2735839, chromosome 19, p = 0.0046), and RTEL1 (rs6062509, chromosome 20, p = 0.001). Previous reports [24], [39] have highlighted the NS SNP rs17632542 in KLK3 as highly associated with PSA level and a highly significant risk variant in fine-mapping of the locus near rs2735839 [39]; while no report for prostate cancer exists for coding SNPs in RTEL1, another NS SNP, rs3208008, in RTEL1 has been found to be associated with glioma risk [40].
Other coding SNPs that could include causal variants producing synthetic associations (associations of rare with common SNPs of high penetrance) include SNPs in genes INS-IGF2, ZFYVE26, C16orf46, UNC13A, NRIP1 and CCDC91 for breast cancer and SNPs in SNED1 and PASK for prostate cancer. These do not have high r2 with the GWAS variants as they are mostly rare (and are >100 kb away from the index signal) but their nominally strong associations (p-values<1×10−3) might possibly be indicative of signals extending for many thousands of base pairs, although it will take much larger studies to verify or refute this.
We found little evidence that the NS, SP, or nonsense variants captured by the HumanExome SNP array that fall within known or suspected high risk genes for breast or prostate cancer are meaningfully associated with either cancer. The Illumina array does not directly interrogate the rare, high-risk mutations, such as frameshift mutations in BRCA1 or BRCA2 (e.g. c.68_69delAG) [41], as very few indels are included on this array (just 136 were examined here). The inability to address frameshift mutations either within known risk genes or more widely is a limitation of this report. Other limitations include the focus on Europeans in the development of the array (as seems to be particularly reflected in the relatively small fraction of SNPs found to be polymorphic in Japanese Americans), and the loss of some targeted SNPs in the manufacturing process and in our QC procedures. In addition, this technology (unlike exome sequencing) cannot address the role of either private variation or of variants too rare to have been reliably identified during the discovery phase of the development of the array.
Genotyping cases and controls from our prospective cohort allowed us an opportunity to examine other cancer-related phenotypes and traits for which data and specimens had been collected prior to breast or prostate cancer diagnosis. While two of these endpoints (BMI, alcohol) were based on self-report, we were able to strongly replicate a number of known associations such as rs671 in ALDH2 with alcohol intake which is proof of principle that the exome array has the potential to reveal biologically relevant coding variants. Apparently novel findings for PSA, BMI, and alcohol consumption will need to be replicated in large-scale exome association analyses; hopefully making the results from these preliminary analyses in a multiethnic population broadly available will contribute to novel discoveries and further understanding the genetic basis of these traits.
In order for rare variants to play an important role in explaining missing heritability [42] even in composite they must have effects that are larger in magnitude than those observed for common SNPs. Roughly speaking, for a given allele the contribution to additive heritability (under a liability model for example [43]) is proportional to 2b2p(1-p) where b is the log odds ratio (OR) and p is the frequency for that allele. Under simplifying assumptions (such as limited selection and constant population sizes) population genetics theory [44] indicates that there should be approximately as many variants “moderately rare” with frequency in the range 0.1 to 1% as there are the common variants in the range 5 to 50% that have been the targets of GWAS studies to date. However, in order that variants in the frequency range from 0.1 to 1% have the same composite effects on risk as do those in the frequency range from 5 to 50% then the magnitude of effect sizes must be considerably larger than for the common variants; if ORs for common variants lie in the range from 1.1 to 1.3 then ORs in the range from 2 to 6 are needed for the rare and common alleles to have similarly sized roles in disease susceptibility (assuming that the same fraction of all rare alleles are risk variants as for common alleles). Moreover, under the hypothesis that the coding regions of the genome (∼1% of the total genome) by themselves play an profound role in disease susceptibility these ORs would likely need to be skewed even higher – i.e. if rarer variation in 1% of the genome was to play as much a role as does common variation over the entire genome then the existence of ORs above 10 or even greater for such variation may arguably be a necessary consequence.
Realistically our study only begins the assessment of whether a range of effects for “moderately rare” coding variants is possible: the detectable ORs in this study range from approximately 3 to 13 for alleles with frequency 1 to 0.1%, respectively. While these are large ORs the above argument indicates that such effect sizes are not unreasonable if rarer protein coding variation plays a similar role in the heritability of risk as does common variation genome-wide. Our failure to find such ORs for the rarer alleles may be providing evidence against coding variation having a predominant role in breast and prostate cancer heritability and risk (outside of high risk families).
In summary, the analyses and methods described here do not support NS variants on the current exome chip as conveying moderate to high risk for breast and prostate cancer. While some suggestive findings are noted it is likely that very large sample sizes of the order that can be only developed through collaborative efforts such as those now engaged in the NCI GAME-ON post-GWAS meta-analysis of common variants, will be required in order to further the understanding of the role of rare NS and other coding variation in disease genetics. Exome sequencing of high-risk families will continue to be important to reveal biologically relevant coding variants for these cancers, both for insertion/deletion variants that were not covered by the current array, and to capture rarer variation (including private variants) that cannot be captured except by sequencing.
This work has been performed according to relevant national and international guidelines. Written consent was obtained at the time of DNA sample collection. The Institutional Review Boards at the University of Southern California and University of Hawaii approved of the study protocol.
The MEC consists of more than 215,000 men and women in California and Hawaii aged 45–75 at recruitment, and comprises mainly five self-reported racial/ethnic populations: African Americans, Japanese, Latinos, Native Hawaiians, and European Americans [45]. Between 1993 and 1996, adults enrolled in the study by completing a 26-page mailed questionnaire asking detailed information about demographic factors, personal behaviors, and prior medical conditions. Potential participants were identified through driver's license files from Departments of Motor Vehicles, voter registration lists, and Health Care Financing Administration data files. Incident breast and prostate cancer, as well as stage and hormone receptor status was identified by linkage of the cohort to the Surveillance, Epidemiology, and End Results cancer registries covering Hawaii and California. Between 1995 and 2006, blood specimens were collected prospectively from ∼67,000 participants for genetic and biomarker analyses. Currently, the breast cancer case-control study nested in the MEC includes 3,141 women diagnosed with invasive breast cancer and 3,721 frequency-matched controls without breast cancer, matched by race/ethnicity and age (in 5-year age categories). The case-control study of prostate cancer includes 4,675 men diagnosed with incident prostate cancer and 4,300 male controls without prostate cancer. The Institutional Review Boards at the University of Southern California and University of Hawaii approved of the study protocol.
Genotyping of the Illumina Human Exome BeadChip (n = 247,895 SNPs) was conducted at the USC Genomics Core Laboratory.
DNA extraction of buffy coat fractions was conducted using the Qiagen protocol. Cases and controls were randomly placed across ethnic-specific plates for each cancer type. All samples had DNA concentrations >10 ng/ul. Initial genotype definitions were based on auto-clustering 6,404 samples across all populations which had call rate >0.99 (African American 1883, Japanese American 1823, Latino 1008, European American 1690) using the GenomeStudio software (V2011.1). Following genotype calling on all samples (>16,000), manual inspection was conducted of the following SNPs: 1) SNPs with call rate <0.98 (n = 3,317), 2) monomorphic SNPs with call rate <1 (n = ∼15,000), 3) SNPs with minor allele frequency between 0 and 0.001 and call rate <1 (n = ∼31,500), 4) SNPs with >1 replicate error based on sample duplicates (∼1,000, discussed below), 5) SNPs with apparent differences in minor alleles frequencies >15% across ethnic-specific 96 sample plates (n = 798), or other evidence of batch/plate effects on allele frequency (n = 18,188), 6) all mitochondrial SNPs and all SNPs on the X and Y chromosomes (n = 5,574), and 7) autosomal SNPs out of Hardy-Weinberg Equilibrium in more than one ethnic group with p value<0.001 and at least one ethnic group with p value<0.00001 (n = 827). During the inspections we in total inspected cluster plots for approximately 70,000 SNPs (counting overlapping SNPs in the categories above) and genotypes were manually edited for 27,506 SNPs.
Of the 15,837 samples described above genotyping was successful with call rates ≥98% for 15,573 samples; of these we removed 17 samples for which reported sex conflicted with assessment of X chromosome heterozygosity, and 651 samples based on relatedness. Relatedness was determined using the IBD calculation in plink [46], and we removed one of each estimated MZ twin, sibling, parent-offspring, half sibling, or first cousin pairs. In the analysis, we also removed SNPs with <98% call rates (n = 2,531). To assess genotyping reproducibility we included 338 replicate samples which passed genotyping QC; among these samples the concordance rate of heterozygote calls, number concordant/(number concordant+number discordant), was 99.6% or greater for all replicate samples (average 99.99%). The final analysis dataset included 245,339 SNPs genotyped on 2,984 breast cancer cases and 3,568 controls, and 4,376 prostate cancer cases and 3,977 controls.
We relied on documentation files obtained from the University of Michigan posted on ftp://share.sph.umich.edu/exomeChip/IlluminaDesigns/ for the assessment of SNP type (i.e. NS, SP), and the amino acid affected. The array also includes SNPs that do not code for protein changes including synonymous SNPs, and other intergenic SNPs including ancestry informative markers, and GWAS identified risk SNPs for a number of diseases and outcomes. All SNPs were analyzed and their results shown in Tables S1, S2, S3, S4, S5, S6, S7, S8, S9. However our primary analysis focused on the 191,032 putative functional variants in the following categories (NS, SP and stop gain or loss) that passed quality control procedures discussed above.
We estimated principal components in the entire sample using EIGENSTRAT [47] based on 2,887 autosomal ancestry informative markers on the array. We adjusted for the top 10 principal components in all analyses.
Recognizing that many variants are only polymorphic in a few racial/ethnic groups, we give power analysis for a study with 1,000 cases and 2,000 controls (roughly the number of cases and controls in each of the four largest populations) by odds ratio (1–200) and allele frequencies ranging from 0.0001 to 0.1 (Figure S2). The Bonferroni criteria for significance in this study is calculated to be 0.05 divided by the largest number of polymorphic SNPs in any population (African Americans, ∼125,000) or roughly 3.9×10−7. For the gene burden analysis the Bonferroni criteria is 0.05 divided by the number of genes considered or roughly 3×10−6. We had 80% power to detect odds ratios of 3.3 or above for SNPs with a frequency of 0.01 and odds ratios in the range 13 or above for SNPs of frequency 0.001 in single SNP analyses. Power for the gene burden analysis depends upon the number of polymorphic SNPs in a given gene. Using a Poisson approximation (i.e. with variance assumed to be equal to the mean) a gene with 10 variants each of frequency 0.001 gives power of 80% to detect a per minor allele OR of 3.1. For genes with many more variants (100) of the same frequency detectable ORs per minor allele are 1.6 or greater. For common variants present in all ethnic groups we had much greater power to detect associations, for example we had 80% power to detect a 20% allele with an OR of 1.24 in the global analyses; for the region-specific analyses we have 80% power to detect a 20% allele with an OR of 1.17 in a region with 100 variants and 1.14 in a region with 10 variants.
|
10.1371/journal.ppat.1007238 | Diverse pathways of escape from all well-characterized VRC01-class broadly neutralizing HIV-1 antibodies | Many broadly neutralizing antibodies (bNAbs) against human immunodeficiency virus type 1 (HIV-1) were shown effective in animal models, and are currently evaluated in clinical trials. However, use of these antibodies in humans is hampered by the rapid emergence of resistant viruses. Here we show that soft-randomization can be used to accelerate the parallel identification of viral escape pathways. As a proof of principle, we soft-randomized the epitope regions of VRC01-class bNAbs in replication-competent HIV-1 and selected for resistant variants. After only a few passages, a surprisingly diverse population of antibody-resistant viruses emerged, bearing both novel and previously described escape mutations. We observed that the escape variants resistant to some VRC01-class bNAbs are resistant to most other bNAbs in the same class, and that a subset of variants was completely resistant to every well characterized VRC01-class bNAB, including VRC01, NIH45-46, 3BNC117, VRC07, N6, VRC-CH31, and VRC-PG04. Thus, our data demonstrate that soft randomization is a suitable approach for accelerated detection of viral escape, and highlight the challenges inherent in administering or attempting to elicit VRC01-class antibodies.
| Several potent antibodies against human immunodeficiency virus type 1 (HIV-1) have been evaluated in clinical trials. Use of these antibodies in humans, however, is problematic, because easy viral escape remains a major concern. To gain greater insights, we sought to develop an approach to rapidly assess the likelihood of viral escape from such antibodies. We show here that soft-randomization mutagenesis is a suitable approach to introduce a controlled number of changes into defined target regions. As a proof of concept, we used this approach to detect the HIV-1 variants fully resistant to VRC01-class of antibodies. We observed that within a few passages of the soft-randomized library of viruses in the presence of potent HIV-1 antibodies, a remarkably wide array of variants emerged, including variants resistant to every VRC01-class antibody. This study provides insights into a wide range of escape pathways, and describes a method for rapidly assessing the likelihood of viral escape from antibodies or small molecules targeting the HIV-1 envelope glycoprotein.
| A large number of potent broadly neutralizing antibodies (bNAbs) have been generated from HIV-1-infected individuals (reviewed in [1–5]). Many of these bNAbs, including 2F5, 4E10, PGT121, VRC01, 3BNC117 and 10–1074, have been or will be evaluated in clinical trials [6–12]. A number of animal studies have shown that administration of bNAbs can prevent infection, and reduce viral loads in an established infection [13–22]. Previous human studies also show that bNAbs can reduce viral loads or delay viral rebound upon treatment interruption [6–8, 11, 12, 23]. However, HIV-1 generally escapes these bNAbs when they are administered to infected animals or humans [6–8, 11–15, 17, 18, 20, 21, 23–26].
Escape pathways of HIV-1 from inhibitors against protease and reverse transcriptase have been comprehensively documented through large-scale studies of patients treated with these inhibitors. In contrast, escape from neutralizing antibodies has not been similarly described, in part because no antibody has yet been tested in a sufficiently large cohort of HIV-1 positive individuals. Most reported in vivo escape variants emerged from a limited number of animals or volunteers who participated in clinical trials [6–8, 11, 13, 20, 24–27]. In addition, although most bNAbs have been characterized with large panels of HIV-1 isolates [28–30], these studies do not provide direct insight into the likelihood of viral escape, because most HIV-1 isolates have never been exposed to or selected by these exceptional and rare antibodies. In some cases, the number of avenues to escape may be small because of structural and functional constraints on conserved epitopes of the envelope glycoproteins (Env). Nevertheless, to use bNAbs in humans or to design vaccines based on their epitopes, it is essential to understand how viruses can escape them. Although such information is now available for some antibodies from clinical trials, and other mutations have been observed in animal models or selected in cell culture, only a limited number of escape variants have been described for any given antibody or combination of antibodies. This small number is a consequence of inefficient process by which resistant variants emerge. This process relies on errors naturally generated by the viral reverse transcriptase, which introduces approximately one mutation per newly integrated provirus. Because these mutations are distributed throughout the viral genome, fewer than 1 in 10 of such mutations affect the env gene, the target of all bNAbs. Typically, the emergence of resistant variants is much faster in vivo than in vitro—escape variants usually emerge within a few weeks after antibody infusion [6–8, 11, 13, 23–26]—likely owing to much higher number and diversity of viruses found in each individual compared to that used in in vitro experiments. Nonetheless, only a limited number of escape pathways were detected in vivo, possibly because once a resistant virus emerges, it soon dominates the viral swarm, eliminating the selection pressure that would otherwise lead to additional resistant mutations. Therefore, to accumulate a meaningful number of escape mutations, this process needs to be repeated great many times or in many individuals or animals.
Several mutagenesis methods have also been adopted to accelerate viral evolution. Of these, the mutation rate of error-prone PCR can be difficult to control, and alanine-scanning mutagenesis is limited to mutating residues only to alanine, precluding other substitutions or combinations of substitutions available to the virus. A library constructed by codon mutagenesis, Exceedingly Meticulous and Parallel Investigation of Randomized Individual Codons (EMPIRIC) and Single-site saturation mutagenesis (SSM) were used to accelerate the evolution of HIV-1 or influenza A virus [31–33]. These methods are comprehensive, but labor intensive and only one mutation is introduced per copy of genome. On the other hand, random mutagenesis would generate high number of mutations, but in an overwhelming number of cases, the resulting Envs would not fold or function properly. Another mutagenesis technique, soft randomization, originally developed for phage display applications, allows generation of libraries whose members have a small, desired number of mutations distributed throughout a target region [34, 35]. This technique permits it by using primers synthesized with nucleotides mixed at a non-equimolar ratio, favoring original nucleotides [36]. The optimal composition and use of a hand mix ratio depend on the number of target codons to be altered, the library size, and how wobble codons are addressed.
Here, we developed a method that can help anticipate a range of viral escape pathways from antibodies with a well-defined epitope. To assess its usefulness, as a proof of concept, we tested this method against the CD4-binding site (CD4bs) bNAbs. We generated a replication-competent HIV-1 library expressing Envs diversified by soft randomization in the CD4bs, and selected the escape variants that were resistant to neutralization by VRC01-class CD4bs antibodies. Soft randomization introduced a controlled number of mutations into the majority of library members, and a large number of antibody-resistant isolates could be identified after only a few passages. Most randomly chosen clones of escape variants were resistant to all well-characterized VRC01-class antibodies, including N6, an exceptionally broad antibody of this class [37], in addition to the antibodies used in the selection. Because the epitope regions of CD4bs antibodies overlap with the CD4 binding sites, the resistant variants exhibit growth kinetics slower than that of the parental virus, as previously shown, but nonetheless they readily outgrew the parental virus at antibody concentrations commonly targeted in clinical trial [8, 10, 38]. Collectively, our data demonstrate that soft randomization can be usefully applied to identifying viral escape pathways in the face of a specific selection pressure.
To determine if soft randomization could generate a useful library of HIV-1 proviruses, we soft randomized the epitope regions of VRC01-class antibodies in the env gene, and selected the resulting virus library with the VRC01-class bNAb NIH45-46. We chose NIH45-46 because it was isolated from the same patient from which VRC01 was derived, but it is more potent than VRC01 [39], and because it has been evaluated in a clinical trial. As shown in Fig 1A, the binding epitopes of NIH45-46 are primarily located in the four regions of Env (Los Alamos National Laboratory (LANL) database CATNAP): Loop D, CD4 binding loop, bridging sheet and variable region 5 (V5). To minimize interference with CD4 binding, only the residues in Loop D, V5 and in their vicinity were randomized, whereas the CD4-binding loop and the bridging sheet were left unmodified. The compositions of the PCR primers used to build our library are shown in Fig 1B, and based on the sequence of ADA, a well-characterized clade B R5 isolate. Soft-randomizing primers were synthesized by handmixing original nucleotide and the other three nucleotides at a selected ratio for the first two positions of each codon. To incorporate a small number (1 to 3) of mutations per region, we use a handmixing ratio of 88:4:4:4 for the 11 amino acids of Loop D, and 91:3:3:3 for the longer V5 region (17 residues). An equimolar mix of G and T for the wobble positions of all amino acids encoded by 4 or 6 codons in Loop D, and an equimolar mix of C and A was used for V5 regions, because the soft-randomizing primer for V5 is anti-sense. As shown in S1 Fig, Loop D and V5 libraries were separately generated by PCR amplification of the entire pBR322 plasmid containing ADA env gene, pBR322-Env(ADA), as a template, and ligated using 5’ phosphate in the soft- randomizing primers. V5 library was subcloned into the plasmid containing Loop D library using an enzyme site engineered between Loop D and V5 regions. The resulting env library containing Loop D and V5 regions was then subcloned into the pNL4-3 proviral plasmid carrying ADA env gene, pNL-ADA, yielding a proviral library encoding diverse Env proteins, pNL-ADA-Lib.
An NL-ADA-Lib virus population (described here as a “swarm”) was then produced by transfecting HEK-293T cells with the pNL-ADA-Lib plasmid. To assess the composition of the library, a fragment encompassing both Loop D and V5 was amplified by RT-PCR and sequenced by paired-end Illumina MiSeq, and the total number of mutations in Loop D and V5 regions of each library member was analyzed. Only 21% of the total reads contained either stop codons or unknown amino acids, fewer than what is typically observed with less controlled mutagenesis approaches. These reads were excluded before analysis. The most common number of amino-acid substitutions in Loop D and V5 were 1 to 2 and 2 to 4, respectively, typically yielding 3 to 6 substitutions combined (Fig 2A), consistent with our predicted values. The proportion of non-mutated members was relatively low: 15% for Loop D and 4% for V5. More than 74% of the library members contain 1–3 mutations in Loop D, and 69% contain 1–4 mutations in V5, indicating that the majority of NL-ADA-Lib members contain any number of substitutions between 2 and 7. The substitution profile of each soft-randomized residue is shown in Fig 2B. Substitutions of each residue were relatively evenly distributed among 19 amino acids, with substitution biases reflecting one or two shared nucleotides between mutated and parental codons. Thus, soft-randomization can introduce a high degree of controlled diversity in selected regions of a viral genome.
To validate that escape variants can be easily detected from the soft-randomized library of viruses, we passaged the library swarm in the presence of indicated concentration of NIH45-46 (S2A Fig). Virus-containing culture supernatants were harvested two days later (NIH45-46 passage 1 swarm), and this process was repeated four additional times at the indicated antibody concentrations (S2A Fig). The virus swarm from each passage was then assessed for their resistance against NIH45-46 in TZM-bl luciferase reporter cells (Fig 3A). Both the parental NL-ADA-Lib swarm and that passaged 5 times in GHOST cells in the absence of any antibody (control passage 5) were used as controls. We observed that a single passage already conferred high level of resistance, and five passages, taking only two days per passage, were sufficient to select near complete resistance. We next assessed the NIH45-46 passage 5 swarm for their resistance to other VRC01-class CD4bs antibodies (Fig 3B). VRC01 and VRC07 were isolated from the same individual from whom NIH45-46 was derived, but 3BNC117, VRC-CH31, VRC-PGV04, and VRC-PG20 were isolated from different donors. Average IC50 and IC80 values of these antibodies for various HIV-1 isolates are shown in S2B Fig. As shown in Fig 3B, NIH45-46 passage 5 swarm was resistant to all VRC01-class CD4bs antibodies assessed, with the exception of 3BNC117 and VRC07 to a limited degree.
To determine whether complete resistance against all VRC01-class CD4bs antibodies can be obtained, we grew NIH45-46 passage 5 swarm in the presence of 3BNC117. The resulting swarm was then passaged four additional times at the concentrations indicated in S2A Fig. To prevent de-selection of NIH45-46 resistant viruses, NIH45-46 was included in all passages. 3BNC117 passage 5 swarm was then assessed for its resistance to 3BNC117 and VRC07 (Fig 4A). Interestingly, although the virus has become completely resistant to 3BNC117 even at 30 times of its IC80, its sensitivity to VRC07 was not changed. We therefore passaged this swarm in the presence of VRC07. The resulting VRC07 passage 5 swarm was resistant to all the VRC01-class antibodies we tested (Fig 4B), including the antibody, N6, a recently isolated CD4bs bNAb with the highest breadth thus far described in the class. We further assessed the same swarm against non-VRC01-class CD4bs antibody, b12, and non-CD4bs antibodies 10–1074, 10E8 and PGDM1400 (Fig 4C). VRC07 passage 5 swarm was sensitive to all of these antibodies, while their vulnerability to PGDM1400 was modestly reduced, suggesting its epitopes might partially overlap with those of CD4bs antibodies. These data show that observed resistance was not due to a generalized resistance mechanism or any non-specific growth advantages, and indicate that escape variants can readily be detected from a soft-randomized library swarm.
To identify the sequence variation that contributes to escape phenotype, VRC07 passage 5 swarm was then deep sequenced in the Loop D and V5 regions and compared to the sequences of the parental library swarm or that passaged 15 times in GHOST cells in the absence of any antibody (control passage 15). The total number of variants containing a substitution at the indicated soft-randomized residue are shown in Fig 5A, and their specific substitutions in Fig 5B. Although the original library swarm had fewer mutations in Loop D than in V5 region (Fig 2), higher number of Loop D mutations was detected in the escape variants (Fig 5A). Some substitutions are enriched in the control swarm, as expected, likely because they provided growth advantages in this experimental setting. The residues found most frequently substituted in the escape virus in this study are N276, D279 and A281 in the Loop D, and N461 and S465 in the V5 regions, similar to previous observations [7, 11, 25, 27, 40]. These residues are indicated in the crystal structure of an Env trimer (Fig 5C).
To rank enriched sequences, copy number of each sequence found in VRC07 passage 5 swarm was normalized by the copy number of the same sequence detected in the control passage 15 swarm. To include in normalization the sequences that were detected in VRC07 passage 5 swarm but not in the control passage 15 swarm, a control copy number of 1.0 was added to all sequences. The sequences of the top 150 escape variants after normalization are listed in S3A Fig. The same sequences are presented in S3B Fig, with each residue marked with a different color to highlight their similarities and substitution patterns. To confirm the resistance of the escape variants to CD4bs antibodies, we first chose six clones (1, 5, 10, 15, 20 and 25) from the top 25. The sequences of these clones—771 bp fragments containing only the mutations found in the Loop D and V5—were synthesized and cloned first into the pBR322-env (ADA) using engineered NcoI and MluI sites (S1 Fig) and then into the pNL-ADA plasmid. Replication-competent clonal viruses were produced from 293T cells by transfecting corresponding plasmids. When assessed with the three antibodies they were selected against, we observed that all these virus clones were completely resistant to them (Fig 6A, lower panel). We also characterized several additional clones from the top 150 (S3 Fig), which contained a small number (2 or 3) of substitutions (Fig 6B, top panel). As shown in the lower panel of Fig 6B, most of these clones were also resistant to all three antibodies they were selected against. These results suggest that many, if not most, of the top 150 escape variants are resistant to all three antibodies. These clones were then tested against additional VRC01-class antibodies: N6, VRC-CH31, VRC-PG04 and VRC01. All clones except clone 1 were resistant to all of these antibodies (Fig 6C). Clone 1 was sensitive to N6 but resistant to all other antibodies. Of note, clone 142, bearing only two substitutions (D279G, N280Y), was resistant to all VRC01-class antibodies tested, including N6. Although D279G was not sufficient by itself to confer resistance to VRC07 (S4A Fig), more than half of known HIV-1 isolates contain an amino acid other than D at the residue 279 (Fig 7A and LANL sequence database), indicating high flexibility at this position.
Whereas none of the top 150 clones gained an N-linked glycan, a majority (136) has lost the well-characterized N-linked glycan at N276 in Loop D through a substitution either at N276 and/or at T278, and 66 clones have substitution either at N461 and/or at S463 in V5 region. The loss of a N276 glycan is consistent with the reports that this glycan is required for virus neutralization by VRC01-class antibodies [22, 29, 41]. While it was previously detected in in vivo escape variants [25, 27], most natural isolates maintain this glycan (Fig 7A and LANL sequence database). When examined, however, loss of the N-glycan at residue 276 alone did not confer resistance to VRC07 (S4B Fig). This mutation was also not necessary for resistance to other VRC01-class antibodies; indeed, three clones (25, 14 and 142) we tested retained the N-glycan at N276, but they were nonetheless resistant to all VRC01-class antibodies (Fig 6).
In addition to the aforementioned mutations at the residues 276, 279 and 461, several other mutations identified in our top 150 escape variants are found in naturally-occurring isolates (LANL sequence database), including L277, S278, I283, S460, T/S/K at reside 461, and N/T at 462 (Fig 7A). Although the sequence variation among different parental swarms in different patients complicates analysis, a subset of the mutations found in our escape variants has also been previously reported in the clinical trials evaluating CD4bs bNAbs, VRC01 and 3BNC117 [6, 7, 11, 12]. These include K276, L277, S278, R282, S460, S/D at residue 461, N/T at residue 462, and R463 (Fig 7B). Thus, some of the mutations identified through our approach are found in infected humans. Collectively, our data demonstrate that there are abundant and diverse pathways through which HIV-1 can escape CD4bs antibodies.
Many of CD4bs bNAb escape variants were shown to exhibit fitness cost [42, 43], because changes in the antibody epitopes are likely alter virus binding to CD4. Typically compensatory mutation emerge that restore fitness while maintaining resistance. To assess fitness, we first measured the ability of WT and escape variants to use CD4 by measuring their neutralization sensitivity to CD4-Ig (Fig 8A–8C). Infection of TZM-bl cells by WT virus and clone 1 was similarly inhibited by CD4-Ig. However, CD4-Ig less potently neutralized other escape clones, indicating that they have reduced affinity for cellular CD4. Next, we assessed their fitness in the CD4+ T cells prepared by activating peripheral blood mononuclear cells with phytohemagglutinin-L (Fig 8D). Progeny virus production was measured by RT-qPCR in the culture supernatants of the infected cells every two days until day 8 post infection. The growth difference between WT and escape clones was substantial, except clone 1, especially in early time points, indicating the fitness of most escape variants was compromised. However, progeny virus production by clone 1 reached similar level as that of WT virus by day 4. In addition, as was shown in Fig 6, whereas replication of escape variants was not affected by the presence of VRC07, more than 99% of WT replication was inhibited. These data are consistent with the observations repeatedly made in the clinical trials evaluating CD4bs bNAbs [6, 7, 10–12, 23] that although generally less fit than WT virus, once selected, the escape variants outcompete WT virus in the presence of bNAbs.
Most bNAbs are able to neutralize majority of known HIV-1 isolates in vitro and prevent a new infection or control an established infection in animal studies. It is well established, however, that resistant viruses easily emerge in vitro and in vivo in the presence of these antibodies (reviewed in [1, 4]). While much effort is focused on using passively administered bNAbs and eliciting bNAbs through vaccination, less effort has been dedicated to understanding how viral escape will impact the utility of those approaches. Comprehensive insight into the ways HIV-1 can escape bNAbs, and methods by which this escape potential could be rapidly assessed, are critical to the use of bNAbs in humans. Without such insight, it is difficult to determine whether an antibody will be therapeutically useful, how it might be improved, whether it would work best in concert with other antibodies or antiviral drugs, or whether its epitope would be a useful target for a therapeutic or prophylactic vaccine. Most importantly, such information is necessary to determine whether the use of bNAbs in humans will easily promote emergence and spread of resistant variants.
Although there are a number of reports describing in vivo bNAb escape mutations, such studies have been limited to only a few antibodies, and by no means comprehensive. Whereas these in vivo escape variants emerge rapidly, within a few weeks after antibody infusion [6–8, 11, 13, 24–26], conventional in vitro escape studies typically take much longer. These studies are slow because viral reverse transcriptase introduces only approximately one nucleotide mutation per viral genome per replication cycle. Because a typical antibody epitope is encoded by 1–2% of the genome, it takes at least 50 replication cycles to introduce a single relevant mutation into a virus population. To accelerate this process, Dingens et al. recently adopted codon mutagenesis, a scanning mutagenesis-based approach [44]. This method generates a library of the env gene encoding every possible single amino-acid change in the Env ectodomain. Although comprehensive, this approach has several disadvantages relative to the soft-randomization approach we used here. First, the costs and labor of library generation are markedly greater; library generation with codon mutagenesis requires hundreds of primer pairs and corresponding numbers of PCR reactions, whereas soft randomization uses only a single pair of primers and one PCR reaction per target region. More critically, with codon mutagenesis, one cannot identify escape variants bearing more than one mutation. It therefore can only identify short pathways of escape unique to the specific Env under study. In contrast, as we show here, soft randomization can identify many escape variants with multiple changes, better reflecting the pathways available to a huge number of highly diverse viruses in circulation. In fact, most previously characterized bNAb escape variants identified in vivo bear multiple mutation [6–8, 11, 13, 24–26]. Nonetheless, our approach has one key disadvantage relative to codon mutagenesis, namely it requires structural knowledge of the epitopes of the bNAbs under investigation, and excludes escape pathways involving changes outside of those epitopes. In the case of HIV-1 Env and bNAbs, however, the amount of available structural and functional data largely compensates for this limitation. As both approaches and their limitations are complementary, in the future, they may be combined to provide maximum insight into viral escape.
Here we extended soft-randomization to HIV-1 and diversified the regions encoding a key antibody epitope, and showed that this approach could create a library of functional replication-competent HIV-1 proviruses with a controlled numbers of amino acid substitutions in each Env. Using well-characterized VRC01-class antibodies as an example, we further showed that a disturbingly high number of such substitutions facilitated viral escape from those antibodies. In addition, a subset of the escape mutations identified in this study was also detected in in vivo studies and in natural isolates. For example, among the five residues found most frequently substituted in this study (N276, D279, A281, N461 and N465), loss of glycosylation at N276 was also observed in escape variants derived from humans and humanized mice [25, 27, 45]. This glycan is highly conserved (in 92% of known HIV-1 isolates, HIV Sequence Compendium 2017 [46]), and its loss further exposes the CD4-binding site to humoral immunity. However, loss of this glycan among escape variants is not unexpected because it participates in the binding of VRC01 antibodies [29, 41, 45]. On the other hand, glycosylation at N461 is present only in 27% of known isolates, indicating that mutations can be readily accommodated at this position. Over 60% of HIV-1 isolates listed in the HIV Sequence Compendium 2017 [46] has residues other than D at the position 279, indicating that variations at this position would also not be difficult, as indicated by its rapid evolution in an infected individual [27]. The similarities between the escape variants from our study and those derived from humans or humanized mice validate the utility of a soft-randomized library approach for rapidly assessing viral escape from a bNAb.
We observed a very high number of escape pathways from VRC01-class antibodies in vitro. This large number of escape pathways is important and problematic for the use of these antibodies in humans, for several reasons. First, it suggests that any effort to “checkmate” the virus by using a cocktail of antibodies, in which viruses escaped from one antibody are designed neutralized by others of the same class, is unlikely to be successful. There are simply too many possible escape pathways to cover them all. Second, our data make clear that the current panels of Env used to assess antibody breadth and potency do not in any way encompass the possible ways through which HIV-1 responds to bNAbs. Occasionally breadth is discussed as a surrogate for difficulty of escape, but our data indicate that these concepts are dissociable. For example, N6 is among the broadest CD4bs bNAbs thus far described, but the virus appears to have many available pathways to escape it (Figs 4B and 6C). In fact, in some cases an antibody may be broad because it is rare, and thus its epitope has not been under pressure to diversify. Third, our data indicate that there is considerable overlap in the ways viruses can become resistant to different VRC01-class antibodies. For example, resistance to N6 readily emerged when viruses were selected against other VRC01-class members. Thus population-level escape from VRC01 may easily promote escape from 3BNC117 or N6. Finally, because many different sets of substitutions can resist all VRC01-class antibodies, engineered vaccines designed to mainly elicit VRC01-class antibodies [47, 48] may not sufficiently suppress population-level escape, and therefore may need to be supplemented by constructs designed to elicit bNAbs targeting complementary epitopes [49, 50].
Although, as a proof of principle, we focused here only on VRC01-class antibodies and ADA isolate, HIV-1 likely has a similarly wide range of pathways of escaping other bNAbs. Indeed, escape might pose greater challenges with other bNAbs such as V2-loop/apex antibodies and 332-glycan antibodies, whose epitopes are less conserved and less functionally important than the CD4 binding site. One possible exception may be the antibodies recognizing the highly-conserved gp41 MPER epitope. A similar study with an MPER-region library may reveal greater constraints on escape than observed here for VRC01-class antibodies.
In summary, soft-randomization of the key epitopes of HIV-1 Env is a useful approach for rapidly and extensively identifying antibody-resistant viruses, and thereby provides important insights into the propensity of a bNAb to promote viral escape and the potential pathways of escape. Its first application to the CD4bs bNAbs in this study extends previous observations and shows that there are a large number of discrete pathways by which HIV-1 can escape all VRC01-class antibodies.
Human embryonic kidney (HEK)-293T cells were obtained from the American Type Culture Collection (ATCC, CRL-3216) and used to generate library and clonal viruses by transfection. The TZM-bl cells were obtained from NIH AIDS Reagent Program and used as an indicator cell line to measure the infectivity of various viruses [51]. Both cell lines were maintained in high-glucose Dulbecco’s minimal essential medium (DMEM) containing 10% fetal bovine serum (FBS). GHOST cells are derived from the human osteosarcoma cells line, HOS. GHOST (3) CCR3+CXCR4+CCR5+ cells were obtained from NIH AIDS Reagent Program and used to passage virus library in the presence of antibodies, and were maintained in DMEM containing 10% FBS, 500 μg/ml G418, 100 μg/ml hygromycin, and 1 μg/ml puromycin [52]. In later text, this media is referred as GHOST-cell complete media, and GHOST (3) CCR3+CXCR4+CCR5+ cells are referred as GHOST-R3/X4/R5. All cells were grown at 37°C under 5% CO2.
The gene for the envelope glycoprotein (Env) of HIV-1 ADA (GenBank AY426119.1) was cloned into pBR322 using SalI and BamHI sites, and used as a template to generate soft-randomized libraries. An AscI site was engineered between Loop D and V5 regions at amino acid positions 309–311 in order to combine independently generated Loop D and V5 libraries. The Env library from pBR322 was cloned into pNL4-3 proviral plasmid containing the env gene from ADA isolate (pNL-ADA), using SalI and BamHI sites, to generate a replication-competent virus library. Loop D and V5 fragments containing the escape mutations identified by deep sequencing were synthesized by Integrated DNA Technologies (IDT) and cloned into pBR322, and then into pNL-ADA. Plasmids encoding antibodies and CD4-Ig are described in Antibodies section.
The broadly-neutralizing antibodies (bNAbs) used in this study are VRC01, 3BNC117, NIH45-46, VRC07, VRC-PG04, VRC-PG20, VRC-CH31, b12, N6, 10E8, 10–1074, PGDM1400. Of these, VRC01, 3BNC117, NIH45-46, VRC07, VRC-PG04, VRC-PG20, VRC-CH31, b12, N6 are CD4 binding-site (CD4bs) antibodies, 10E8 binds to the gp41 MPER epitope, 10–1074 binds a glycan on the V3 loop, and PGDM1400 recognizes an Env oligomer. IC80 and IC50 of these antibodies for ADA were obtained from the LANL database CATNAP (http://hiv.lanl.gov/catnap) [53], and are provided in S2B Fig. The expressor plasmids for antibodies NIH45-46, 3BNC117 and 10–1074 were kindly provided by Michael Nussenzweig (The Rockfeller University). Those for VRC01 and 10E8 were obtained from AIDS Reagent Program, and the rest were constructed by cloning the genes for the heavy- and light-chain variable regions synthesized by IDT into the plasmids encoding the constant regions of the light chain and human heavy chain of IgG1, as previously described [54]. GenBank numbers for the synthesized heavy and light chains are: VRC07 (H, KT365998.1; L, KM408147.1); VR-PG04 (H, JN159464.1; L, JN159466.1); VRC-PG20 (H, KF515514.1; L, KF515513.1); VRC-CH31 (H, JN159435.1; L, JN159438.1); N6 (H, KX595108; L, KX595112); b12 (H, AAB26315.1; L, AAB26306.1); PGDM1400 (H, KP006370.1; L, KP006383.1). All antibodies were produced in Expi293 Expression Medium (Life Technologies) by transfecting Expi293 cells with the corresponding expressor plasmids. Cells were grown for three days at 37°C with humidified air containing 8% CO2 on an orbital shaker platform rotating at 125 rpm. Proteins were purified from the culture supernatants using Protein A-Sepharose beads.
The protocol for soft randomization was previously described [36]. In brief, primers were designed for Loop D (amino acids 274–284 by HXB2 numbering) and V5 (amino acids 455–471) regions of ADA env gene. The soft-randomizing primer for Loop D was synthesized by hand-mixing 88% of the original nucleotide and 4% each of the other three nucleotides (88:4:4:4) for the first two positions of each codon of the 11 residues. A ratio of 91:3:3:3 was used for the longer (17 residues) V5 region. For the wobble positions of the Loop D primer, an equimolar mix (50:50) of G and T was used for amino acids encoded by 4 or 6 codons. For the wobble positions of other codons, the ratios of 88:4:4:4 (for Loop D) or 91:3:3:3 (for V5) was used. Both the 5’ and 3’ ends of the primers were extended outside the soft-randomized regions to match the melting temperature of pairing primers. Soft-randomizing primers were phosphorylated at the 5’ ends. The sequences of the soft-randomizing primers and pairing primers are shown in Fig 1B.
The Loop D and V5 soft randomization was performed independently via whole-plasmid PCR of pBR322 carrying SalI-BamHI fragment of ADA env gene, using Q5 HotStart DNA Polymerase (New England Biolabs). 1 μg of DpnI-digested and purified PCR product was ligated in 40 μl with 0.1 μl of concentrated T4 ligase (New England Biolabs) at 16°C overnight. 1 μg of ethanol-precipitated ligate was electroporated at 1700V into 25 μl of Electrocompetent NEB 10-beta (New England Biolabs), and the culture was grown in S.O.C. media with shaking at room temperature for 1.5 hours. The culture was then added to 300 mL Lennox broth containing 50 μg /ml ampicillin and grown with shaking until the culture reached an OD600 of 1.7. Chloramphenicol was added to final 170 μg/mL and the culture was grown for additional 36 h. To combine Loop D and V5 libraries, V5 library was subcloned into pBR322-Env (ADA) containing Loop D library, using AscI, engineered between Loop D and V5 regions, and BamHI sites. Ligation, DNA precipitation, electroporation, and plasmid preparation were performed as described above. pNL-ADA library was then constructed by subcloning the SalI-BamHI fragment of pBR322-Env(ADA) containing soft-randomized Loop D and V5 regions. Ligation, DNA precipitation, electroporation and plasmid preparation were again performed as described above. A total of 24 x 300 ml cultures, each 300 ml of which was derived from 1 ug ligate electroporated into 25 μl of NEB 10-beta, were grown to an OD600 0.8–1.0. DNA was prepared as described above, pooled and used as the final pNL-ADA-Lib.
WT pNL-ADA and its library, and all clonal viruses (sequences are shown in S3 Fig) were produced by calcium phosphate transfection of HEK-293T cells with corresponding plasmids, and cells were grown in DMEM containing 10% FBS. The virus-containing supernatants were harvested 48 hours post infection (hpi), aliquoted, and stored at -80°C. All experiments involving replication-competent viruses were performed in biosafety level 3 laboratory following the protocols approved by the Institutional Biosafety Committee of The Scripps Research Institute.
Library swarms, produced by transient transfection of HEK-293T described above, or obtained from a previous passage were incubated in GHOST-cell complete media for 30 minutes with indicated concentrations of NIH45-46, 3BNC117 or VRC07 antibody. Antibody concentrations used for selection are provided in S2A Fig for each antibody and passage. For this study, we chose virus dilution that yielded 60–70% infection of GHOST cells at 48–72 hpi. The virus-antibody mixture was then added to GHOST-R3/X4/R5 cells pre-seeded at ~40% confluence in 10 x T75 flasks. After 6–8 h, cells were washed twice with Phosphate-Buffered Saline (PBS) and further incubated in GHOST-cell complete media containing the same concentration of an antibody used for selection. The Tat-regulated GFP expression in GHOST cells was used to assess virus infection levels. When infection level reaches 60–70%, the virus-containing culture supernatants were harvested, spun to remove cell debris, aliquoted, stored in the -80°C, and used for subsequent viral passages or RNA extraction for deep sequencing.
RNA was extracted from 250 ul of cell-free culture supernatants of library or antibody-escaped swarms, using RNAqueous Total RNA Isolation kit (Ambion, ThermoFisher Scientific). cDNA was synthesized using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) using a gene-specific primer (5’ GAACCCAAGGAACATAGCTCCTATC 3’). An env fragment encompassing Loop D and V5 regions was amplified via 18 cycles of PCR reaction using Takara Taq Hot-Start DNA polymerase (Takara Bio Inc.) and ADA-Env-F3 (5’ GGCAGTCTAGCAGAAGAAGAGGTAGTAATTAG 3’) and ADA-Env-B3 (5’ CACTTCTCCAATTGTCCCTCATATCTCCTC 3’) primers. The PCR products (amplicons) were purified by AMPure XP beads (Beckman Coulter) at a bead to sample ratio of 3:2, quantified using Qubit dsDNA HS (high sensitivity) assay kit (Life Technologies) and run on the DNA high sensitivity chip in the Bioanalyzer 2100 (Agilent) to confirm their length. 100ng of these amplicons were then end-repaired, 5’ phosphorylated, and A-tailed at the 3’ ends using TruSeq Nano kit (FC-121-4001, Illumina). Next, these amplicons were ligated with illumina barcoded-partial adapters, size selected using AMPure XP beads, and further amplified for 4 PCR cycles. The final products were validated on the Bioanalyzer 2100 and quantified using Qubit dsDNA HS assay, pooled at equimolar and sequenced by paired-end 150bp reads on the Illumina MiSeq at the Scripps Genomics Core Facility in La Jolla, California.
The reads were trimmed of adapters, and cropped to isolate Loop D and V5 fragments. These sequences were translated, filtered for open-reading frames by eliminating those with stop codons and unidentifiable amino acids, and analyzed using in-house python scripts. To avoid removal of rare but intended mutations, reads were not corrected for the low level of errors that are potentially introduced during reverse transcription, PCR and sequencing. Enrichment was determined by normalizing the copy number of each sequence found in VRC07 passage 5 swarm by the copy number of the same sequence detected in the control passage 15 swarm. To include the sequences in normalization, which were detected in VRC07 passage 5 swarm but not in the control swarm, a control copy number of 1 is added to all sequences.
Neutralization assays were performed in TZM-bl cells, as described previously [51]. To use TZM-bl cells, antibiotics were removed from virus stocks by growing them in GHOST cells in the absence of antibiotics. This process also removed antibodies present in the culture supernatants. Virus-containing supernatants were harvested at 48–72 hpi, spun to remove debris, aliquoted, and used for neutralization assays. Viruses were incubated with 0–10 ug/ml of antibodies at room temperature for 30 minutes and then added to TZM-bl cells plated at 10,000 cells per 96 well one day before the assay. At 48 hpi, infection levels were measured by luciferase assays using Luc-Pair Firefly Luciferase HS assay kit (GeneCopoeia). The relative light unit was read at 575 nm using a Victor X3 plate reader (PerkinElmer).
To assess virus fitness, the growth of WT virus or virus clones that are resistant to bNAbs was assessed in CD4+ T cells. Cryopreserved human peripheral blood mononuclear cells (PBMC, StemCell Technologies), were thawed and incubated in RPMI containing 15% FBS and 20 U/ml human IL-2 (Roche) for overnight. Next day, cells were enriched with CD4+ T cells population by negative selection using Human CD4+ T Cell Isolation kit (Biolegend) and activated with 1 μg/ml PHA-L (Sigma) for 48 hours in RPMI supplemented with 15% FBS, 20 U/ml IL-2 at 1 x 106 cells/ml. Viruses (5 x 108 genome copy number, quantified by RT-qPCR) were pre-incubated for 20 min at room temperature with or without 10 μg/ml VRC07 in 100 μl RPMI supplemented with 15% FBS and 20 U/ml IL-2, and added to 1.5 x 105 PHA-L-activated CD4+ T cells in 150 μl. After 6 h incubation at 37°C, cells were washed with PBS and resuspended in 500 μl fresh RPMI containing 15% FBS with or without 10 μg/ml VRC07. Every 2 days, 180 μl of supernatant was harvested and replaced with same amount of fresh media. RNA was extracted from 150 μl these supernatants using TRIzol LS (Ambion), and cDNA synthesized using High Capacity cDNA Reverse Transcription kit (Applied Biosystems) and an ADA Env-specific primer (5’-GAACCCAAGGAACATAGCTCCTATC-3’). Probe qPCR was performed using iTaq Universal Probes Super mix (Bio-Rad), ADA-Env-qPCR-sense (5’-CAAAGCCTAAAGCCATGTGTAAA-3’) and ADA-Env-qPCR-antisense (5’-CTCCTCTCATTCCCTCACTACTA-3’), primers, and ADA-Env-qPCR probe (5’-/56-FAM/CCCATCCTG/ZEN/TGTTACTTTAAATTGCACTGA/3IABkFQ/-3’) in CFX96 Touch Real-Time PCR Detection System (Bio-Rad).
The frequency of mutations in the Loop D and V5 regions is analyzed using the AnalizeAlign tool available at LANL (www.hiv.lanl.gov). The sequences included in the analyses are: 5471 natural isolates available at LANL database, 298 Loop D and 250 V5 sequences of in vivo escape variants previously identified in the clinical trials for CD4bs antibodies, VRC01 and 3BNC117 [6, 7, 11, 12], and the top 150 escape variants identified in this study. Because of the large number of parental sequences in in vivo studies, parental residues are excluded from the analyses of in vivo escape mutations. Insertion mutations found in some of in vivo variants are also excluded.
Statistical analysis of the data was performed using GraphPad Prism software. The difference between groups for all neutralizing assays and virus growth curves in CD4+ T cells were tested using a two-way ANOVA. The null hypothesis was rejected when p<0.05 in all cases.
|
10.1371/journal.pntd.0006514 | The human-snail transmission environment shapes long term schistosomiasis control outcomes: Implications for improving the accuracy of predictive modeling | Schistosomiasis is a chronic parasitic trematode disease that affects over 240 million people worldwide. The Schistosoma lifecycle is complex, involving transmission via specific intermediate-host freshwater snails. Predictive mathematical models of Schistosoma transmission have often chosen to simplify or ignore the details of environmental human-snail interaction in their analyses. Schistosome transmission models now aim to provide better precision for policy planning of elimination of transmission. This heightens the importance of including the environmental complexity of vector-pathogen interaction in order to make more accurate projections.
We propose a nonlinear snail force of infection (FOI) that takes into account an intermediate larval stage (miracidium) and snail biology. We focused, in particular, on the effects of snail force of infection (FOI) on the impact of mass drug administration (MDA) in human communities. The proposed (modified) model was compared to a conventional model in terms of their predictions. A longitudinal dataset generated in Kenya field studies was used for model calibration and validation. For each sample community, we calibrated modified and conventional model systems, then used them to model outcomes for a range of MDA regimens. In most cases, the modified model predicted more vigorous post-MDA rebound, with faster relapse to baseline levels of infection. The effect was pronounced in higher risk communities. When compared to observed data, only the modified system was able to successfully predict persistent rebound of Schistosoma infection.
The observed impact of varying location-specific snail inputs sheds light on the diverse MDA response patterns noted in operational research on schistosomiasis control, such as the recent SCORE project. Efficiency of human-to-snail transmission is likely to be much higher than predicted by standard models, which, in practice, will make local elimination by implementation of MDA alone highly unlikely, even over a multi-decade period.
| Infection with blood fluke Schistosoma parasites is a major cause of disease burden around the world. Control of schistosomiasis, which is transmitted through intermediate host freshwater snails, is a priority for national and global health programs working in at-risk regions of Africa, the Mideast, Asia, the Philippines, and South America. Program planning often relies on mathematical models to project the impact of different schedules of mass drug administration (MDA) of the anti-schistosomal drug, praziquantel, in these areas. In practice, though, recent projections of standard models have failed to capture the variability of MDA program impact on community levels of infection, especially in high-risk zones. In the present study, we developed a modification of the conventional modeling approach that takes more detailed account of human-to-snail transmission. Inclusion of a revised, nonlinear form for the model’s snail infection function had profound effects on long term predictions of the impact of MDA programs for Schistosoma control. In specific, our proposed snail parameters helped to explain the persistent rebound of Schistosoma prevalence in certain high risk communities. The efficiency of human-to-snail transmission is likely to be much higher than predicted in standard models, which makes local elimination by implementation of MDA, alone, highly unlikely.
| Schistosomiasis is a neglected tropical disease (NTD) having an estimated global prevalence of 240 million infected persons, many of whom experience significant morbidity within the infected communities of Africa, the Mideast, South America, Asia, and the Philippines [1]. For global control of the disease schistosomiasis, the World Health Organization (WHO) recommends delivery of the anti-helminthic drug, praziquantel, via mass drug administration (MDA), with attempts at local elimination, where possible [1, 2]. Unlike the very effective MDA experience obtained for other helminthic NTDs such as onchocerciasis and lymphatic filariasis [3, 4], there remain significant concerns about the feasibility of schistosomiasis elimination using MDA alone [5]. This is in part due to that fact that MDA has been unable to interrupt schistosomiasis transmission in many endemic areas, even after a decade or more of repeated MDA [6, 7]. This failure to interrupt transmission has often been marked by a significant rebound of infection prevalence following termination of MDA [8–10], or of concurrent mollusciciding interventions [11]. The highly uneven landscape distribution of suitable intermediate host snail habitat, combined with weather- and climate-related seasonal differences in snail abundance, mean that there is often a quite varied patchwork of transmission zones within any given region slated for parasite control [12–16].
Understanding the mechanisms that drive infection rebound is crucial for the development and implementation of more efficient control strategies [1]. Conventional predictive models of transmission suggest that where rebound is slow, there can be progressive reduction of parasite burden after each MDA cycle, so we may expect to bring Schistosoma burden under control and achieve elimination of transmission [17]. On the other hand, rapid rebound of parasite burden following treatment serves to impede long term progress towards elimination goals, and necessitates additional MDA effort and/or introduction of complementary environmental control measures to achieve parasite elimination [5, 18, 19].
In developing transmission models, important but often overlooked determinants of schistosome transmission are the ecology and population biology of the intermediate snail host and accurate assessment of the human-to-snail force of infection (FOI). As part of the transmission cycle, Schistosoma must infect very specific intermediate host snail species, then undergo a process of extensive asexual multiplication within the snail’s body in order to create the free-swimming cercariae that will infect the next round of human hosts [20, 21]. For Schistosoma parasites of humans, local presence of freshwater snail species of genera Bulinus, Biomphalaria, Oncomelania, or Neotricula, is essential to the transmission of Schistosoma haematobium, S. mansoni, S. japonicum, and S. mekongi, respectively [22]. Because snail infection is an obligate stage for parasite transmission, ecological factors that favor the presence and abundance of these ‘vector’ snails also foster local risk for these Schistosoma spp. infections and for their related human disease states, either urogenital or intestinal schistosomiasis [15, 20, 23].
Conventional transmission models assume the snails’ FOI is a linear function of human infectivity (see, e.g. [24–26]). Under this assumption, any drop in human infectivity (e.g. via MDA-related reduction in local egg excretion), will proportionately reduce the rate of local snail infections, which in turn will slow reinfection of human hosts. However, the empiric field data from recent large-scale, cluster- randomized operational research trials of anti-schistosomal MDA [27] have demonstrated a broad range of community-level parasitological responses, ranging from highly effective reductions in prevalence and intensity at some locations, to the existence of highly resistant “hotspots” (Fig 1), where infection levels persist at or near baseline levels despite effective implementation of MDA [28, 29]. While the current simplified deterministic models mimic the average effects of MDA across all communities, the failure to account for broad village-by-village variability is a challenge to the general utility of transmission model-based predictions.
Prior modeling studies of other micro-and macro-parasite systems have established that the form assumed for the transmission coefficient (beta) can have a significant impact on the projected outcomes for disease ecology models [30–32]. As a simplifying initiative, most modelling approaches typically assume ‘density-dependent’ kinetics, in which two well-mixed populations provide a constant per capita rate of exposure [33]. However, some models have elected to employ ‘frequency-dependent’ kinetics, which can exhibit saturation of transmission at higher host densities or in the presence of non-random mixing [30, 34, 35]. In such models, landscape patchiness, associative movement networks, time-dependence, and heterogeneity in host susceptibility can explain the failure of standard ‘mass action’ transmission coefficients to accurately capture the trajectory of disease transmission in real-world settings [30, 33, 36–38]. These transmission features are common among vector-borne macroparasites such as the Schistosoma species studied here. Where such features exist, it is apparent that fine-scale transmission events in linked territories can serve to drive larger meta-population patterns of infection prevalence [30].
To explain observed heterogeneities in Schistosoma transmission, we undertook a closer examination of the intermediate snail host and its infection by humans. Schistosoma transmission and parasite development have multiple time scales, ranging from “fast” larval dynamics (hours, days), to “slow” (month, years) host-parasite-snail dynamics. In the current study, we focused on these slower dynamics, so larval stages did not enter our model formulation explicitly. However, we saw the need to have an accurate account of their effect on human and snail infection. Conventional modeling approaches assume each FOI to be proportional to its source infectivity and population size [33]. We reexamined the conventional model assumptions, and derived a newer formulation of human-to-snail FOI that combines human host infectivity, demographics and snail population inputs. Among other salient features of the proposed FOI is its nonlinear dependence on human egg output. This functional form could be linked to the magnitude of the post-MDA prevalence rebound and to the consequent success or failure of long-term control.
We explored the effect of modifying snail FOI in simulating MDA responses for typical endemic communities, comparing “nonlinear” vs. “linear” models. The two models produced markedly different outcomes, particularly in high-intensity transmission settings.
The schistosome parasite maintains a complex life cycle, transiting between human and snail hosts, with the transition mediated by two larval stages, the egg-derived miracidium (for human-to snail movement), and the snail-derived cercaria (for snail-to-human movement) [21]. For this study, we applied a previously developed dynamic model that describes this biological process. We denote the corresponding forces of infection λ (for snail-to-human), and Λ (for human-to-snail). The former (λ) represents the mean rates of worm accumulation by human hosts, the latter (Λ), the mean rate of snail invasion by miracidia.
Each force depends on its host carrier’s infectivity, population abundances, and the frequency and pattern of their contact (human water exposure and water contamination rates). In our setup, human FOI is proportional to infected snail prevalence (0 < y < 1), λ = A y, with transmission coefficient A. Snail FOI is a function of human infectivity, E (mean egg release), but its derivation requires careful analysis. Most conventional models employ linear Λ = B E with transmission coefficient B [25, 26, 39]. Here, instead, we propose a nonlinear (saturable) form of snail FOI,
Λ=Λ0(1−e−bE)
(1)
The derivation of (1) is outlined in the supporting information S1 File. It employs some natural assumptions on miracidial dynamics from human egg release, its diffusive spread, and the process of snail invasion. We assumed miracidia randomly cluster about snail host, with a Poisson distributed “miracidia/snail” ratio. The resulting saturable (exponential) function (1) is the probability of successful invasion (see S1 File).
To study the effect of nonlinear FOI, we programmed two coupled human-snail model systems, termed M1 (having a linear snail FOI factor), and M2 (having a nonlinear FOI given by Eq (1)). Nonlinear Λ had two coefficients (Λ0, b) that, through local model calibration, reflected important local environmental, biological, and behavioral inputs. Λ0 can be viewed as maximal rate of miracidial invasion in a given environment. It depends on local snail density (which determines “mean travel time” to reach target), and on search strategies employed by miracidia (see [40] for a general discussion of encounter rates). Coefficient b is related to the mean miracidia production by human hosts and the probability of snail invasion by miracidia. Additional factors that enter b include mean population density (host/snail), and human-snail contact (exposure/contamination) rates.
Different types of human and snail models can be coupled via FOI terms λ,Λ. Here we adopted a stratified worm burden (SWB) approach (for the human part), developed in earlier works [41–44], but one can also use a simpler MacDonald-type mean worm burden (MWB) system [24, 45]. The basic differences between models M1 and M2, and their projected control outcomes, are due primarily to the Λ -function, whereas a specific formulation for the human side of the coupled model proved less influential.
Importantly, there can be a convergence between linear and nonlinear FOI systems: Function (1) can be approximated by a linear function
Λ(E)≈Λ0bE
(2)
at small contagion levels (i.e., b E ≪ 1). So our nonlinear Λ (1) can be viewed as an extension of linear form (2) to reflect larger values of human infectivity. Specifically, the M1 and M2 FOIs depart significantly as E or b grow large; the latter, in particular, embodies higher human-to-snail ratios or higher contact rates. Notably, the two FOI systems can also give markedly different values of transmission coefficients, even when calibrated against the same datasets.
For snail infection modeling, we used a standard simple S-I transmission system (x–susceptible (S), y–infected (I)) with stationary population density (x + y = 1). The prevalence variable 0 < y(t) < 1, solves differential equation
dydt=Λ(1−y)−νy,
(3)
with snail FOI, Λ.
For the present analysis, a human SWB model was used, consisting of variables h→(t)={hm(t)}- (worm burden strata) that undergo dynamic changes due to worm accumulation and loss processes. The detailed exposition of SWB approach has been described in detail in previous publications [39, 41, 42], and it is briefly summarized in S1 File.
A conventional MWB setup [39, 41, 42] can also be used if desired. It has a single dynamic variable, MWB w(t), that obeys differential equation
dwdt=λ−(γ+μ)w
(4)
with human FOI (λ = Ay) depending on snail prevalence (3), and loss term (γ + μ) which combines worm mortality, γ, and host turnover, μ. The two models, MWB and SWB, share common input parameters (λ,γ,μ). In fact, MWB Eq (4) follows from the SWB if one takes the first moment (mean) of the {hm}-distribution,
w(t)=∑m>0mhm(t).
The main difference between the SWB and MWB approaches lies in their assumptions on within-humans worm distribution patterns, {hm}, and the resulting human infectivity E (see, e.g. [41]). The SWB imposes no constraints on variables {hm}, whereas MWB uses a priori assumptions to express E as a function of w(t). Typically in the MWB model, {hm}are assumed to follow a negative binomial (NB) with prescribed aggregation constant, k.
In both systems, human infectivity is a product of mean mated worm count (MMC) Φ, and worm fecundity ρ, with E = ρΦ. The MWB gives MMC as a function of variable w, Φ(w,k), while SWB function Φ(h→) depends on worm burden strata h→={hm}.
Two different FOI {λ(y),Λ(E)} couple transmission dynamics between human and snail hosts, and give rise to a coupled SWB-snail model. The setup can be can be extended to demographically-structured populations made of multiple risk/age groups, each carrying specific burden distributions. In our analysis we employed structured host communities made of child (C) and adult (A) age groups, with age-specific FOI and transmission coefficients, λC = ACy, λa = Aay.
The combined infectivity of such system depends on MMC Φi of each group, their age-specific worm fecundities ρi, population fractions (Hc + Ha = 1), and contact (exposure/ contamination) rates ωi. Their combination gives the following dimensionless form
E=ρ(HcΦc+ωHaΦa)
(5)
Factor ρC is the mean worm fecundity of the child group, while weight ω is the product of relative (child-to-adult) fecundity and exposure factors (S1 File). The child age-group worm fecundity is subsumed as a factor in the transmission coefficient, b, so doesn’t enter the model explicitly.
Calibration of the coupled systems proceeded in two steps: (i) human egg-count (diagnostic test) data were employed to estimate snail-to-human FOI and worm fecundity (λi,ρi) for each human subgroup. The outcome was a best-fit posterior distribution of the model parameter space; (ii) next, the calibrated human parameters were combined with additional environmental/behavioral (snail) data to estimate transmission coefficients Ai (snail-to-human), and either {B,ω} (for linear FOI), or triplet {Λ0,b,ω} (for nonlinear, Λ; see S1 File, Part B for details).
In our predictions, we used similar snail inputs (baseline prevalence, y*) and the relative adult/child exposure factor, ω, in both model systems M1 and M2. However, nonlinear FOI (M2) had an additional parameter, b, which encoded the relative human/snail population factor (H/N). In our sensitivity analysis, we varied b to simulate a broad range of environments and explore its effect on MDA outcomes.
Drug treatment with praziquantel kills a large fraction of adult Schistosoma worms, and its clearing efficacy is estimated at 80–95% [5]. In our simulations, we have set this value at 85% (using a surviving worm fraction, ε = .15). The key inputs for MDA program simulation consisted of target group sizes (children, adults), their coverage levels (e.g. 0 < fc < 1, 0 < fA < 1), and the timing or frequency of MDA delivery (annual, biennial, etc.).
In our numeric simulations, MDA was implemented as an instantaneous event, whereby worm burden of each group is reduced depending on its coverage and drug efficacy, so the dynamical transmission system was reinitialized at time td after each control event. For the SWB system, an MDA event results in reshuffling of burden strata, so that each higher-burden stratum shifts to lower-burden strata hm → hεm (see [42, 43]). For a corresponding MacDonald-like MWB system, each MDA event with coverage f, and efficacy ε, would reduce MWB w(td) by a factor ε f + (1 − f).
For analysis and MDA simulations we modeled three communities from past Kenyan control-surveillance studies 1983–92 [46], and 2000–2009 [12, 13], having heavy (H), moderate (M) or light (L) infection levels (see Table 1).This dataset was extensively used in our previous SWB work [10], and in more recent papers [41–43, 47]. The latter have employed refined SWB methodology to account for in-host biology (worm mating, aggregation, random egg release), and have introduced more advanced calibration methodologies.
The modeled high-intensity community (H) was subject to longitudinal study spanning nine years, with two MDA sessions (in 2001 and 2003), and three population-wide surveillance screenings (in 2001, 2003, and 2009). For the purpose of the current comparative modeling analysis, we divided the village population into child (0–20 year old) and adult (20+ years) age groups (Table 2) based on Kenyan demographics. Additional model parameters included in the simulations were worm mortality and snail survival as described in S1 File, Table A1.
The two study models (M1, M2) were calibrated for each of our high-, moderate- and low-intensity sample communities following [42]. The calibration procedure involved two-steps: (i) individual egg-count test data at baseline (Year 2001) were employed to define a posterior distribution of likely parameter choices (λ,ρ,k) for age-groups C and A. The calibration results (marginal distributions of human parameters and their statistics) are described in S1 File, part B.
The next step used the estimated human parameters (from our first-stage calibration’s posterior distribution) to estimate transmission coefficients. Snail-to-human transmission coefficients Ai (i = C,A) were identical for M1 and M2. The human-to-snail components were different: {B,ω} for the linear-FOI model M1, and {Λ0,b,ω} for the nonlinear M2. All depended on infected snail prevalence (both prepatent and patent (i.e., cercaria-shedding)), which was fixed at value y* = 0.3, consistent with PCR-based snail surveillance findings in the Kenyan environment [48]. Patent snail density, which is responsible for transmission, was assumed to be proportional to infected snail prevalence, y(t). There were two additional inputs (y*,ω) for M1, and 3 additional inputs (y*,ω,b) for M2. The relative adult/child exposure ratio, ω, was set at 1.5, and b combined a transmission coefficient (miracidia contagion release the by the child age group) times relative host population abundance (human/snail) (see S1 File, part B). Because these values have been less well studied, in sensitivity analysis we allowed broad range of uncertainties: 0.5 < ω < 5; 0.5 < b < 5, for both of these transmission variables.
The calibrated model community, using a consistent choice of transmission uncertainties (y,ω,b), was subjected to a series of control experiments to explore the effect of snail FOI assumptions (model M1 vs. M2) and the role of (y,ω,b) on long term MDA outcome patterns in different environmental settings. A typical 10-year history for a high-risk community is shown in Fig 2. The model parameters used in this simulation are listed in Table 3. For this analysis, annual community MDA was used, with an estimated 75% annual coverage for children, and 35% biennial coverage was used for adults.
The simulation results show large differences between M1 and M2 projections, with the M1 system rapidly approaching elimination, whereas M2 becomes locked in a limit-cycle pattern and does not approach elimination (Fig 2). This qualitative distinction between the models—mainly that M2 model was considerably less likely to achieve MDA-mediated elimination—persisted for a range of parameter choices and MDA coverage. In sensitivity analysis of our prediction by random sampling of model parameters (human and environmental) over a broad range of values with identical M1 and M2 communities subjected to the same MDA regimen, significant differences remained in projected outcomes. History envelopes (Fig 3) show ensemble mean and 95% CI for the multiple simulated 10-year MDA programs. The M1 histories consistently go to elimination, while the M2 histories settle into recurrent limit cycles that fail to achieve elimination.
To help validate our approach, we used an observed longitudinal dataset collected over 9-year period for the base case high-risk community, Milalani, in Kwale County, Kenya [46, 49]. The community was screened in (2001, 2003, 2009), with two MDA sessions run in 2001 (community-wide coverage 79%), and in 2003 (community-wide coverage 41%). The results of study are summarized in Table 4.
To assess prediction potential of linear and nonlinear models, both systems were fitted to the baseline infection dataset (2001). As explained in Methods, this yields a posterior ensemble of best-fit calibrated human parameters (λ,k,ρ). We then sampled random choices from this posterior ensemble, along with three additional environmental inputs, (ω,b,y*), to get the estimated transmission parameters for M1 and M2 (see Table 3). Each virtual community (parameter choice) was simulated over a 9-year period subject to two MDAs. Typical model outcomes are shown in Fig 4, with comparison to observed field data. On both follow-up years (2003, 2009), we observed significant relapse toward pre-control (endemic) levels of infection. Of the two calibrated models, the nonlinear M2 was able to reproduce this pattern for child and adult groups. However, the M1 model did not capture post-treatment prevalence values with its slower intrinsic relapse rate.
We again tested parameter sensitivity for robustness of our predictions. This test was run separately for three environmental inputs: i) the relative exposure factor was varied in the range 0.5 < ω < 5, ii) the child transmission rate was varied in the range 0.5 < b < 5 (for M2), and iii) a random variation of best-fit panel parameter inputs (λi,ρi,ki) of the calibrated community was used in each replicate simulation. In each case, an ensemble of 9-year histories was simulated. Solution envelopes of these ensembles along with their mean path are plotted in Fig 5 (panels a, b, and c). The envelopes are less sensitive to relative exposure factor ω, but child transmission b had more pronounced effect. The uncertainties due to human inputs {(λi,ρi,ki)}, come from the baseline posterior calibration, as shown in panel (c) of Fig 5. In all cases, observed data points lie within prediction envelopes.
As discussed earlier in Methods, nonlinear snail FOI becomes approximately linear at low levels of human infectivity. To explore the effect of a reduced transmission environment on long term MDA, we subjected three sample communities with heavy (H), moderate (M) or light (L) transmission intensity, respectively, to the same 10-year control regimen, and compared projected prevalence outcomes for M1 vs. M2 simulations (infection prevalence). Fig 6 shows the comparative results. The difference in simulations is unambiguous for the high risk community, where M1 predicts gradual decline towards elimination, whereas M2 shows strong rebound to moderately high prevalence levels (15–30% for children) each year. For moderate risk areas, the two curves for M1 and M2 are closer, although M2 still predicts a persistent cycle of reinfection. For the low risk community (L) the discrepancy between models appears marginal, with M1 and M2 closely following each other.
In this modeling study, we systematically compared two model structures for Schistosoma transmission to better understand the importance of non-linear snail vector dynamics for model prediction of long-term intervention outcomes. We calibrated two transmission models with identical human host inputs but different human-to-snail transmission coupling—a conventional model with linear FOI assumption (M1) and a more complex model assuming a nonlinear saturable FOI for snails (M2)–using longitudinal data collected in coastal Kenya [10, 42]. We subjected both models to a series of numeric experiments simulating different MDA regimens, and found marked differences in long-term epidemiologic predictions. The conventional M1 model predicted efficient control (reaching targeted reductions, then elimination) after relatively few rounds of MDA, even in the face of low or moderate treatment coverage levels. The proposed M2 model, however, found many settings to be highly refractory to MDA treatment impact, with persistent Schistosoma re-infection even with high treatment coverage levels. In the model validation, we found the M2 model with its non-linear snail FOI formulation to be more reflective of empirically observed data [10, 29]. Going forward, these findings have clear implications for program monitoring and evaluation and future control implementation for schistosomiasis control, suggesting that a non-linear FOI function should be incorporated for more realistic projections in future Schistosoma transmission modeling.
Empirical evidence from other host-pathogen systems [33, 38, 50–53] suggest that there is likely to be a continuum in transmission kinetics that must be considered when modeling the observed transmission patterns found in settings where host numbers and distribution are varied. Although they are more complex and require more data, more nuanced modelling systems are expected to yield better understanding of parasite dynamics and the impact of control interventions [33]. Previous modeling work on S. japonicum transmission by Liang and colleagues [54] has incorporated multiple human risk groups identified by location and occupation, as well as seasonal aspects of snail reproduction and development. When calibrated against field data, this model more accurately projected the re-emergence of infection in high-risk communities when MDA and other interventions were stopped. Prediction of ‘bounce-back’ risk will be essential in determining the design of follow-up surveillance programs as local elimination is attempted. As noted above, the accurate calibration of such models requires more information about the control areas. However, the greater precision of model projections should improve the efficiency of program interventions [54].
In the presence of nonlinear FOI, a relatively small infective human host pool can exert a disproportionate, leveraged effect on snail infection. Hence, even a steep drop of human infectivity post MDA may result in only a marginal drop of snail infections, and this phenomenon, in turn, may result in a vigorous rebound or human infection to pre-treatment levels as noted in the SCORE project persistent hotspots [28, 29]. In our analysis, we have used independent longitudinal data from communities in rural Kenya to formally compare the proposed non-linear snail FOI models with more conventional models to understand the impact of this effect on long-term model prediction.
In our study system, the concept of a nonlinear, saturable pattern for snail FOI in Schistosoma transmission environments (as proposed in the M2 model) has biological plausibility: i) Water contamination occurs in pulses, as infected humans only intermittently contaminate their environment with urine or feces [55]. Human treatment coverage is non-random, with people who are non-adherent to MDA perhaps the most likely ones to contaminate the snail environment (i.e., as effective superspreaders); ii) the miracidia that hatch from contaminating eggs selectively home onto local vector snails in order to infect them [56], iii) because of substantial asexual reproduction of the sporocyst, each infected intermediate host snail has the potential to release thousands of infective cercariae [57, 58], and iv) cercariae sense human skin lipids, and preferentially swim toward any persons coming into contact with affected water bodies [59]. These nonlinear features all bias the transmission process in favor of higher levels of human infection and post-MDA reinfection. Specifically, this means that the extra-human phase of Schistosoma transmission is not a random, mass action process, although, for simplicity’s sake, many current models of transmission have assumed that it is.
The coupled human-snail transmission dynamics in a model of schistosomiasis transmission are driven by two FOI: human-to-snail (Λ), and snail-to-human (λ). Each FOI is dependent on its source population size and infectivity, and given the predictive limitations of conventional models, our findings suggest that future models should include an updated accounting of these parasite invasion processes. The two obligate trematode hosts (human and snail) are treated differently in mathematical models of schistosome transmission but their FOIs are often assumed to be linear functions of the combined host infectivity. While such an assumption appears justified for human FOI, λ, snail FOI Λ requires more careful elaboration. In our current analysis, we derived a nonlinear saturable snail FOI function, which embodied several essential environmental (e.g. type of water source, sanitation), demographic (e.g. age distribution), and behavioral inputs (e.g. contact with water, defecation practices), including human/snail population densities (H, N) and their contact/exposure rates. Given the difficulty of empirically measuring many of these aspects, we calibrated a composite estimate of FOI that reflected many complex and often heterogeneous factors. The conventional linear and proposed nonlinear functions were approximately equal at low levels of human contagion, where the linear FOI could be viewed as an adequate approximation of what is actually a nonlinear Λ. However, the two FOI versions diverged at higher levels of contagion, and so yielded very different transmission parameter estimates when fitted to the same human-snail infection data. The M1 and M2 models, based on the two different systems, also responded differently to strong perturbations, as occurs with MDA interventions; the M2 models predicting substantially faster post-MDA rebound as compared to M1 models. The human part of our present coupled system analysis employed SWB methodology [39, 41–43], but the qualitative conclusions of the M1-M2 comparison would remain true for other transmission models, including MacDonald-type MWB models [24, 45].
Only the nonlinear (M2) was able to accurately reproduce the strong rebound of infection seen in the dataset in years 3 and 9 of the Kenya project. This would predict that such communities will be resilient to any attempts at targeted elimination of transmission. In many cases the temporal differences between the two model systems (M1 & M2) were large, in that M1 community model projections typically achieved control targets over a short time-span with moderate effort, compared to M2 models, where infection was projected to persist much longer and to require extended treatment intervention. In a separate project, we have explored, in greater depth, possible elimination strategies using combined MDA and environmental snail control, and we predict that the only way to achieve target reduction in high transmission communities would be via implementation of additional environmental interventions, e.g. combining MDA with molluscicide-based snail control [11, 44].
For the nonlinear M2 system, three factors contribute independently to snail FOI estimation, accounting for a variety of MDA responses ranging from near-linear, efficient reduction /elimination in lower prevalence communities, to a highly resilient “locked” pattern of reinfection, whereby each MDA-mediated drop in prevalence is matched by post-treatment rebound. This latter feature could provide a key to the hotspot phenomenon observed in many control programs (see, e.g. [28, 29]). Indeed, it can explain why adjacent communities with near identical baseline human infection can produce divergent MDA responses based on variations in their local snail environment and in human behavior [60]. Importantly, while the proposed non-linear model demonstrates improved predictive value, this benefit should be balanced with the need for additional community data and more complex parameter estimation. The principle finding of this study is that a relatively simple non-linear function, on average, outperforms a linear function even when considering parameter uncertainty.
Our analysis suggests a defining role of transmission environment (and its resultant snail FOI) for predicting MDA control outcomes. The heterogeneity and connectedness across Schistosoma transmission landscapes [16, 45, 61], along with substantial parasite replication in the snail host, appear to make Schistosoma infection control much more challenging than for the filarial parasites that are transmitted by insect vectors [3, 4]. In particular, MDA-based ‘transmission control’ for schistosomes will be particularly fragile in the face of persistent non-adherence to treatment (or sanitation) by a small group of infected residents or migrants [44, 45, 62].
In summary, there are substantial complexities in the human and snail factors that can affect Schistosoma transmission dynamics and related predictions of MDA-based schistosomiasis control outcomes. This study finds that nonlinear human-snail coupling (FOI) can improve model prediction. Although other model structures could also provide broad agreement with the data, nonlinear snail FOI could provide a plausible explanation of strong MDA resilience (hotspots) observed in the SCORE studies and the observed heterogeneous community responses reported elsewhere [28, 29]. The present work will motivate future studies to apply these ideas to connected human-snail environments (see [14], [63]), and to the analysis of recent control datasets to develop tools to more accurately predict hotspots and explore strategies for their efficient control.
|
10.1371/journal.ppat.1002032 | Investigating the Host Binding Signature on the Plasmodium
falciparum PfEMP1 Protein Family | The Plasmodium falciparum erythrocyte membrane protein 1
(PfEMP1) family plays a central role in antigenic variation and cytoadhesion of
P. falciparum infected erythrocytes. PfEMP1
proteins/var genes are classified into three main
subfamilies (UpsA, UpsB, and UpsC) that are hypothesized to have different roles
in binding and disease. To investigate whether these subfamilies have diverged
in binding specificity and test if binding could be predicted by adhesion domain
classification, we generated a panel of 19 parasite lines that primarily
expressed a single dominant var transcript and assayed binding
against 12 known host receptors. By limited dilution cloning, only UpsB and UpsC
var genes were isolated, indicating that UpsA
var gene expression is rare under in vitro
culture conditions. Consequently, three UpsA variants were obtained by rosette
purification and selection with specific monoclonal antibodies to create a more
representative panel. Binding assays showed that CD36 was the most common
adhesion partner of the parasite panel, followed by ICAM-1 and TSP-1, and that
CD36 and ICAM-1 binding variants were highly predicted by adhesion domain
sequence classification. Binding to other host receptors, including CSA, VCAM-1,
HABP1, CD31/PECAM, E-selectin, Endoglin, CHO receptor “X”, and
Fractalkine, was rare or absent. Our findings identify a category of larger
PfEMP1 proteins that are under dual selection for ICAM-1 and CD36 binding. They
also support that the UpsA group, in contrast to UpsB and UpsC
var genes, has diverged from binding to the major
microvasculature receptor CD36 and likely uses other mechanisms to sequester in
the microvasculature. These results demonstrate that CD36 and ICAM-1 have left
strong signatures of selection on the PfEMP1 family that can be detected by
adhesion domain sequence classification and have implications for how this
family of proteins is specializing to exploit hosts with varying levels of
anti-malaria immunity.
| The malaria parasite Plasmodium falciparum persists in the human
host partly by avoiding elimination in the spleen during blood stage infection.
This strategy depends principally upon members of the large and diverse PfEMP1
family of proteins that are exported to the surface of infected erythrocytes.
PfEMP1 proteins are important targets for host protective antibody responses and
encode binding to several different host receptor proteins. Switches in PfEMP1
expression allow parasites to evade host antibodies and may precipitate severe
disease when infected erythrocytes accumulate in brain or placenta.
Consequently, the severity of malaria infection may depend on the type of PfEMP1
protein expressed. In this study, we employ a representative panel of distinct
PfEMP1 types and host receptor proteins to demonstrate that CD36 and ICAM-1
binding properties of full-length PfEMP1 are highly predicted by their domain
composition. We also find that CD36 binding is under strong selection in many
PfEMP1 proteins, but that a group of PfEMP1s associated with more severe
infections does not bind CD36 and may utilize alternative means to sequester
infected erythrocytes. These findings have implications for understanding the
molecular basis for severe malaria.
| Plasmodium falciparum erythrocyte membrane protein 1 (PfEMP1) is a
clonally variant adhesion protein that mediates binding of infected erythrocytes
(IE) to blood microvasculature and other host cells [1]. Adherence of IEs to
microvascular endothelium is a major virulence factor and, in conjunction with the
related phenomenon of rosetting with uninfected erythrocytes, prevents parasitized
erythrocyte circulation to the spleen where parasites may be destroyed [2]. Each parasite
strain encodes ∼60 PfEMP1 proteins, or var genes, which are
expressed in a mutually exclusive fashion [3], [4]. Switches in
var gene expression enable infected erythrocytes to evade host
immunity and may modify disease manifestations by changing parasite binding tropism
[5]–[7].
Efforts to unravel the role of PfEMP1 proteins in disease are complicated by the vast
diversity of var genes. Each parasite has a diverse repertoire of
genes, and there is limited overlap of repertoires between parasite genomes [8]–[10]. However, genes can be classified into three main
subfamilies denoted Groups A, B, and C [11], plus three unusual
strain-transcendent variants (var1csa, var2csa,
and type 3 var) [12]–[15]. The var gene subfamilies possess
distinctive upstream flanking regions termed UpsA, UpsB, and UpsC and are found in
characteristic locations in the subtelomeric or central regions of chromosomes [4], [9], [11], [12]. It has been
hypothesized that var gene organization may contribute to a gene
recombination hierarchy that influences gene function and evolution [1].
A number of studies have sought to correlate specific parasite adhesion traits with
disease outcome [16]–[19]. To date, at least 12 host receptors have been reported
to mediate P. falciparum IE binding [20]. CD36 binding is the most common
adhesion trait in the parasite population, followed by intercellular adhesion
molecule 1 (ICAM-1) [17], [19]. These two receptors can synergize under flow conditions
to mediate infected erythrocyte binding to microvasculature endothelium [21]–[23]. Most other binding
properties appear to be rarer or have not been studied in more than one or a few
parasite isolates. ICAM-1 binding has been associated with cerebral malaria in some
studies [17],
[24], but not
in others [19],
[25]. In
addition, infected erythrocyte rosetting, or binding of parasitized red blood cells
to uninfected red blood cells, has been associated with disease severity in African
children [26]–[28]. The clearest disease association is placental malaria,
in which parasites express the unusually strain-transcendent VAR2CSA PfEMP1 protein
and adhere to chondroitin sulfate A (CSA) in the placenta [14], [29]. VAR2CSA is a leading candidate
for a pregnancy malaria vaccine and a paradigm for syndrome-specific anti-disease
vaccine efforts.
Although the molecular basis for other adhesion-based complications of P.
falciparum is less established than for pregnancy malaria, several
observations suggest the antigenic diversity of severe malaria isolates may also be
limited. For instance, immunity to severe malaria appears to be acquired after
relatively few infections [30], [31]. In addition, isolates from severe malaria cases appear
to express a relatively restricted variant antigen surface repertoire [32]–[34]. Furthermore,
seroepidemiological and var transcriptional profiling studies
suggest that UpsA variants are more commonly expressed in young African children
with limited immunity and in severe malaria infections [35]–[39]. Therefore it is possible the
UpsA group has become specialized to exploit individuals with limited anti-malaria
immunity, and it is important to understand what may account for this expression
profile.
To gain insight into PfEMP1 binding properties, sequence classification has been
performed [40]. The
extracellular binding region of PfEMP1 proteins is comprised of 2–7
receptor-like domains called Duffy Binding-Like (DBL) and Cysteine Rich Interdomain
Region (CIDR) [41], [42]. DBL and CIDR domains are classified into different major
types (α to ε) and sub-types by sequence criteria [10], [40]. PfEMP1 proteins can be further
subdivided by protein architecture into small proteins with a four-domain
extracellular binding region (DBL-CIDR-DBL-CIDR) and large proteins with a more
complex domain composition [43]. By comparison to other groups, nearly all of the UpsA
proteins are in the large protein category [9], [10]. The best characterized binding
interactions are between CIDR::CD36 and DBLβ::ICAM-1 [44]–[46]. In a repertoire-wide binding
comparison with CIDR recombinant proteins, the majority of proteins encoded CD36
binding function, except for the UpsA group, which had different CIDR sequence types
than the UpsB and UpsC groups [11], [12], [46]. UpsA proteins may also be under less selection to bind
ICAM-1, as 7 of 23 DBLβ domains from the IT4 parasite strain bound ICAM-1, but
none of the 9 DBLβ domains tested from the UpsA group were ICAM-1 binders [44]. However, using
a different binding analysis in a BioPlex system, only a single DBLβ recombinant
protein from the 3D7 parasite strain bound ICAM-1, and it was from an UpsA protein
[45]. UpsA
proteins have also been reported to bind ICAM-1 (PFD1235w) and PECAM-1 (PF11_0008)
[47].
Taken together, sequence and binding analysis suggest the UpsA group forms a
preferential gene recombination group that is under less selection to bind the
primary microvasculature receptor CD36. Furthermore, it is possible UpsA genes may
have evolved specialized binding properties that contribute to their preferential
expression in the malaria non-immune.
While sequence and binding analysis of isolated domains have provided significant
insights into PfEMP1 function, few binding predictions have been confirmed for
native proteins at the IE surface, and it is not yet established whether binding
differences truly exist between var gene subfamilies. Furthermore,
it is possible that recombinant protein binding properties may be modified by
adjacent domains [48] or may not extrapolate to the native PfEMP1 molecule
[49]. Thus,
there remain significant uncertainties in our ability to predict IE binding, and
there is still limited understanding of how host selection is shaping the PfEMP1
variant antigen repertoire for parasite survival and transmission. For this study,
we generated a large panel of cloned parasite lines from the cytoadhesive IT4/FCR3
parasite strain and selected three highly enriched UpsA parasite lines with specific
monoclonal antibodies. This panel was employed to both investigate the major host
selection binding pressures operating on the protein family and to evaluate binding
predictions based on sequence information and isolated domain binding assays.
To create a panel of parasites for phenotypic analysis, parasites were cloned
from a long-term, continuous culture of the IT4/25/5 clone A4 (Figure 1) [6]. The
IT4/25/5 (IT4) parasite genotype is unusual because the parasite maintained its
cytoadhesion capabilities after in vitro adaptation [50], [51], making it
a primary model for this virulence determinant. A set of 54 var
genes has been reported from the IT4 parasite genotype [9], [10]. The A4 cloned parasite line
expresses a var gene (A4var/IT4var14) that has
an unusually high switch frequency (∼1–2% per generation) [6], [52],
resulting in PfEMP1 heterogeneity at the population level in the long term
culture. After 70 parasite divisions in continuous culture, the long-term A4
culture had completely switched away from the A4var gene
(IT4var14) and expressed a mixture of different
var genes at low levels with IT4var26,
IT4var31, and IT4var37 predominating
(Figure 2). Both
IT4var31 (previously referred to as
C18var) and IT4var37 (previously referred to
as AFBR6) were also found to be common switch events in two
previous studies of var gene switching within the A4 parasite
lineage [7],
[52],
suggesting that these particular genes may have high “on” rates in
unselected cultures.
Initially, 17 subclones were isolated from the long-term A4 parasite culture by
limited dilution cloning (Figure
1). From var transcription profiling, 6 of the
subclones transcribed IT4var31 as either the primary or
secondary var transcript, 8 transcribed dominant
var gene transcripts distinct from each other, and a
dominant var transcript (present at greater than 50% of
the total var transcripts) could not be identified in 3 of the
subclones by qRT-PCR analysis (Table 1, and data not shown). Ten subclones that primarily expressed
single dominant var transcripts, including two that expressed
IT4var31, were selected for phenotypic analysis (Figure 3).
Of interest, there was negligible UpsA transcription in the long-term A4 culture
(Figure 2), and none of
the isolated subclones expressed an UpsA var gene (Figure 3). To attempt to
enrich for UpsA variants, the long-term A4 culture was panned on CD36 receptor
protein and non-adherent parasites were selected. Although the
var transcriptional profile was modified after CD36
negative selection, this approach did not enrich for UpsA variants. Instead, the
frequent switch variant IT4var31 was the resulting major
transcript (data not shown). This again indicates that UpsA genes are rare
switch events in long-term A4 cultures.
To create a more representative panel for phenotypic analysis, six previously
isolated parasite lines from the IT4/FCR3 strain and three UpsA parasite lines
from different parasite strains (IT4/FCR3, Palo Alto 89F5, and 3D7) (Figure 1) were included in the
binding studies. The three UpsA parasite lines (R29, VarO, and Pf13) were
isolated by rosette enrichment and selected for high purity using specific
monoclonal antibodies to the respective PfEMP1 proteins [53]. Altogether, 19
parasite lines were examined representing all three major var
gene groups. Three of the parasites in the panel expressed an UpsA protein as
the dominant var transcript, ten expressed an UpsB
var gene, three expressed an UpsC var, one
expressed the unique UpsE linked transcript (IT4var4,
var2CSA), and for one parasite, the Ups category of the
dominant var transcript has yet to be determined (Figure 4). The remaining
parasite in the panel, 2G2, is knobless and was employed as a negative binding
control (Table 1) [54]. Most
parasites in the panel expressed distinct dominant var
transcripts, except two subclones (P6G2 and P5B6) expressed
IT4var31, and two others (P6A8 and 4E12) expressed
IT4var37/AFBR6 as either the dominant or secondary
var transcript (Figure 3 and Table 1).
To confirm the presence of knobs on the IE surface, which are known to be
important in PfEMP1 anchoring and infected erythrocyte binding [41], [54], [55], parasites
were tested for transcription of the knob associated histidine rich protein
(kahrp+) by RT-PCR and floated by gelatin
sedimentation (gelatin+). All parasites in the panel were
positive in both assays, except for the negative control 2G2 parasite line,
which lacks kahrp and therefore sedimented in gelatin. In
addition, the three rosette-forming UpsA parasites all transcribed
kahrp but sedimented in gelatin because they were
originally isolated on the basis of their property to sediment in Ficoll (Table 1). To confirm the
identity of var gene transcription at the time of binding
assays, RNA was harvested within the same growth cycle that binding assays were
performed. For these assays, thawed parasite stabilates were grown for 4 to 5
cycles to generate sufficient parasite material, and parasites were generally
analyzed a total of 18–20 cycles from initial parasite cloning. In
general, the dominant var transcript did not change between the
initial qRT-PCR characterization performed after limited dilution cloning and
the second round of -typing done at the binding assay (Table 1). In only one parasite line, P6G2,
the previous dominant transcript was replaced by the secondary
var transcript that was present before freezing (Table 1). At the time of the
binding analysis, the average fold transcription of dominant
var transcripts relative to the asl
housekeeping gene was 14.2 (range 2.8–28.1). Furthermore, most parasite
lines were significantly enriched for a single predominant var
transcript (Figure 3), and
only 8 parasite lines contained a secondary var transcript at
greater than 5% of the total var transcripts (Table 1). In most cases, the
secondary transcript was present at much lower levels than the dominant
var transcript. Thus, var gene
transcription was stable over the short-term culture period used to perform
these assays. For the three UpsA variants, PfEMP1 expression was established by
flow cytometry with specific monoclonal antibodies to be 79% or higher
using conservative gating criteria (Figure S1). Furthermore, all three lines
formed rosettes in O-type RBCs: R29 (rosetting
rate = 37%, 89% mAb reactivity R29), VarO
(rosetting rate = 73%, 79% mAb reactivity
VarO), Pf13 (rosetting rate = 52%, 94% mAb
reactivity Pf13_0003). Therefore, all of the parasites in the panel were highly
homogenous for one or two var transcripts, and UpsA parasite
lines were highly pure for a single expressed PfEMP1 variant.
To investigate whether infected erythrocyte binding to CD36 could be predicted
from sequence information and binding studies of isolated CIDR domains [46], the
complete panel of parasite lines was analyzed for binding to both CHO745-CD36
and immobilized CD36 recombinant protein. Because rosettes of uninfected red
blood cells can interfere with binding or make bound IEs more susceptible to
disruption during washing stages, the rosettes of the three UpsA variants were
first disrupted using heparin sulfate prior to binding analysis. Previous work
has shown that sulfated glycoconjugates can enhance binding to CD36 on cell
surfaces [56]. Therefore, as a control for the three rosetting
parasite lines, all of the parasites in the panel were treated with heparin
sulfate and tested for binding to immobilized CD36 recombinant protein. Heparin
sulfate treatment greatly diminished rosette formation in the three UpsA
parasite lines (<10%) (Figure S2), but had minimal effect on
infected erythrocyte binding to immobilized CD36 recombinant protein. Overall,
parasites had comparable binding levels in the presence or absence of heparin
sulfate (Figure
S2). In addition, two non-rosetting, CD36 binding parasite lines
(A4ultra and ItG-ICAM-1) were tested for binding to CHO745-CD36 cells in the
presence or absence of heparin sulfate. Similar to what has been reported
previously [56], sulfated glycoconjugates increased IE binding to
CHO745-CD36 (Figure S3). Because heparin sulfate may slightly enhance IE adhesion
to CHO-CD36 and did not modify IE adhesion to immobilized CD36, the binding
assay was then repeated for all of the non-UpsA parasites in the absence of
sulfated glycoconjugates. In contrast, binding of the three UpsA lines to
CHO-CD36 and immobilized CD36 was repeated in the presence of sulfated
glyconjugates to prevent infected erythrocyte rosetting interfering with the
binding results.
Overall, there was a significant correlation between CHO745-CD36 and spotted CD36
protein formats (Figure 5,
Spearman's Rho = 0.75, p<0.001). Although the level
of CD36 binding varied between parasite lines, most of the parasites bound CD36,
with the exception of UpsA/E groups (Figure 6). The three UpsA parasites were at
the lower spectrum of CD36 binding in both cell and recombinant protein binding
assays, and were basically indistinguishable from the negative control, knobless
parasite line, and the UpsE parasite line that does not bind CD36 (Figure 6). Furthermore, CD36
binding was highly predicted by the type of CIDR1 domain in the PfEMP1 head
structure (Figure 4).
Indeed, only two parasites in the panel that were predicted to bind CD36 did not
bind to CHO-CD36 cells. However, both exceptions (P4H12 and P3G5) bound at a low
level to 50 µg/mL rCD36, but not to 5 µg/mL rCD36 (Figure 5), and therefore may
be lower affinity CD36 binders. In group-wide comparisons, UpsB and UpsC had a
higher mean CD36 binding level than UpsA. This difference was significantly
different in the immobilized CD36 binding assay and between the UpsC and UpsA
groups in the CHO-CD36 assay, and just missed significance between the UpsB and
UpsA groups in the CHO-CD36 assay (Figure 6). Taken together, infected erythrocyte binding was highly
predictable based on the type of CIDR domain (Figure 4), and the UpsA group appears to be
under less selection to bind CD36.
To test whether ICAM-1 binding was associated with larger PfEMP1 proteins
containing DBLβ domains [44], the parasite panel was analyzed for binding to
CHO745-ICAM-1 and recombinant ICAM-1 protein. Again, to prevent rosettes from
interfering with the binding analysis, the three UpsA parasite lines were
treated with sulfated glycoconjugates prior to binding analysis, and as a
control, two non-rosetting, ICAM-1 binding parasite lines (A4ultra and
ItG-ICAM-1) were tested for ICAM-1 binding in the presence or absence of
sulfated glycoconjugates. Sulfated glyconjugates reduced binding of A4ultra in
the CHO745-ICAM-1 assay and binding of both parasite lines to spotted ICAM-1
recombinant protein (Figure S3), similar to what has been reported
before [56]. Because of the potential for sulfated glyconjugates to
interfere with ICAM-1 binding in the cell and recombinant protein assays, the
three UpsA parasite lines were not considered in the ICAM-1 binding
analysis.
In the cell binding assay, two parasite lines bound at a high level (>2
IEs/CHO745-ICAM-1), three bound at moderate level (0.5–2
IEs/CHO745-ICAM-1), and the remaining parasite lines bound at a low level or did
not bind ICAM-1 (Figure 5).
While there was good consistency between the cell and recombinant protein assays
for the two high level ICAM-1 binders, there was more discordance for weaker
ICAM-1 binders (Figure 5).
Only three parasite lines bound ICAM-1 in both platforms (3G8, ItG-ICAM-1, and
A4ultra), and two parasite lines that bound at a moderate level to CHO745-ICAM-1
did not bind to immobilized ICAM-1 protein (Figure 5). Notably, both parasite lines
express the IT4var31 transcript, which has been suggested to be
a weaker ICAM-1 binding variant that is trypsin-resistant [57], [58]. To confirm whether binding
was trypsin-resistant, P5B6-infected erythrocytes expressing
IT4var31 were treated with 1 mg/mL trypsin prior to ICAM-1
binding analysis. Trypsin treatment reduced CD36 binding and increased binding
to recombinant ICAM-1 (Figure S4), and therefore may have cleaved or
truncated the PfEMP1 head structure. The increase in ICAM-1 binding could be
blocked by anti-ICAM-1 antibody (mAb 15.2) and not by anti-CD36 isotype control
antibody (FA6-152) (Figure S4). In contrast, identical trypsin
treatment of 3G8 (IT4var1) and ItG-ICAM-1 parasite lines
(IT4var16) abolished binding to both CD36 and ICAM-1 (data
not shown). Thus, as predicted from binding of the isolated DBLβ domain
[58],
IT4var31 was associated with ICAM-1 binding, but the cell binding assay was more
sensitive than immobilized protein in detecting this interaction. Two of the
parasite lines also bound at a low level to immobilized ICAM-1 recombinant
protein but did not bind CHO745-ICAM-1. Thus, there may be differences in the
sensitivity of the two platforms to detect lower affinity ICAM-1 interactions,
or some of the low level binding interactions may not have been specific.
Overall, ICAM-1 binding was strongly associated with larger PfEMP1 proteins that
contained a DBLβ domain. Seven of the ten parasites lines that expressed a
dominant var transcript containing a DBLβ domain bound to
ICAM-1 in either the cell or recombinant protein platform (Figure 4), and parasite lines without a
DBLβ either bound extremely weakly or did not bind ICAM-1 (Figure 6). This difference was
significant in the immobilized ICAM-1 assays (1-tailed t-test,
p = 0.020) and just missed significance in the
CHO745-ICAM-1 assay (1-tailed t-test, p = 0.103). Recently,
there has been a reclassification of DBL and CIDR domains into additional
subtypes based on a comparison of 7 parasite genomes in which DBLβ domains
were subclassified into 13 sub-types [10]. Of interest, all three
parasites that bound in both the CHO-ICAM-1 and immobilized ICAM-1 assays
expressed a DBLβ5 domain (Figure 4). To investigate if DBLβ5 could be a marker for ICAM-1
binding, we reanalyzed the recombinant DBLβ-ICAM-1 binding data [44]. In the IT4
parasite genotype, 7 of 23 DBLβ domains bound ICAM-1. Of the 7 ICAM-1
binders, 6 were DBLβ5 sequences, and there were no DBLβ5 domains that
did not bind ICAM-1 (Figure
7). Significantly, an ICAM-1 binding parasite from India
(JDP8-ICAM-1, AY028643) [59] also uses a DBLβ5 domain to bind ICAM-1 (Figure 7). The fact that
ICAM-1 binding was 100% predictable in the IT4 parasite genotype, and
that a different parasite isolate from India also uses DBLβ5 for binding,
strongly supports this domain as a marker for ICAM-1 binding. There are also two
DBLβ3 sequences that bound ICAM-1, one from the IT4 parasite genotype and
one from the 3D7 parasite genotype [45], but several other
DBLβ3 sequences did not bind ICAM-1 as recombinant proteins (Figure 7). Taken together,
ICAM-1 binding was strongly associated with the DBLβ domain, and the
DBLβ5 marks a category of larger PfEMP1 variants that encode this adhesion
property.
Infected erythrocytes have been reported to bind a number of host receptors [20], but for the
most part binding has only been tested on one or a few parasite lines. Using
transfected cells or recombinant proteins, the 19 parasite lines were assayed
against 8 additional receptors: Endothelial Leukocyte Adhesion Molecule 1
(E-selectin), Vascular Cell Adhesion Molecule 1 (VCAM-1), CHO receptor
“X”, Hyaluronan Binding Protein 1 (HABP1), Platelet Endothelial Cell
Adhesion Molecule-1 (CD31/PECAM-1), Thrombospondin-1 (TSP-1), CSA, and
Fractalkine. Whereas a few parasite lines bound at a low level to TSP-1 and
CHO-ELAM-1, there was negligible binding to most receptors tested (Figure 8). Two of the UpsA
parasites (Pf13 and VarO) bound at a low level to HABP1, CD31, and CSA. However,
binding of UpsA parasites was performed in the presence of sulfated
glycoconjugates to disrupt rosettes, and they also had higher background binding
to bovine serum albumin (BSA) employed as a blocking agent for binding assays
(Figure 8, and data not
shown). As expected, the strongest CSA-binder in the panel was the CS2 parasite
line in both the CHO-K1 cell and CSA spot formats (Figure 8). CS2 expresses the VAR2CSA PfEMP1
protein that has been shown to be the primary PfEMP1 variant associated with CSA
binding [60],
[61]. Most
of the other receptors tested did not support strong adhesion of infected
erythrocytes in these binding assays and it is questionable whether all of the
observed weak interactions are physiologically relevant.
PfEMP1 proteins/var genes are classified into three main subfamilies
(UpsA, UpsB, and UpsC) that have different host expression profiles [35]–[37], [39]. Both binding
strength and specificity of IEs are likely to influence disease severity during an
infection; therefore, it is important to understand whether PfEMP1 subfamilies have
evolved specialized properties for distinct host/biological niches. Studies of
malaria during pregnancy have demonstrated how a specific PfEMP1 variant can
precipitate severe disease in otherwise immune women by altering IE tropism for the
placenta [14],
[29], [62]. Although
VAR2CSA appears to be unique in its ability to confer high-affinity binding to CSA
in the placenta [60], [61], [63], it offers a paradigm for the role of specific PfEMP1s in
disease. UpsA classified PfEMP1 proteins are frequently observed in young children
with limited anti-malaria immunity or experiencing severe malaria [35]–[39]. Unlike
VAR2CSA, the adherence characteristics of UpsA proteins are poorly understood and
limited largely to predictions of binding based on studies of isolated adhesion
domains [44]–[46]. To investigate a correlation of PfEMP1 binding
specificities with disease outcome, the binding characteristics of at least a
representative sample of the three main subgroups (UpsA, UpsB and UpsC) have to be
known. In this study, we employed a panel of different PfEMP1 types to test binding
predictions based upon studies of single PfEMP1 domains.
While UpsA variants appear to be commonly expressed in early childhood infections and
non-immune individuals [35]–[39], very little is known about what may account for this
preferential expression in the malaria naïve. Investigation is hampered because
most P. falciparum infections contain a mixture of PfEMP1 variants
and even minor parasite subsets may obscure binding analysis. In addition, gene
silencing of UpsA variants has been observed upon in vitro
adaptation [64].
In long term in vitro adapted parasite cultures grown without
selection for specific var gene expression, UpsA variants were
expressed at a low level, and an UpsB (IT4var31/C18var) and an UpsC
(IT4var37/AFBR6) var gene
appeared to be the most common switch events. Both were also found to be frequently
activated in previous clonal analyses in this strain background [7], [52] and thus may
have a higher “on” rate under in vitro culture
conditions. One study found that var genes in central chromosome
regions had lower switch rates than those in telomeric regions [65], but inherent differences were
not consistently observed in a different parasite line [52]. The chromosome positions of
IT4var31 (UpsB) and IT4var37 (UpsC) have not
been mapped and therefore we cannot comment on whether this observation held true in
our study or not. However, our findings indicate that promoter type is not the main
determinant of var gene “on” rate as far as UpsB and
UpsC type var genes are concerned. In the case of UpsA variants,
the promoter type did seem to determine var gene expression rate by
significantly reducing it. To overcome these problems, we used specific monoclonal
antibodies to generate three distinct UpsA parasite lines of high purity for the
parasite panel.
In epidemiological studies, CD36 and ICAM-1 binding are the most common adhesion
traits in the parasite population [17], [19], but their distribution among different members of the
PfEMP1 family is only partially understood [44]–[46], [58], [66]. In the parasite panel, CD36
was by far the most common binding partner, followed by ICAM-1 and TSP-1.
CD36-binding was nearly 100% predictable and was always associated with a
CIDRα type domain in the protein head structure, while the three UpsA variants
had different sequence types (CIDRγ and CIDRδ) and did not bind CD36 or only
bound at a low level. Thus, in the absence of a CIDRα domain, other potential
CD36 ligands [67], [68] were unable to compensate for infected erythrocyte
binding. Moreover, the level of CD36 binding differed between isogenic parasites
expressing different PfEMP1 variants, suggesting that PfEMP1 sequence variability or
surface expression levels have an important role in influencing the overall binding
affinity of infected erythrocytes.
The UpsA group contains three different types of CIDR1 sequences (α1, γ, or
δ) [10], [12], [40], [46]. Although the
three UpsA parasites in the panel were all selected for rosetting,
“rosetting” and “non-CD36 binding” can exist as independent
phenotypes. For instance, the non-CD36 binding CIDR domains identified in this study
may potentially be found in non-rosetting group A genes, and there is evidence that
CD36 is able to act as a host receptor for rosetting in the Malayan Camp parasite
strain and some field isolates [69]. This parasite panel did not contain any representation
of the CIDRα1 subtype, which is found in approximately half of UpsA proteins
[10]. However,
it has previously been shown that recombinant CIDRα1 subtype domains do not bind
CD36 [46], and
CD36 selection led to loss of expression of an UpsA gene in a mixed parasite culture
that expressed a CIDRα1 subtype [70]. Taken together, the
results suggest the UpsA group is not under strong selection for CD36 binding, and
it will be interesting to determine if the UpsA protein head structure is selected
for specific binding properties that support microvasculature sequestration by a
mechanism different from CD36 binding. Part of this selection may be for infected
erythrocyte rosetting [71], [72], but the UpsA group may encode other adhesion properties
[47].
After CD36, ICAM-1 is one of the most common adhesion properties, and the two
receptors synergize to mediate infected erythrocyte binding under flow [22], [23]. ICAM-1 is
upregulated on brain endothelium during malaria infections and has been proposed to
be a potential cerebral sequestration receptor [24]. ICAM-1 binding has previously
been mapped to the DBLβ domain [44], [45], [58], [59], [73]. Our study confirms this association as the DBLβ5
domain was 100% associated with ICAM-1 binding in both parasite lines and
recombinant proteins. It also shows that not all DBLβ domains bind to ICAM-1. In
future work using patient samples it may be interesting to investigate how well
transcription of var genes containing a DBLβ5 domain can
predict ICAM-1 binding. Overall, this study identifies a category of large UpsB and
UpsC PfEMP1 containing CIDRα and DBLβ5 subtype domains that were 100%
associated with CD36 and ICAM-1 binding. In a comparison of var
gene repertoires from 7 parasite strains, the CIDRα and DBLβ5 domains were
always found together in tandem arrangement (27 of 399 full or partial length
var genes), and the DBLβ5 domain was never associated with
a predicted “non-CD36 binding” CIDR domain. This suggests the
association has not evolved by chance and that the CIDRα-DBLβ5 domain
combination may be under dual selection for binding to CD36 and ICAM-1. Both
receptors are co-displayed on many of the same cell types (endothelial, monocyte,
and dendritic cells) and may provide the parasite opportunities to manipulate host
cells [74], [75], thus
contributing to their strong selection in the PfEMP1 repertoire. There were also a
few DBLβ3 domains that bound to ICAM-1, but these were found in association with
both CD36 binding and non-CD36 binding CIDR domains. Thus, CD36 and ICAM-1 have left
strong signatures of selection detectable by PfEMP1 adhesion domain sequence
classification, despite the extensive sequence diversity in the family.
Other PfEMP1 adhesion properties examined appear to be much rarer or may only play an
additive role in overall binding affinity. Nearly all PfEMP1 proteins have four or
more extracellular domains. In addition to undefined binding properties, other
PfEMP1 domains may also function as “spacers” to position the PfEMP1
head structure and adjacent DBLβ away from the IE surface in order to engage
CD36 and ICAM-1 [76]. A potential caveat is that binding was performed under
static adhesion conditions, and individual host recombinant proteins were employed
in the protein binding assays. However, all host receptors examined were originally
defined under similar static adhesion conditions. Furthermore, static adhesion
assays are capable of detecting host receptor interactions that support both rolling
(ICAM-1, TSP-1) and stationary (CD36) cytoadhesion of infected erythrocytes under
flow conditions [21]. Cooperative binding is likely necessary to mediate firm
adhesion under flow [21]–[23], but from this analysis CD36 binding is under greatest
selection and contributes the greatest binding avidity in different PfEMP1
proteins.
These results reveal a fundamental difference in CD36 binding between Ups groups that
has important implications for how parasites establish infections in individuals of
varying levels of immunity. UpsA proteins are more commonly expressed in children
with low immunity [35], [36], [39]. Later, as malaria immunity develops, it may be
significant that the proportion of non-UpsA types and CD36 binding variants
increases. It is interesting to speculate that non-CD36-binding parasites may
experience a selective advantage over their CD36-binding counterparts in patients
with limited exposure to malaria. CD36-binding parasites are thought to manipulate
both host innate and adaptive immune responses by interacting with monocytes and
dendritic cells [74], [75], [77], [78]. In the malaria naïve, these interactions may be
less important, or UpsA variants may possess other advantages or means of host
manipulation. While UpsA variants have not been clearly associated with disease in
all studies [79], they are more abundant in patients with severe malaria [80], [81] and have been
associated with cerebral malaria infections in children in Mali [38]. A greater
proportion of UpsA variants in early infections could potentially contribute to why
CD36 binding levels are very low in children with severe malaria anemia [17], [19], or these
variants could alter the pattern of sequestration to microvascular beds, such as
brain endothelium, where CD36 binding levels are extremely low [24]. Therefore, it will be
important to learn more about this group of proteins.
In conclusion, the PfEMP1 protein family has diversified under dual selection to
evade host immunity and mediate infected erythrocyte binding. The development of a
parasite panel enriched for distinct PfEMP1 expression from the major Ups groups has
facilitated the testing of binding predictions, and may have potential applications
for investigating immune acquisition to the family of proteins. This comparative
analysis demonstrates the predictability of P. falciparum-IE
binding to the two major cytoadhesion receptors CD36 and ICAM-1 and provides new
insight into how natural selection may be shaping the PfEMP1 binding repertoire to
exploit distinct host niches of varying anti-malaria immunity.
Human blood was used for P. falciparum culture in this study.
Donor blood was obtained from healthy volunteers under a minimal risk,
standardized, Institute protocol (protocol number HS013) that was approved by
the Western Institutional Review Board. Written informed consent was obtained
from all blood donor study participants.
The three UpsA variants were isolated by gelatin sedimentation followed by
positive selection with specific monoclonal antibodies against the respective
NTS-DBLα domain. The VarO parasite clone was generated from the Palo Alto
strain as described by rosette enrichment and selection with monoclonal antibody
D15–50 [82]. The R29 parasite (IT4 parasite strain) has been
described previously [6], [7], [83]. Highly enriched parasite cultures expressing the R29
PfEMP1 protein and Pf13 (3D7 strain) were isolated by similar methodologies to
the VarO parasite line using rosette enrichment and specific monoclonal
antibodies against the R29-DBLα domain (3B13C5) or the Pf13_0003-DBLα
domain (J3.21) [53]. The ItG-ICAM-1 parasite line was derived by ICAM-1
selection [18], CS2 by CSA selection [84], and the 3G8, 4E12, and 2G2
parasite lines by limited dilution cloning [52]. The remaining parasite
lines were derived from IT4/25/5 clone A4 [6] by limited dilution
cloning. Infected erythrocytes were cultured under standard conditions using
human O red blood cells (RBCs) in RPMI-1640 medium (Invitrogen) supplemented
with 10% pooled human A+ serum and an atmosphere of
5% CO2, 5% O2, and 95% N2
at 37°C. Synchronization of parasite growth was achieved by treatment with
5% sorbitol in PBS. Gelatin sedimentation assays were performed in
RPMI-1640 medium containing 0.7% porcine gelatin (Sigma) for 45 minutes
at 37°C. Enrichment of infected erythrocytes (IE) in the gelatin supernatant
was determined by counting >300 methanol-fixed, Giemsa-stained RBCs under
1000X magnification. Rosette formation was visualized after infected red blood
cell nuclei were stained by ethidium bromide. The rosetting rate was calculated
by determining the percentage of rosette-forming infected cells in the mature
parasite population.
CHO-K1, CHO745, and CHO745 transfectants expressing CD36, ICAM-1, E-selectin, or
VCAM-1 were cultured in F-12 Kaighn's medium supplemented with 10%
fetal calf serum and 0.5 mg/mL geneticin (Gibco). The CHO745 transfectants were
described in Buffet et al. [85]. Recombinant protein surface expression was monitored
by flow cytometry on a monthly basis using receptor-specific monoclonal
antibodies (R&D Systems), and cells were replaced if the percentage of
transfected cells or mean fluorescence intensity diminished by greater than
20%.
An A4 parasite clonal line [6] was grown continuously under standard conditions for
more than 70 growth cycles in the absence of overt selection. IEs were
periodically enriched for knob expression by floatation in 0.7% porcine
gelatin (Sigma) dissolved in RPMI-1640 (Invitrogen) at 37°C. Prior to
limited dilution cloning, RNA was collected and a profile of
var transcription was determined by quantitative real-time
polymerase chain reaction (qRT-PCR) using a primer set designed to amplify
unique sequence tags within the repertoire of IT4 var genes
[86].
Individual infected erythrocytes were obtained on two separate occasions by
limited dilution cloning after more than 78 and 84 cycles of continuous parasite
growth, respectively, at a seeding rate of 0.5 infected erythrocytes per well.
Initial frozen stabilates were collected after approximately 14–15 cycles
of growth and parasite lines were typed for var gene expression
by qRT-PCR.
The determination of var gene transcription profiles was
performed using primers and PCR conditions as previously described [86]. In brief,
RNA was extracted in Trizol LS (Invitrogen) from ring stage parasites at
∼6–12 hours post-invasion and purified on RNeasy Micro columns with
on-column DNaseI treatment (QIAGEN) according to manufacturer's protocols.
cDNA was synthesized from 4 µg total RNA using Multi-Scribe reverse
transcriptase (Applied Biosystems) and one half of this material was used for
each real-time reaction against the complete set of primers. Real-time reactions
were performed on an ABI Prism 7500 thermocycler at optimized final primer
concentrations of 0.05 µM-0.5 µM using Power-SYBR Green Master Mix
in 20 µL reaction volumes under the following PCR conditions: 50°C for
1 min, 95°C for 10 min, then 40 cycles of dissociation, annealing, and
extension at 95°C for 15 sec, 52°C for 15 sec, and 60°C for 45 sec,
respectively. Relative transcription was determined by normalization to the
adenylosuccinate lyase (ASL, PFB0295w) control housekeeping gene. After
optimizing primer efficiencies, residual primer bias was corrected by
calculating the average difference in CT values between each
optimized IT4 var primer pair and ASL using genomic DNA as
template to provide a final normalized correction.
Parasite RNA was collected and binding assays performed within the same growth
cycle to accurately assess var transcription at the time of the
binding assay. For binding assays, individual CHO cell lines were grown to
subconfluent levels on 60-mm tissue culture-treated dishes (BD Falcon) and
recombinant proteins were immobilized by overnight incubation onto 60-mm
polystyrene dishes (Corning). The following proteins were analyzed: CD36-Fc
(R&D Systems), ICAM-1-Fc (R&D Systems), HABP1/gC1qR-6x HIS (R&D
Systems), Fractalkine-6x-HIS (R&D Systems), CD31/PECAM-1 (R&D Systems),
TSP-1-10x HIS (R&D Systems), and CSA (Sigma). All proteins and CSA were
applied at 50 µg/mL except for CD36 and ICAM-1, which were additionally
applied at 5 µg/mL and 100 µg/mL. On the day of the assay, dishes
containing CHO cells were washed twice with pre-warmed cell binding medium
(BMcell: RPMI-1640 medium containing 0.1% bovine serum
albumin, pH 7.2) and protein spots were blocked with 2% bovine serum
albumin for 45 min at 37°C, then washed twice with pre-warmed protein
binding medium (BMprotein: RPMI-1640 medium containing 0.1%
bovine serum albumin, pH 6.8). Infected erythrocytes (3-8% parasitemia)
were washed and resuspended to 1% hematocrit in either BMcell
or BMprotein then overlayed onto CHO cells or spotted onto
immobilized proteins, respectively, and incubated for 1 hr at 37°C. Prior to
binding assays, rosettes in the three UpsA parasite lines were disrupted in
binding medium containing 100 Units/mL heparin sulfate (Sigma). The same
conditions were used when testing the effect of heparin sulfate on all of the
parasites in the panel. In additional assays to test the effect of sulfated
glycoconjugates on IE binding, either 10 µg/mL dextran sulfate (MW
>500,000; Sigma) or 100 Units/mL heparin sulfate were included during binding
assays. Non-binding erythrocytes were removed by gently flooding each dish with
warm binding medium, rocking the dish back and forth several times to resuspend
non-binding erythrocytes, then pouring off and replacing the medium. The initial
washing procedure was performed on CHO745 cells and 2% BSA spots and was
repeated until non-binding erythrocytes were sufficiently removed by observation
under 400X magnification. The remaining cells and spots then received the same
number of washes. For quantification, dishes were fixed in 1%
glutaraldehyde for 20 min at room temperature, then stained with 1X Giemsa for
15 minutes. Binding was quantified by determining the number of IE adhering to
at least 300 cells under 1000X magnification or the number of IE per
mm2 in 4 random fields under 400X magnification. All binding
assays were repeated in duplicate.
Trophozoite stage infected RBCs were incubated for one hour at room temperature
with specific monoclonal mouse antibodies against R29var NTS-DBLα (mAb
3B13C5, 1∶500) Pf13_0003 NTS-DBLα (mAb J3.21, 1∶20), or VarO
NTS-DBLα (mAb D15-50, 1∶20). Antibody labeling was detected with goat
anti-mouse IgG-R-Phycoerythrin (Sigma) (1∶20) for 30 minutes. Infected
erythrocyte nuclei were detected with SYTO 61 DNA dye (Invitrogen)
(1∶1000) added with the secondary antibody. Stained cells were washed in
PBS and analyzed on an LSRII FACS machine (BD Biosciences). Analysis was
performed using FlowJo 8 (Tree Star, Inc).
|
10.1371/journal.pbio.2000106 | Decoding Spontaneous Emotional States in the Human Brain | Pattern classification of human brain activity provides unique insight into the neural underpinnings of diverse mental states. These multivariate tools have recently been used within the field of affective neuroscience to classify distributed patterns of brain activation evoked during emotion induction procedures. Here we assess whether neural models developed to discriminate among distinct emotion categories exhibit predictive validity in the absence of exteroceptive emotional stimulation. In two experiments, we show that spontaneous fluctuations in human resting-state brain activity can be decoded into categories of experience delineating unique emotional states that exhibit spatiotemporal coherence, covary with individual differences in mood and personality traits, and predict on-line, self-reported feelings. These findings validate objective, brain-based models of emotion and show how emotional states dynamically emerge from the activity of separable neural systems.
| Functional brain imaging techniques provide a window into neural activity underpinning diverse cognitive processes, including visual perception, decision-making, and memory, among many others. By treating functional imaging data as a pattern-recognition problem, similar to face- or character-recognition, researchers have successfully identified patterns of brain activity that predict specific mental states; for example, the kind of an object being viewed. Moreover, these methods are capable of predicting mental states in the absence of external stimulation. For example, pattern-classifiers trained on brain responses to visual stimuli can successfully predict the contents of imagery during sleep. This research shows that internally mediated brain activity can be used to infer subjective mental states; however, it is not known whether more complex emotional mental states can be decoded from neuroimaging data in the absence of experimental manipulations. Here we show that brain-based models of specific emotions can detect individual differences in mood and emotional traits and are consistent with self-reports of emotional experience during intermittent periods of wakeful rest. These findings show that the brain dynamically fluctuates among multiple distinct emotional states at rest. More practically, the results suggest that brain-based models of emotion may help assess emotional status in clinical settings, particularly in individuals incapable of providing self-report of their own emotional experience.
| Functional neuroimaging offers unique insight into how mental representations are encoded in brain activity [1,2]. Seminal cognitive neuroscience studies demonstrated that distributed patterns of cortical activity measured with functional magnetic resonance imaging (fMRI) contain information capable of differentiating among visual percepts, including object categories [3] and basic visual features [4]. Extending findings from these studies, subsequent work demonstrated that machine learning models trained on stimulus-evoked brain activity, termed “decoding” or “mind-reading” [5], can be used to predict the contents of working memory [6–8] and mental imagery [9,10], even during sleep [11]. Thus, pattern recognition approaches can identify defining features of mental processes, even when driven solely on the basis of endogenous brain activity. The approach was further shown to accurately discriminate among multiple cognitive processes (e.g., decision-making, working memory, response inhibition, among others) in independent subjects [12], establishing the efficacy of assessing diverse mental states with fMRI across individuals.
Paralleling cognitive studies decoding task-evoked brain activity, multivariate decoding approaches have recently been used to map patterns of neural activity evoked by emotion elicitors onto discrete feeling states [13,14]. However, a key piece of missing evidence is whether categorically distinct emotional brain states occur intrinsically [15,16] in the absence of external eliciting stimuli. If so, then it should be possible to classify the emotional status of a human being based on analysis of spontaneous fluctuations of brain activity during rest. Successful classification would validate multivariate decoding of unconstrained brain activity and provides insight into the nature of emotional brain activity during the resting state.
Adapting the logic of other cognitive imaging studies [16,17], we postulate that the presence of spontaneous emotional brain states should be detectable using multivariate models derived from prior investigations of emotion elicitation. We previously developed decoding algorithms to classify stimulus-evoked responses to emotionally evocative cinematic films and instrumental music [13]. These neural models (Fig 1) accurately classify patterns of neural activation associated with six different emotions (contentment, amusement, surprise, fear, anger, and sadness) and a neutral control state in independent subjects, generalizing across induction modality. Importantly, these neural biomarkers track the subjective experience of discrete emotions independent of differences in the more general dimensions of valence and arousal [18]. By indexing the extent to which a pattern of neural activation to extrinsic stimuli reflects a specific emotion, these models can be used to test whether intrinsic spatiotemporal patterns of brain activity correspond to stimulus-evoked emotional states.
Here, we evaluate whether these neural models of discrete emotions generalize to spontaneous brain activation measured via fMRI in two experiments. The first experiment assesses if model predictions are convergent with individual differences in self-reported mood and emotional traits. Because individual differences are linked to mental health and subjective well-being [19–21], this evaluation provides insight into the potential clinical utility of quantifying spontaneous emotional states, as they may be associated with risk factors for mental illness. The second experiment employs an experience sampling procedure to evaluate whether model predictions based on brain activity during periods of rest are congruent with on-line measures of emotional experience. Together, these studies probe how brain-based models of specific emotion categories quantify changes in extemporaneous affect both between and within individuals.
We applied the multivariate models of emotional experience to brain activation acquired from young adults during resting-state fMRI (n = 499; Fig 2A). Two consecutive runs of resting-state scans were acquired, spanning a total duration of 8.53 min. Following preprocessing of data, we computed the scalar product of the resting-state signal and emotion category-specific model weights at every time point of data acquisition. This procedure yielded scores that reflect the relative evidence for each of seven emotional states across the full scanning period. A confirmatory analysis revealed that voxels distributed across the whole brain informed this prediction, as opposed to activity in a small number of brain regions (S1 Fig).
If emotional brain states occur spontaneously, the frequency of classifications from our decoding models should be more varied than the uniform distribution that would be expected by chance. To test this hypothesis, we sought to identify whether the total time (or absolute frequency) in each state differed across emotion categories. Such an analysis informs the degree to which discrete emotional brain states may spontaneously occur and, by extension, could contribute to the identification of individual differences that map onto the likelihood of experiencing specific spontaneous states. To perform this comparison, we identified the single model with the maximum score at each time point (one-versus-all classification) and summed the number of time points assigned to each category. The frequency of emotional states clearly differed across categories (Fig 2B, χ2 = 1491.52, P < .0001, Friedman test), in contrast to the uniform distribution that would be expected if emotional brain-states did not occur in spontaneous activity (see S2 Fig).
Follow-up comparisons revealed that neutral states occurred more frequently than chance rates (20.1 ± 3.59% [s.d.], z = 20.50, Punc = 2.03E-93), followed by states of surprise (18.37 ± 3.87% [s.d.], z = 16.38, Punc = 2.47E-60) and amusement (14.71 ± 3.78% [s.d.], z = 1.25, Punc = 0.21). States of sadness (13.49 ± 3.76% [s.d.], z = -3.31, Punc = 9.24E-4), fear (13.26 ± 3.42% [s.d.], z = -5.28, Punc = 1.28E-7), and anger (11.31 ± 3.62% [s.d.], z = -13.07, Punc = 4.78E-39) occurred with lower frequency, while states of contentment occurred the least often (8.74% ± 3.42% [s.d.], z = -19.61, Punc = 1.33E-85; see Table 1).
Although patterns of neural activation were most often classified as neutral as a whole, it is possible that consistent fluctuations in the time course of emotional states occur against this background. Research on MRI scanner-related anxiety has shown that self-report [22,23] and peripheral physiological [24] measures of anxiety peak at the beginning of scanning, when subjects first enter the scanner bore. This literature predicts that brain states indicative of fear should be most prevalent at the beginning of resting-state runs, and that neutral states should emerge over time, given their overall high prevalence (Fig 2B).
To assess gradual changes in the emotional states over time, we performed Friedman tests separately for each emotion category, all of which revealed significant effects of time (see S1 Table). Next, we quantified the direction of these effects using general linear models to predict classifier scores using scan time as an input. We found the scores for fear decreased over time (β^=−0.001, t498 = -4.92, Punc = 1.20E-006, Fig 3 gray lines), whereas neutral states exhibited an increasing trend throughout the scanning period (β^=0.0017, t498 = 7.36, Punc = 7.66E-013), consistent with predictions (additional effects were observed for scores for contentment [β^=0.0017, t498 = 7.37, Punc = 7.05E-13], surprise [β^=0.0010, t498 = 4.07, Punc = 5.51E-05], anger [β^=−0.0007, t498 = -3.36, Punc = 0.00085], and sadness [β^=−0.0034, t498 = -15.59, Punc < 2.52E-038]).
To determine whether emotional states exhibited consistent dynamics over the course of the scanning period, we fit smoothing spline models [25] for each subject and assessed the correlation between each subject and the average time course of other subjects in a cross-validation procedure. This analysis showed that there is substantial moment-to-moment variability in the time course of emotional states across subjects (which cannot simply be explained by scaling differences in the emotion models or resting-state data; see S3 Fig). Consistent with the linear models using time as a predictor, evidence for neutral brain states was most prevalent in the second scanning session, especially during a peak at the beginning of the run, whereas the time course for fear peaked at the beginning of the first run and decreased throughout the scanning session. The model for surprise exhibited a similar time course as neutral states but peaked at the end of the second run. Additionally, this analysis showed that evidence for sad classifications peaked in the middle of the first run and decreased over time. Overall, these time series revealed a gradual change in evidence from negative emotions (fear and sadness in run 1) to non-valenced or bi-valenced emotions (neutral and surprise in run 2).
To ensure that our emotion-specific brain states are not proxies for more general resting-state networks thought to subserve other functions, we examined the spatial overlap between our models and those commonly derived by connectivity-based analysis of resting-state fMRI data [26]. On average, we observed little overlap (Jaccard index = 13.1 ± 1.97% [s.d.]; range 10.8%–16.7%) with the seven most prominent networks found in resting-state data, implicating a substantial degree of independence.
To further establish the construct validity of the spontaneous emotional brain states, we reasoned that their incidence should vary with individual differences in self-reported mood and personality traits associated with specific emotions. We assayed depressive mood with the Center for Epidemiologic Studies Depression Scale (CESD) [27] and state anxiety using the State-Trait Anxiety Inventory State Version (STAI-S) [28], instructing participants to indicate how they felt during the resting-state scan itself. Binomial regression models revealed that higher depression scores were associated with increases in the frequency of sadness (β^=0.0025, t497 = 2.673, Punc = .0075, Fig 4A, see S4 Fig for scatter plots of predictions) and no other emotional state (all Punc > .24). State anxiety was associated with increasing classifications of fear (β^=0.0033, t497 = 2.608, Punc = .0091) and decreasing frequency of contentment (β^=−0.0031, t497 = -2.015, Punc = .0439). Viewing these beta estimates as odds ratios (computed as eβ^) reveals how a one-unit increase in self-reported mood is associated with differences in the occurrence of spontaneous emotional states. Applying this approach to CESD scores reveals that individuals with a score of 16 (the cutoff for identifying individuals at risk for depression) have 5.92% increased odds of being in a sad state compared to those with a score of 0. In more practical terms, this corresponds to approximately seven extra minutes a day of exhibiting a brain state that would be classified as sadness.
Drawing from the Revised NEO Personality Inventory (NEO-PI-R) [29], we focused personality trait assessment on the specific Neuroticism subfacets of Anxiety, Angry Hostility, and Depression, due to their discriminant validity [30], heritability [31], universality [32], and close theoretical ties to the experience of fear, anger, and sadness. We found that increasing Anxiety scores were associated with more frequent classification of fear (β^=0.003, t497 = 1.978, Punc = 0.0479, Fig 4B) and fewer classifications of anger (β^=−0.004, t497 = -2.407, Punc = 0.0161). Angry Hostility scores were positively associated with the number of anger classifications (β^=0.0042, t497 = 2.400, Punc = 0.0164). Depression scores were positively associated with the frequency of fear (β^=0.003, t497 = 2.058, Punc = 0.0396) and sadness (β^=0.0037, t497 = 2.546, Punc = 0.0109). These results provide converging evidence across both state and trait markers that individual differences uniquely and differentially bias the spontaneous occurrence of brain states indicative of fear, anger, and sadness.
Finally, we examined whether the predictions of our decoding models were consistent with self-report of emotional experience during periods of unconstrained rest. We conducted a separate fMRI experiment in which an independent sample of young adult participants (n = 21) performed an experience sampling task in the absence of external stimulation (Fig 5A). Participants were instructed to rest and let their mind wander freely with their eyes open during scanning. Following intervals of rest of at least 30 s, a rating screen appeared during which participants moved a cursor to the location on the screen that best indicated how they currently felt.
If spontaneous emotional states are accessible to conscious awareness, then scores should be greater for emotion models congruent with self-report relative to scores for models incongruent with self-report. Contrasting emotion models in this manner is advantageous from a signal detection standpoint because it minimizes noise by averaging across emotions, as some were reported infrequently or not at all in some subjects (see [33] for an analogous approach to predict the contents of memory retrieval during similarly unconstrained free-recall). To test our hypothesis, we extracted resting-state fMRI data from the 10-s interval preceding each self-report query and applied multivariate models to determine the extent to which evidence for the emotional brain states in this window predicted the participants’ conscious emotional experience.
Consistent with our hypothesis, we found that scores for models congruent with self-report were positive (0.016 ± 0.0093 [s.e.m.], z = 2.068, Punc = 0.0386; Wilcoxon signed rank test), whereas scores for incongruent models were negative (-0.0048 ± 0.0017 [s.e.m.], z = -3.041, Punc = 0.0024). Classification of individual trials into the seven emotion categories exhibited an overall accuracy of 27.9 ± 2.1% (s.e.m.) of trials, where chance agreement is 21.47% (Punc = 0.001; binomial test). Not only do these results demonstrate that classification models are sensitive to changes in emotional state reported by participants, but also that there is selectivity in their predictions, as negative scores indicate evidence against emotion labels that are incongruent with self-report. Establishing both sensitivity and selectivity is important for the potential use of these brain-based models as diagnostic biomarkers of emotional states.
As an additional validation of our decoding models, we examined the correspondence between the prevalence of individual emotional brain states as detected via pattern classification and participant self-report. Classifications based on self-report and multivariate decoding yielded similar frequency distributions (Fig 5C), in which neutral and amusement were the most frequent. We found a positive correlation between the frequency of classifications based on participant ratings and multivariate decoding (r = .3876 ± 0.102 [s.e.m.], t20 = 2.537, Punc = .0196; one sample t test), further demonstrating a link between patterning of brain states and subjective ratings of emotional experience in the absence of external stimuli or contextual cues.
Converging findings from our experiments provide evidence that brain states associated with distinct emotional experiences emerge during unconstrained rest. Whereas prior work has decoded stimulus-evoked responses to emotional events, our study demonstrates that spontaneous neural activity dynamically fluctuates among multiple emotional states in a reliable manner over time. Observing such coherent, emotion-specific patterns in spontaneous fMRI activation provides evidence to support theories that posit emotions are represented categorically in the coordinated activity of separable neural substrates [34,35].
Validating the neural biomarkers in the absence of external stimulation suggests that they track information of functional significance, and do not merely reflect properties of the stimuli used in their development. It is possible that these classifiers detect the endogenous activity of distributed neural circuits, consistent with recent views that emotions are not represented in modular functional units [36,37]. However, the extent to which such activity is the result of innate emotion-dedicated circuitry, a series of cognitive appraisals, or constructive processes shaped by social and environmental factors remains to be determined (for a review of these viewpoints, see [38]). Regardless of the relative influence of such factors, the present findings suggest that the emotion-specific biomarkers track the expression of functionally distinct brain systems, as opposed to idiosyncrasies of the particular machine-learning problem.
Our findings complement recent studies demonstrating that a variety of emotion manipulations have lasting effects on resting brain activity [39–41]. For instance, one study revealed elevated striatal activity following gratifying outcomes in a decision-making task—an effect that was diminished in individuals with higher depressive tendencies [39]. Because these effects immediately followed emotional stimulation, they could plausibly reflect regulatory processes or lingering effects of mood. The present results, on the other hand, show that resting brain activity transiently fluctuates among multiple emotional states and that these fluctuations vary depending on the emotional status of an individual. Thus, emotional processes unfolding at both long and short time scales likely contribute to spontaneous brain activity.
Findings from our resting state experiment stand in contrast to recent work investigating emotion-specific functional connectivity [42]. In this study, whole-brain resting-state functional connectivity was assessed using seeds identified from a meta-analytic summary of emotion research [43]. This latter approach failed to reveal unique patterns of resting-state connectivity for individual emotions but showed that seed regions were commonly correlated with domain-general resting-state networks, such as the salience network [44]. In light of the present results, it is important to consider methodological differences between studies. Seed-based correlation highlights connectivity between brain regions whose time course of activation is maximally similar to the activity of a small number of voxels (which are averaged together to create a single time series), whereas pattern classification identifies combinations of voxels that maximally discriminate among mental states. Because individual voxels sample diverse neural populations [45], it is plausible that seed-based correlation is biased towards identifying networks that have large amplitudes in seeded regions as opposed to exhibiting specificity (e.g., see [46]). Thus, our approach may have greater sensitivity to detect discriminable categorical patterns.
Results of the experience sampling study provide external validation of our emotion-specific biomarkers [13]. Consistent with the resting-state study, the overall distribution of emotional states was clearly non-uniform, and classifications of neutral states occurred with high frequency. Beyond these commonalities, the inclusion of behavioral self-report led to differences in emotion-related brain activity. States of contentment and amusement were more frequently predicted during experience sampling compared to resting-state (46.31% versus 23.45%), a finding that was corroborated by higher ratings for these emotions in the self-report data. It is possible that this difference in the frequency of positive brain states is the result of a self-presentation bias [47], wherein participants may have employed emotion regulation in order to project a more positive image. Alternatively, it is possible that the self-reporting task requirement elicited more introspection between trials, which contributed to the pattern of altered emotional states [48]. Future work will be necessary to fully characterize how such cognitive-emotional interactions shape the landscape of emotional brain states [36,49].
We found that individual differences in mood states and personality traits are associated with the relative incidence of brain states associated with fear, anger, and sadness. These findings further establish the construct validity of our brain-based models of emotion and link subfacets of Neuroticism to the expression of emotion-specific brain systems. Given their sensitivity to individual differences linked to the symptomology of anxiety and depression, spontaneous emotional brain states may serve as a novel diagnostic tool to determine susceptibility to affective illness or as an outcome measure for clinical interventions aimed at reducing the spontaneous elicitation of specific emotions. This tool may be particularly useful to objectively assess the emotional status of individuals who do not have good insight into their emotions, as in alexithymia, or for those who cannot report on their own feelings, including patients in a vegetative or minimally conscious state.
All participants provided written informed consent in accordance with the National Institutes of Health guidelines as approved by the Duke University IRB. The resting state experiment was approved as part of the Duke Neurogenetics Study (Pro00019095) with an associated database (Pro00014717). The experience sampling project was approved separately (Pro00027404).
Classification of emotional states was performed using neural biomarkers that were developed based on blood oxygen level dependent (BOLD) responses to cinematic films and instrumental music [13]. This induction procedure was selected because it reliably elicits emotional responses over a 1 to 2 min period, as opposed to longer-lasting moods. These models were developed to identify neural patterning specific to states of contentment, amusement, surprise, fear, anger, and sadness (in addition to a neutral control state). These particular emotions were modeled to broadly sample both valence and arousal, as selecting common sets of basic emotions (e.g., fear, anger, sadness, disgust, and happiness) undersamples positive emotions. In selecting these particular emotions, we verified that the accuracy of these models tracked the experience of specific emotion categories (average R2 across emotions = .57) independent of subjective valence and arousal. Thus, the models offer unique insight into the emotional state of individuals and characterize the likelihood they would endorse each of the seven emotion labels, independent of general factors such as valence or arousal.
A total of 499 subjects (age = 19.65 ± 1.22 years [mean ± s.d.], 274 women) were included as part of the Duke Neurogenetics Study (DNS), which assesses a wide range of behavioral and biological traits among healthy, young adult university students. For access to this data, see information provided in S1 Text. This sample was independent of that used to develop the classification models. This sample size is sufficient to reliably detect (β = .01) a moderate effect (r = .2) with a type-I error rate of .05, which is particularly important when studying individual differences in neural activity. All participants provided informed consent in accordance with Duke University guidelines and were in good general health. The participants were free of the following study exclusions: (1) medical diagnoses of cancer, stroke, head injury with loss of consciousness, untreated migraine headaches, diabetes requiring insulin treatment, chronic kidney or liver disease, or lifetime history of psychotic symptoms; (2) use of psychotropic, glucocorticoid, or hypolipidemic medication; and (3) conditions affecting cerebral blood flow and metabolism (e.g., hypertension). Diagnosis of any current DSM-IV Axis I disorder or select Axis II disorders (antisocial personality disorder and borderline personality disorder), assessed with the electronic Mini International Neuropsychiatric Interview [50] and Structured Clinical Interview for the DSM-IV subtests [51], were not an exclusion, as the DNS seeks to establish broad variability in multiple behavioral phenotypes related to psychopathology. No participants met criteria for a personality disorder, and 72 (14.4%) participants from our final sample met criteria for at least one Axis I disorder (10 Agoraphobia, 33 Alcohol Abuse, 3 Substance Abuse, 25 Past Major Depressive Episode, 5 Social Phobia). However, as noted above, none of the participants were using psychotropic medication during the course of the DNS.
Participants were scanned on one of two identical 3 Tesla General Electric MR 750 system with 50-mT/m gradients and an eight channel head coil for parallel imaging (General Electric, Waukesha, Wisconsin, USA). High-resolution 3-dimensional structural images were acquired coplanar with the functional scans (repetition time [TR] = 7.7 s; echo time [TE] = 3.0 ms; flip angle [α] = 12°; voxel size = 0.9 × 0.9 × 4 mm; field of view [FOV] = 240 mm; 34 contiguous slices). For the two 4 min, 16 s resting-state scans, a series of interleaved axial functional slices aligned with the anterior commissure—posterior commissure plane were acquired for whole-brain coverage using an inverse-spiral pulse sequence to reduce susceptibility artifact (TR = 2000 ms; TE = 30 ms; α = 60°; FOV = 240 mm; voxel size = 3.75 × 3.75 × 4 mm; 34 contiguous slices). Four initial radiofrequency excitations were performed (and discarded) to achieve steady-state equilibrium. Participants were shown a blank gray screen and instructed to lie still with their eyes open, think about nothing in particular, and remain awake.
Preprocessing of all resting-state fMRI data was conducted using SPM8 (Wellcome Department of Imaging Neuroscience). Images for each subject were slice-time-corrected, realigned to the first volume in the time series to correct for head motion, spatially normalized into a standard stereotactic space (Montreal Neurological Institute template) using a 12-parameter affine model (final resolution of functional images = 2 mm isotropic voxels), and smoothed with a 6 mm FWHM Gaussian filter. Low-frequency noise was attenuated by high-pass filtering with a 0.0078 Hz cutoff.
A total of 22 subjects (age = 26.04 ± 5.16 years [mean ± s.d.], 11 women) provided informed consent and participated in the study. Data from one participant was excluded from analyses because of excessive head movement (in excess of 1 cm) during scanning. While no statistical test was performed to determine sample size a priori, this sample size is similar to those demonstrating a correspondence between self-report of affect and neural activity [13,52,53].
Participants engaged in an experience sampling task in which they rated their current feelings during unconstrained rest. Participants were instructed to keep their eyes open and let their mind wander freely and that a rating screen [54] would occasionally appear, which they should use to indicate the intensity of the emotion that best describes how they currently feel. This validated assay of emotional self-report consists of 16 emotion words organized radially about the center of the screen. Four circles emanate from the center of the screen to each word (similar to a spoke of a wheel), which were used to indicate the intensity of each emotion by moving the cursor about the screen. During four runs of scanning, participants completed 40 trials (10 per run) with an inter-stimulus interval (ISI) of 30 s plus pseudo-random jitter (Poisson distribution, λ = 4 s).
Self-report data were transformed from two-dimensional cursor locations to categorical labels. Polygonal masks were created by hand corresponding to each emotion term on the response screen. A circular mask in the center of the screen was created for neutral responses. Because terms in the standard response screen did not perfectly match those in the neural models, the item “relief” was scored as “content,” whereas “joy” and “satisfaction” were scored as “amusement.” The items “surprise,” “fear,” “anger,” “sadness,” and “neutral” were scored as normal.
Scanning was performed on a 3 Tesla General Electric MR 750 system with 50-mT/m gradients and an eight channel head coil for parallel imaging (General Electric, Waukesha, Wisconsin, USA). High-resolution images were acquired using a 3D fast SPGR BRAVO pulse sequence (TR = 7.58 ms; TE = 2.936 ms; image matrix = 2562; α = 12°; voxel size = 1 × 1 × 1 mm; 206 contiguous slices) for coregistration with the functional data. These structural images were aligned in the near-axial plane defined by the anterior and posterior commissures. Whole-brain functional images were acquired using a spiral-in pulse sequence with sensitivity encoding along the axial plane (TR = 2000 ms; TE = 30 ms; image matrix = 64 × 64; α = 70°; voxel size = 3.8 × 3.8 × 3.8 mm; 34 contiguous slices). Four initial radiofrequency excitations were performed (and discarded) to achieve steady-state equilibrium.
Processing of MR data was performed using SPM8 (Wellcome Department of Imaging Neuroscience). Functional images were slice-time-corrected, spatially realigned to correct for motion artifacts, coregistered to high resolution anatomical scans, and normalized to Montreal Neurologic Institute (MNI) space using high-dimensional warping implemented in the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm.html). Low-frequency noise was attenuated by high-pass filtering with a 0.0078 Hz cutoff.
To rescale data for classification, preprocessed time series were standardized by subtracting their mean and dividing by their standard deviation. Maps of partial least squares (PLS) regression coefficients from stimulus-evoked decoding models [13] were resliced to match the voxel size of functional data. These coefficients are conceptually similar to those in multiple linear regression, only they are computed by identifying a small number of factors (reducing the dimensionality of the problem) that maximize the covariance between patterns of neural activation and emotion labels (for specifics on their computation, see [55]). Classifier scores were computed by taking the scalar product of functional data at each time point and PLS regression coefficients from content, amusement, surprise, fear, anger, sad, and neutral models. Individual time points were assigned categorical labels by identifying the model with the maximal score.
In order to determine if relatively focal or diffuse patterns of resting-state activity informed classification, we computed importance maps for each subject (S1 Fig). This was accomplished by calculating the voxel-wise product between PLS regression coefficients for each emotion model and the average activity of acquisition time points labeled as the corresponding emotion. We made inference on these maps by conducting a mass-univariate one-sample t test for each of the seven models, thresholding at FDR q = .05.
To address the potential overlap of the emotion classification models and canonical resting-state networks of the brain, we computed the maximal Jaccard index for each emotion model and the seven most prominent resting-state networks identified in Yeo et al [26]. This index is computed as the intersection of voxels in the two maps (voxels above threshold in both maps) relative to their union (the number of voxels above threshold in either map). Thresholds for classification models were adaptively matched to equate the proportion of voxels assigned to each resting state network.
When conducting inferential tests on classification frequency (count data), non-parametric tests were conducted. To test whether classifications were uniformly distributed across the emotion categories, a Friedman test was performed (n = 499 subjects, k = 7 emotions). Wilcoxon signed-rank tests were performed to test for differences in frequency relative to chance rates (14.3%) in addition to pairwise comparisons between emotion models, and corrected for multiple comparisons based on the false-discovery rate.
Because the models have different levels of accuracy when used for seven-way classification [13], we additionally conducted wavelet resampling of classifier scores in the time domain [33,56] over 100 iterations to ensure that differences in the sensitivity of models did not bias results. This procedure involved scrambling the wavelet coefficients (identified using the discrete wavelet transform) of classifier scores (time series in Fig 3) to generate random time series with similar autocorrelation as the original data. Classifications were then made on these surrogate time series, and Friedman tests were performed to test for differences in frequencies across categories. This procedure yielded a null distribution for the chi-square statistic against which the observed statistic on unscrambled data was compared.
To test whether classifier scores changed over time, Friedman tests were conducted on the outputs of the emotion models separately (concatenating the time series across runs), as classifier scores were found to violate assumptions of normality. Follow-up tests on the direction of these changes (either as increases or decreases) were conducted using general linear models with one constant regressor and another for linearly increasing time for each subject. Inference on the parameter estimate for changes over time was made using a one-sample t test (498 degrees of freedom).
In addition to testing gradual changes over time, smoothing spline models [25] were used to characterize more complex dynamics of emotional states. Because spline models are flexible and may include a different number of parameters for each subject, cross-validation was conducted to assess the coherence of spline fits across subjects. In this procedure, a smoothing spline model was fit for each subject, and its Pearson correlation with the mean fit for all other subjects was computed. The average of resulting correlations accordingly reflects the coherence of nonlinear changes in emotional states across all subjects.
The influence of individual differences in mood and personality was assessed using generalized linear models with a binomial distribution and a logit link function. Multiple models were constructed, each using a single measure from either the CESD, STAI, or facets from the NEO-PI-R to predict the frequency of classifications for the seven emotion categories (seven models per self-report measure). Inference on parameter estimates (characterizing relationships between individual difference measures and classification frequency) was made using a t distribution with 497 degrees of freedom.
To control for multiple comparisons, FDR correction (q = .05) [57,58] was applied for targeted predictions. For individual differences in mood, this procedure included correction for positive associations between the frequency of sad classifications and CESD scores and between fear classification and STAI values (Pthresh = .0091). For differences in emotional traits, correction was applied to models predicting the frequency of fear classification on the basis of Anxiety scores, anger classification using Angry Hostility scores, and sad classifications on the basis of Depression scores (Pthresh = .0479). Scatterplots and predicted outcomes for these regression analyses are displayed in S4 Fig.
To assess concordance in the experience sampling study, classifier scores were averaged for trials congruent and incongruent with self-report for each subject. For instance, all trials in which a participant self-reported “fear,” the classifier outputs from the neural model predicting fear were considered congruent, whereas the remaining six models were averaged as incongruent. Because the frequency of self-report varied across emotions (e.g., endorsement of fear and sadness were very infrequent), scores were averaged across all trials to reduce noise.
In a supplemental analysis, scores were extracted separately for all trials and classified by identifying the model with the highest score. Accuracy was assessed on data from all subjects, using self-reports of emotion as ground truth. Because the frequency of self-reported emotions was non-uniform, chance agreement between self-report and neural models was calculated based on the product of marginal frequencies, under the assumption of independent observer classifications [59]. Inference on the observed classification accuracy was tested against this value using the binomial distribution B(480, 0.2147). Due to infrequent self-reports of surprise, fear, and anger, accuracy on individual models was not computed.
Scores were initially assessed by averaging the 10 s preceding each rating. Subsequent analyses increasing the window length up to 20 s did not alter results. Because the scores for congruent (p = 0.0186, Lilliefors test against normal distribution) and incongruent (p = 0.0453) trials exhibited non-normal distributions, Wilcoxon signed rank tests were used to test each sample against zero mean rank. The correspondence between the frequencies of classification labels from self-report and neural decoding was assessed by computing the Pearson correlation for each subject. The correlation coefficients were Fisher transformed and tested against zero using a one-sample t test.
To ensure that population differences (i.e., inclusion of individuals with psychopathology) did not contribute to differences in the prevalence of emotions in the resting-state and experience sampling studies, we re-calculated the frequency of classifications using repeated random subsampling of healthy participants in the resting-state sample (1,000 iterations, sampling 21 participants without replacement). The average correlation between the healthy subsamples and the full sample was very high (ravg = .981, s.d. = .013), making it unlikely that clinical status accounts for differences in the frequency of classifications across studies.
|
10.1371/journal.ppat.1005009 | Sequence-Specific Fidelity Alterations Associated with West Nile Virus Attenuation in Mosquitoes | High rates of error-prone replication result in the rapid accumulation of genetic diversity of RNA viruses. Recent studies suggest that mutation rates are selected for optimal viral fitness and that modest variations in replicase fidelity may be associated with viral attenuation. Arthropod-borne viruses (arboviruses) are unique in their requirement for host cycling and may necessitate substantial genetic and phenotypic plasticity. In order to more thoroughly investigate the correlates, mechanisms and consequences of arbovirus fidelity, we selected fidelity variants of West Nile virus (WNV; Flaviviridae, Flavivirus) utilizing selection in the presence of a mutagen. We identified two mutations in the WNV RNA-dependent RNA polymerase associated with increased fidelity, V793I and G806R, and a single mutation in the WNV methyltransferase, T248I, associated with decreased fidelity. Both deep-sequencing and in vitro biochemical assays confirmed strain-specific differences in both fidelity and mutational bias. WNV fidelity variants demonstrated host-specific alterations to replicative fitness in vitro, with modest attenuation in mosquito but not vertebrate cell culture. Experimental infections of colonized and field populations of Cx. quinquefaciatus demonstrated that WNV fidelity alterations are associated with a significantly impaired capacity to establish viable infections in mosquitoes. Taken together, these studies (i) demonstrate the importance of allosteric interactions in regulating mutation rates, (ii) establish that mutational spectra can be both sequence and strain-dependent, and (iii) display the profound phenotypic consequences associated with altered replication complex function of flaviviruses.
| West Nile virus (WNV) is the most geographically widespread arthropod-borne virus (arbovirus) in the world. Like most arboviruses, WNV is a RNA virus which is highly mutable and exists in nature as genetically diverse mutant swarms. Although many recent studies have investigated the relationship between virus mutation rate and viral fitness, this had not previously been determined for WNV or other flaviviruses. We identified WNV mutations associated with variation in mutation rate using cell culture passage in the presence of a mutagen and engineered these mutations into an infectious WNV clone in order to investigate the causes and consequences of altered fidelity. Our results demonstrate that interactions among proteins which comprise the WNV replication complex can significantly alter both the extent and types of mutations that occur. In addition, we show that both increasing and decreasing WNV fidelity has host-specific effects on replication in cell culture and is associated with nearly complete ablation of WNV infection in mosquito vectors. These results have significant implications for our understanding of arbovirus evolution, replication complex function and arboviral fitness in mosquitoes, and identify important targets to study the determinants and mechanisms of vector competence and arbovirus fidelity.
| Lack of proofreading mechanisms and high replication rates among most RNA viruses make them inherently error-prone, yet there is also variation in mutation rates among both species and strains of RNA viruses, making fidelity itself a trait with a genetic basis subject to some fine-tuning by selection [1–3]. The generally accepted belief is that genetic diversity provides a benefit for RNA viruses for which success depends on the capacity to effectively proliferate in a range of internal environments and evade host immunity [4–6]. Such plasticity could be particularly beneficial for arthropod-borne viruses (arboviruses), which require successful infection, replication and transmission by taxonomically divergent vertebrate and invertebrate hosts.
On the other hand, some would argue that mutation rate is simply coupled to replication rate and that the low fidelity of RNA viruses is not a requirement, but rather a consequence of selection for maximum replicative fitness [7]. There is indeed a clear relationship between replication fidelity and replication rate [8], but there is also evidence that the two can be uncoupled. For example, the high-fidelity variant of poliovirus (PV), G64S, was shown to have replicative kinetics equivalent to wildtype virus in vitro [9–11]. Pushing mutation rate beyond maximum replicative fitness creates a scenario in which genetic information is lost and selection can no longer outpace the accumulation of deleterious mutations, termed lethal mutagenesis [12,13]. Selection for mutational robustness could buffer somewhat against the negative impacts of increased mutational load, yet there is clearly a limit to this, as demonstrated by the effectiveness of ribavirin and other mutagenic antivirals against a range of RNA viruses [14,15]. In addition, previous studies demonstrate that mutator variants of chikungunya virus (CHIKV), coxsackievirus (CV), SARS-coronavirus and PV are highly attenuated [16–19]. Conversely, high-fidelity variants of PV, CV, foot and mouth disease virus and CHIKV have also been shown to be attenuated in various hosts [10,20–25], suggesting that there is likely a delicate balance between the need for accuracy and diversity among RNA viruses.
With the exception of important studies with CHIKV, studies directly evaluating the phenotypic impact of mutation rates of arboviruses are lacking. Given the species-specific differences in selective pressure and virus-host interactions, there is clearly a need to individually characterize these relationships for other medically important arboviruses [26]. In addition, direct evidence linking specific arbovirus mutations to biochemical alterations affecting fidelity have not been presented, and therefore the mechanism of altered fidelity, including the role of specific structural changes in the RNA-dependent RNA polymerase (RdRp) and allosteric interactions with other proteins in the replication complex, are not well defined. Lastly, there is intriguing evidence that not just the effect of altered mutation rate, but fidelity itself could be host or even cell-specific [27], which could be particularly relevant for arboviruses.
West Nile virus (WNV; Flaviviridae, Flavivirus) is the most geographically widespread arbovirus in the world and there remains no effective therapeutics or prophylactics against WNV disease in humans. Although WNV is one of the most well characterized arboviruses in terms of evolution and host-virus interactions, there are gaps in our understanding of host-specific selective pressures and genetic correlates of viral fitness and pathogenesis. While there is evidence of superior WNV fitness in mosquitoes with highly homogeneous strains [28], and the accumulation of diversity in mosquitoes could simply be a product of relaxed purifying selection as a result of mutational robustness [29], there is also evidence for a correlation between WNV fitness and intrahost diversity in mosquito cell culture and Culex mosquitoes [30–32]. Increased diversity has also been associated with decreased WNV virulence in mice [31], suggesting that altering the capacity to accumulate mutations could have host-specific phenotypic consequences.
In order to gain insight into the phenotypic correlates and mechanism of WNV mutation restriction and expansion, we utilized experimental evolution in the presence of the antiviral ribavirin to identify mutations in the WNV replication complex important in regulating fidelity and characterized WNV mutants possessing these changes. Our results provide new insight into the specificity of genome replication and fidelity, the importance of allosteric changes in the regulation of mutation, and the host-specific consequences of alterations to fidelity.
A WNV infectious clone (WNV-IC), generated from WNV strain 3356 (NY99; AF404756) as previously described [33] was serially passaged in HeLa cells (ATCC) in the presence of the antiviral nucleoside analog ribavirin (Sigma) in duplicate (Lineage I & II). Both lineages were passaged in the presence of 50, 100 and 250uM ribavirin, and the virus with the highest infectious titer at 5 days post-infection (pi) was used to initiate the subsequent passage at all concentrations of ribavirin. Ribavirin-treated HeLa cell monolayers were also infected with fresh stock of WNV-IC at each passage as a naive control for comparative antiviral resistance of serially passaged virus. A multiplicity of infection (MOI) of 0.1 was used to initiate all passages and resistance assays. In addition to ribavirin, susceptibility to 50uM 5-fluorouracil (Sigma) was also determined for select WNV strains. HeLa cells were grown in EMEM supplemented with 100ug/ml penicillin streptomycin and 2% fetal bovine serum (FBS). For all cells treated with antiviral compounds growth media was removed and monolayers in 6-well cell culture plates were overlaid with 1ml media containing the antiviral compound and incubated for 1h at 37°C. Media was then removed and replaced with 100ul of virus diluted in media supplemented with antiviral compound and incubated for 1 hour at 37°C. After incubation, 3mls of media supplemented with desired concentration of antiviral was added to each well. Supernatants were harvested at day 5 pi and titrated by plaque assay on African Green Monkey kidney (Vero) cells (ATCC) according to standard protocol [34]. In order to isolate clonal strains with decreased antiviral susceptibility, 20 individual plaques were harvested from Vero monolayers following the completion of passage 6, re-suspended in 100ul of EMEM, inoculated onto fresh Vero monolayers and grown in liquid culture for 96h. Susceptibility of mutagens was reported as log10 reduction and titer and compared using t-tests following confirmation of normality (GraphPad Prism, Version 5.0).
Full-genome consensus sequencing was performed in order to determine changes accrued with passage and to verify sequences of mutated viruses. RNA was extracted from cell culture supernatant and subjected to reverse transcription (RT) and polymerase chain reactions (PCR) using the SuperScript III one-step RT-PCR kit (Life technologies) with 5–10 overlapping fragments (sequences available upon request). RT-PCR products were concentrated using Zymo-5 DNA spin columns (Zymo Research). Sequencing was carried out using the same RT-PCR primer sets and all sequencing reactions were completed at the Wadsworth Center Applied Genomics Technology Core (WCAGTC) on an ABI 3100 or 3700 automated sequencer (Applied Biosystems). WNV amplicons of nucleotides 1311–3248 (envelope/ns1 genes; [29]) were created for deep-sequencing using the same methodology with WNV RNA isolated following 72 h growth on Aedes albopictus cells (C6/36, ATCC). C6/36 cells were used in order to maximize viral titer for sequencing and were grown in MEM supplemented with 10% FBS, 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 0.1mM non-essential amino acids, 100 U/ml of penicillin, and 100 ug/ml of streptomycin and maintained at 28°C in 5% CO2. Deep-sequencing was performed at the WCAGTC on the Ion Torrent Personal Genome Machine (IT-PGM) using a 316 semiconductor chip.
Sequences were compiled and edited using the SeqMan module of the DNAStar software package (DNAStar) and a minimum of two-fold redundancy throughout each clone or consensus fragment was required for sequence data to be considered complete. Ion Torrent generated sequence data was analyzed by the Wadsworth Center Bioinformatics Core facility using CLC Genomics Workbench (CLC bio) software. Quality trimming of sequence reads, reference mapping and SNP (single-nucleotide polymorphism) detection was done in CLC Genomics Workbench v5.0.1. Quality trimming and reference mapping were done with default parameters. Reference mapping was completed using the WNV-IC sequence (AF404756). SNP detection was performed with default parameters except minimum coverage was set to 20, minimum variant frequency was set to 1.0% and ploidy was set to 1.
WNV mutants including C8423T (T248I), G10057A (V793I), and G10096A (G806R) were generated by site-directed mutagenesis (SDM) of the WNV-IC using mutagenic primer sets along with the QuikChange XLII SDM kit (Stratagene) as per the manufacturer’s protocol. Mutant WNV-IC DNA was then amplified in E. coli and plasmid harvested by Highspeed Midiprep (Qiagen). Full-genome sequencing of NS5 mutant WNV-IC plasmids indicated no other mutations were present except those engineered. Transcription of mutant and control WNV-IC plasmids was carried out by linearization with Xba1 and transcription using the MEGAscript kit (Ambion) supplemented with Anti-reverse cap analog (Ambion) and assembled as per manufacturer’s protocol. Transcription reactions were incubated at 37°C for 4h. Resulting RNA was purified with the MEGAclear kit (Ambion) and quantified on a Nanodrop 2000 (Thermo Fisher Scientific). RNA was stored in 10μg aliquots at -80°C. Wild-type WNV-IC RNA and mutant RNA were electroporated into 0.8 x 107 C6/36 cells in PBS using a GenePulser (BioRad). Transfected cells were seeded into T75 flasks and supernatants were collected from day 3 to 7 post-transfection, aliquoted and stored at -80°C. WNV infectious particles were quantified by plaque titration on Vero cells.
The WNV NS5 gene was cloned into the pET26Ub-CHIS bacterial expression plasmid [35]. This system allows for the production of ubiquitin fusion proteins containing a carboxy-terminal hexahistidine tag that are then co and/or post-translationally processed by the ubiquitin protease, co-expressed from a second plasmid, pUBPS. Briefly, the WNV NS5 coding region was amplified using the WNV NY99 strain (AF404756) as template, oligonucleotides 1 and 2 (Table 1) and Deep Vent DNA polymerase (NEB). The PCR product of WNV NS5 was gel purified and cloned into the pET26Ub-CHIS plasmid using SacII and BamHI sites and by using In-Fusion ligation independent cloning. The final construct (pET26Ub-WNV NS5-CHIS) was confirmed by sequencing at the Pennsylvania State University’s Nucleic Acid Facility. Expression plasmids for the WNV NS5 derivatives (T248I and V793I, G806R) were constructed using the same strategy.
E. coli Rosetta(DE3)pUBPS cells were transformed with the pET26Ub-WNV-NS5-CHIS plasmid for protein expression. Rosetta(DE3)pUBPS cells containing the pET26Ub-WNV-NS5-CHIS plasmid were grown in 100 mL of media (NZCYM) supplemented with kanamycin (25 μg/mL), chloramphenicol (20 μg/mL) and spectinomycin (S50) at 37°C until an OD600 of 1.0 was reached. This culture was then used to inoculate 1L of K75, C60, S150-supplemented ZYP-5052 auto-induction media studier [36], to an OD600 = 0.1. The cells were grown at 37°C to an OD600 of 0.8 to 1.0, cooled to 15°C and then grown for 36–40 h. Typically, after 36–40 h at 15°C the OD600 reached ~7.0–10.0. Cells were harvested by centrifugation (6000 x g, 10 min) and the cell pellet was washed once in 200 mL of TE (10 mM Tris, 1 mM EDTA), centrifuged again, and the cell paste weighed. Typically, yields were 20 g of cell paste per liter of culture. The cells were then frozen and stored at -80°C until used. Frozen cell pellets were thawed on ice and suspended in lysis buffer (100 mM potassium phosphate, pH 8.0, 500 mM NaCl, 5 mM 2-mercaptoethanol, 20% glycerol, 1.4 μg/mL leupeptin, 1.0 μg/mL pepstatin A and one Roche EDTA-free protease tablet per 10 g cell pellet), with 5 mL of lysis buffer per gram of cells. The cell suspension was lysed by passing through a French press (SLM-AMINCO) at 15,000 psi. After lysis, phenylmethylsulfonylfluoride (PMSF) and NP-40 were added to a final concentration of 1 mM and 0.1% (v/v), respectively. While stirring the lysate, polyethylenimine (PEI) was slowly added to a final concentration of 0.25% (v/v). The lysate was stirred for an additional 30 min at 4°C after the last addition of PEI, and then centrifuged at 75,000 x g for 30 min at 4°C. The PEI supernatant was decanted to a fresh beaker, and while stirring, pulverized ammonium sulfate was slowly added to 60% (w/v) saturation. This supernatant was stirred for 30 min after the last addition of ammonium sulfate, and centrifuged at 75,000 x g for 30 min at 4°C. The supernatant was decanted, and the pellet was suspended in buffer A (100 mM potassium phosphate, pH 8.0, 500 mM NaCl, 5 mM 2-mercaptoethanol, 20% glycerol, 1.4 μg/mL leupeptin, 1.0 μg/mL pepstatin A, 5 mM imidazole). The resuspended ammonium sulfate pellet was loaded onto a Ni-NTA column (Qiagen) at a flow rate of 1 mL/min (approximately 1 mL bed volume/100 mg total protein) equilibrated with buffer A. After loading, the column was washed with fifty column volumes of buffer A and five column volumes of buffer A containing 50 mM imidazole. Protein was eluted from the Ni-NTA column with buffer A containing 500 mM imidazole. Fractions were collected and assayed for purity by SDS-PAGE. Fractions with the highest purity were pooled and dialyzed against buffer B (50 mM HEPES pH 7.5, 500 mM NaCl, 5 mM 2-mercaptoethanol and 20% glycerol; MWCO of 24,000 Da). The dialyzed protein was then passed thru a Hi-Load 16/600 Superdex 200 prep grade gel filtration column (GE Healthcare) equilibrated with buffer B at 1 ml/min using an AktaPrime system. Fractions (3 mL) were collected, assayed for purity by SDS-PAGE, pooled and then concentrated to 150 μM (~15 mg/mL) using a Vivaspin concentrator (30,000 MWCO). The protein concentration was determined by measuring the absorbance at 280 nm by using a Nanodrop spectrophotometer and using a calculated molar extinction coefficient of 221,730 M-1 cm-1. Purified, concentrated protein was aliquoted and frozen at -80°C until use. Typical WNV NS5 yields were 1 mg/5 g of E. coli cells, which can be produced from 0.25 L of culture.
To assemble WNV NS5 elongation competent complexes, 1 or 5 μM WNV NS5 was mixed with 10 μM pGGC RNA primer, 1 μM RNA template (either 5’-AAACUGAGAAGGAGAAAGCC-3’ or 5’-AAAUCGAGAAGGAGAAAGCC-3’), 20 μM CTP, 20 μM UTP and 0.1 μCi/μL [γ-32P]-UTP for 30 min in 50 mM HEPES pH 7.5, 5 mM MgCl2 and 10 mM 2-mercaptoethanol. For single nucleotide incorporation assays, the NS5 elongation competent complex was mixed with 25 μM heparin, 50 mM NaCl and 100 μM NTP substrate (either ATP or GTP) in 50 mM HEPES pH 7.5, 5 mM MgCl2 and 10 mM 2-mercaptoethanol at 30°C. After mixing, reactions were quenched at various times by the addition of 50 mM EDTA. Products were resolved from substrates by denaturing PAGE. An equal volume of loading buffer, 5 μL, (70% formamide, 0.025% bromophenol blue and 0.025% xylene cyanol) was added to 5 μL of quenched reaction mixtures and heated to 70°C for 2–5 min prior to loading 5 μL on a denaturing 20% polyacrylamide gel containing 1X TBE (89 mM Tris base, 89 mM boric acid, 2 mM EDTA) and 7 M Urea. Electrophoresis was performed in 1X TBE at 90 W. Gels were visualized by using a PhosphorImager (GE) and quantified by using ImageQuant software (GE).
Data were fit by nonlinear regression using the program KaleidaGraph (Synergy Software). All experiments shown are representative, single experiments that have been performed after at least three individual trials to define the concentration or time range shown. In all cases, values for parameters measured during individual trials were within the limits of the error reported for the final experiments. Kinetic data were fit by nonlinear regression using the program KaleidaGraph (Synergy Software, Reading, PA). Observed rate constants (kobs) for nucleotide incorporation were obtained by fitting product-versus-time data to an equation defining a single exponential (Eq 1), where A is the amplitude, kobs is the observed rate constant and C is the endpoint.
Confluent monolayers of baby hamster kidney cells (BHK; ATCC) and Culex tarsalis mosquito cells (CxT; kindly provided by A. Brault, CDC Fort Collins) were infected with virus, in triplicate, using 6-well plates, at a MOI of 0.01 pfu/cell (multi-step), 10.0 pfu/cell (BHK one-step), or 8.0 pfu/cell (CxT one-step). BHK cells were grown in minimal essential medium (MEM, Gibco) supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 1.5g/L sodium bicarbonate, 100 U/ml of penicillin, and 100 ug/ml of streptomycin and maintained at 37°C in 5% CO2. CxT cells were grown in Schneider’s media (GIBCO) supplemented with 10% FBS, 2 mM L-glutamine, 1.5 g/L sodium bicarbonate and maintained at 28°C in 5% CO2. After a one hour absorption period at 37°C (BHK) or 28°C (CxT), the inoculum was removed, cells were gently washed, overlaid with 2 ml of maintenance media and incubated at appropriate temperatures. Samples consisting of 50ul supernatant were harvested at 24, 48, 72, 96, and 120 (CxT only) hpi for multi-step growth kinetics and 3, 6, 9, 12, 24 and 30 (CxT only) hpi for one-step kinetics, diluted 1:10 in media containing 20% FBS, and stored at -80°C. Titrations were performed in duplicate, by plaque assay on Vero cells and mean titers for each time point were calculated. WNV RNA genomes were also quantified following extraction with QIAamp viral RNA spin columns (Qiagen) using a TaqMan quantitative real-time RT-PCR assay (Applied Biosystems) with a primer/probe designed for WNV E gene amplification [37]. Growth kinetics were compared using repeated measured ANOVA and Tukey’s post hoc tests and infectivity was compared by t-test following confirmation of normality (GraphPad Prism, Version 5.0).
Mosquito susceptibility was determined as previously described [38] in both highly colonized Cx. quinquefasciatus originally obtained from Benzon Research Inc. and F4 Cx. quinquefasciatus collected as egg rafts from Orange County, CA (kindly provided by Robert Cummings, Orange County California Vector Control District). Briefly, individual WNV strains were diluted to equivalent titers (~8.0 log10 pfu/ml), mixed 1:5, virus: defibrinated sheep blood (Colorado Serum) + 2.5% sucrose, and offered to ~500 female mosquitoes using a Hemotek membrane feeding system (Discovery Workshops). Following 1 h, mosquitoes were anesthetized using CO2 and fully engorged mosquitoes were saved and housed at 27°C for subsequent testing. Twenty-five to 50 mosquitoes per strain were saved at -80°C in 1 ml mosquito diluent [MD; 20% heat-inactivated fetal bovine serum (FBS) in Dulbecco’s phosphate-buffered saline (PBS) plus 50 μg/ml penicillin/streptomycin, 50 μg/ml gentamicin, and 2.5 μg/ml Fungizone] at days 5, 7, 10, 14 and 21 days post-feeding. Samples were thawed and homogenized for 30 seconds at 24hz in a Mixer Mill MM301 (Retsch). Debris was then pelleted by centrifugation at 6000 rcf for 5 minutes and the supernatant screened by plaque assay on Vero cells to determine infection status.
Passaging in the presence of the antiviral ribavirin was used to select for WNV variants with decreased susceptibility and putative alterations to polymerase fidelity. Ribavirin susceptibility, as measured by reduction in viral titer following treatment, was monitored throughout the passage series and again assessed following the completion of passage 6 (Fig 1A). Significantly lower reductions in viral titer relative to WNV-IC controls were measured in both lineage I and II following passage 5 and 6 (t-test, df = 5, p<0.05), such that lineage I viral titer decreased just 1.4-fold following ribavirin treatment after 6 passages, as compared to a greater than 25-fold titer reduction measured for WNV-IC. In order to select for clonal strains with decreased mutagen susceptibility, ribavirin sensitivity was assessed for individual biological clones isolated from the lineage I passed virus and strains with the highest levels of resistance (WNV pp3, pp9; Fig 1B) were chosen for further characterization.
In order to identify shared WNV amino acid (aa) substitutions associated with mutagen resistance, full-genome sequencing of clonal strains WNV pp3 and WNV pp9 was completed. A total of 15 (pp3) and 18 (pp9) nt substitutions were identified, resulting in 9 and 8 aa substitutions, respectively (Table 2). Of these, 10 nt and 7 aa substitutions were shared. Given the assumption that substitutions outside of the replication complex were more likely to be associated with adaptation to Hela cell culture or drift, shared aa substitution in the WNV RdRp and methyltransferase (Mtase) genes exclusively were chosen for further characterization. These included C8423T, resulting in a threonine to isoleucine change at position 248 of the Mtase, as well as G10057A and G10096A, resulting in valine to isoleucine and glycine to arginine changes at positions 793 and 806 of the RdRp, respectively (Table 2). Mapping of these residues on the known flavivirus RdRp and Mtase structures demonstrates that T248I is located at the C-terminal loop (aa 245–267), which is expected to interact with the RdRp domain, and both V793I and G806R exist in locations outside of the RdRp active site, although are within the priming loop (Fig 2; [39–42]). Both T248 and G806 are conserved among lineage I WNV strains, yet not across lineages or species, while V793 is shared among flaviviruses. No naturally circulating strains were found to possess the identified mutations at these locations.
To confirm that these amino acid substitutions independently conferred decreased ribavirin susceptibility, and to assess if this corresponded to generalized mutagen resistance, mutations were engineered independently and in combination into the WNV-IC and susceptibility to both ribavirin and 5-fluorouracil was assessed with WNV mutants. Full-genome sequencing following mutagenesis confirmed the exclusive presence of desired mutations. WNV mutants created included WNV T248I, V793I, G806R, double mutant V793I/G806R, and triple mutant (T248I/V793I/G806R). Mutagen resistance assays demonstrated significantly decreased reduction in titer (susceptibility) relative to untreated controls for all mutant strains as compared to WNV-IC following treatment with both ribavirin and 5-fluorouracil (t-test, df = 6 p<0.05; Fig 3). The highest level of mutagen resistance was measured with the RdRp double mutant, WNV V793I/G806R, for which 2.2 and 4.6-fold mean titer reduction were measured following ribavirin and 5-fluorouracil treatments, respectively, as compared to 55 and 37-fold mean titer reductions measured with WNV-IC (Fig 3).
In order to test the hypothesis that mutagen resistance corresponded to alterations in mutation rate, deep-sequencing was used to quantify accumulation of unique SNPs following a single passage on mosquito cell culture (Fig 4A). Levels of WNV RNA for both WNV-IC and mutant strains were statistically similar for all samples chosen for sequencing (~9.0 log10 WNV copies/ml), suggesting differences in replication were not likely to account for differences in the number of mutations accumulated. Assays were completed in duplicate for each WNV strain and sequencing results identified fewer SNPs in all mutant strains relative to WNV-IC, with the exception of the methyltransferase mutant, WNV T248I, for which the number of unique SNPs identified was approximately 2.5 fold higher than WNV-IC. The RdRp double mutant, WNV V793I/G806R, showed the fewest number of unique SNPs; approximately 2.5 fold lower than WNV-IC (Fig 4B). Mutations were distributed throughout the sequenced regions for all WNV strains, yet mutant swarm composition varied significantly among strains (Fig 4C). Specifically, transition to transversion ratios were ~2:1 for WNV-IC and WNV T248I, yet <1 for WNV V793I/G806R. Decreased mutation of WNV V793I/G806R relative to WNV-IC resulted primarily from lack of U to C and G to A mutations, which accounted for 11/28 mutations for WNV-IC and 0/12 mutations for WNV V793I/G806R (chi-squared, p = 0.011). Increased mutation of WNV T248I, on the other hand, resulted primarily from A misincorporations, which accounted for 33/62 mutations for WNV T248I, and just 7/28 mutations of WNV-IC (chi-squared, p = 0.045). Approximately the same number of U to A mutations were identified for WNV V793I/G806R as WNV T248I (4 vs 3), yet no G to A mutations were identified for WNV V793I/G806R, in stark contrast to the 20 identified for WNV T248I (Fig 4C). These results demonstrate strain-specific differences in the misincorporation of different nucleotides and, more specifically, in the propensity for particular mispairs, suggesting mutation frequencies may be dependent on both replication complex and template sequences and/or context.
Purified IC, V793I/G806R, and T248I WNV NS5 proteins were used to quantify and compare the kinetics of nucleotide misincorporation of NS5 elongation—competent complexes (Fig 5). Complexes were assembled using a 5’-phosphorylated trinucleotide primer (pGGC), single stranded RNA template, UTP and CTP (Fig 5A). The template was designed such that two nucleotides led to production of a 15-mer RNA. Labeling of the elongation complex was achieved by using α-32P-labeled nucleotide. Once assembled, the complex was stable and capable of rapid incorporation of the next correct nucleotide substrate (elongation) to produce a 16-mer RNA product. The elongation competent complex was then used to interrogate the kinetics of nucleotide misincorporation. The initial substrate, designed to measure G:U mispairs was used to quantify GMP misincorporation for each NS5 protein (Fig 5B and 5C). Comparing the percentage of RNA product produced over time, it was demonstrated that WNV-IC, V793I/G806R and T248I NS5 proteins displayed similar in vitro kinetics for GMP misincorporation (Fig 5C). These results were consistent with a lack of biological differences in fidelity among NS5 variants, but were also in agreement with deep-sequencing data, for which the number of A to G substitutions were similar among WNV-IC, WNV V793I/G806R, and WNV T248I (Fig 4C). In order to evaluate fidelity differences implied by deep-sequencing data, the template for biochemical assays was redesigned to quantify A:C mispairs, equivalent to G to A substitutions. The number and proportion of G to A substitutions identified following growth in mosquito cells differed among strains, with means of 6, 0 and 20 identified in WNV-IC, WNV V793I/G806R, and WNV T248I, respectively (Fig 4C). Biochemical assays were consistent with these results, clearly demonstrating an increasing rate of A misincorporation for WNV T248I relative to WNV-IC and a decreasing rate of A misincorporation for WNV V793I/G806R (Fig 5C). Taken together, these results demonstrate sequence-specific fidelity differences among WNV mutant strains.
Comparison of one-step and multi-step growth kinetics of WNV mutants to WNV-IC in vertebrate (BHK) and invertebrate (CxT) cell lines demonstrates host-specific effects of replication complex mutations (Fig 6). No differences in overall kinetics (repeated measures ANOVA, F = 0.14, df = 6, p = 0.98) or viral titers at individual time points (t-test, p>0.05) were measured when comparing WNV mutants to WNV-IC on vertebrate cell culture, while significant differences in viral kinetics were measured in mosquito cell culture at both MOIs for all replication complex mutants relative to WNV-IC (repeated measures ANOVA, F = 12.83, df = 6, p<0.0001). Specifically, consistently lower viral titers were measured for all RdRp and Mtase mutants in mosquito cells relative to WNV-IC (Tukey’s multiple comparison test, p<0.05), and kinetics were similar among mutants with the exception of WNV V793I, which despite having significantly lower titers than WNV-IC had modestly higher titers relative to other mutants (Fig 6). In addition, WNV RNA was quantified with qRT-PCR following one-step growth and particle/pfu ratios were quantified and compared among WNV-IC, WNV V793I/G806R and WNV T248I in order to assess the relationship between fidelity and infectivity. Specific infectivity was elevated in mosquito cells as compared to vertebrate cells for all WNV strains (Fig 7). Trends measured with infectivity are consistent with identified fidelity differences in both cell lines, with the highest infectivity measured with the RdRp double mutant WNV V793I/G806R and the lowest infectivity measured with the Mtase mutant WNV T248I, yet differences were only significant relative to WNV-IC for WNV T248I in BHK cells (t-test, df = 4, p = 0.0028; Fig 7). These differences do not therefore account for differences in growth kinetics identified among WNV strains in CxT cells.
In order to determine if modest attenuation in mosquito cell culture and fidelity differences corresponded to differences in mosquito competence in vivo, infectivity of WNV-IC was compared to infectivity of WNV T248I and WNV V793I/G806R in colonized Cx. quinquefasciatus mosquitoes following exposure to infectious bloodmeals. Input titers were comparable to a natural dose and similar among WNV strains and experimental replicates (Table 3; [43]). Despite the fact that levels of infection were somewhat lower than have been measured with other wildtype WNV strains (35.8%), stark and highly significant differences were measured when compared to WNV mutants (chi-squared, p<0.001 for all mutants relative to WNV-IC). Specifically, over the course of 3 experimental replicates and multiple time points, a total of just 2 of 210 (V793I/G806R) and 3 of 203 (T248I) mosquitoes acquired measurable infections. To confirm that this was generalizable phenotype which was relevant in natural populations, infectivity experiments were repeated in Cx. quinquefasciatus recently acquired from the field. Although modestly lower input titers were used, infectivity was slightly higher for WNV-IC relative to colonized mosquitoes (39.2%) and a general lack of infectivity was confirmed for replication complex mutants (Table 3). Given the inefficient infectivity of mutant strains, dissemination and transmission were not evaluated in this study.
Although viral load was not determined for individual mosquitoes, it is notable that the 8 total mosquitoes identified as positive following exposure to WNV mutants all showed relatively low levels (less than 20 plaques) with undiluted plaque screens.
As has been successfully accomplished with other systems [21,22,24,25,44], we exploited selection in the presence of a mutagen to identify mutations altering WNV replicase fidelity and utilized WNV fidelity mutants to interrogate the consequences and mechanisms of altered mutation rates. Although ribavirin is not considered an efficacious antiviral for the treatment of active WNV infections, it has been shown to cause both error-prone replication and WNV attenuation in vitro, particularly in Hela cell culture [45].
The relative decrease in mutation frequency measured for the WNV high fidelity variant V793I/G806R (~2.5 fold) is similar or modestly higher than has been shown with other systems [21,22,24,25,44]. With the exception of coronaviruses, which employ a proofreading exoribonuclease system unique among RNA viruses [46], these data together demonstrate that either the lack of biochemical capacity or the extent of phenotypic consequences by-in-large prevent highly significant alterations to RdRp fidelity. Despite this, data presented here and in previous studies clearly demonstrate that subtle alterations to mutation rates can have profound phenotypic effects on RNA viruses.
Although fully characterizing the mechanism by which these RdRp residues alter fidelity would require further biochemical and biophysical investigations, mapping of T248, V793 and G805 residues on the crystal structures of NS5 [41,42] provides some indication of possible mechanisms (see Fig 2). Despite the lack of full-length NS5 structure from WNV, the relative orientation of the Mtase domain with respect to the polymerase domain can be defined using the recent crystal structure of the full length dengue virus (DENV) NS5 [47] as a guide. The two NS5 structures are highly similar with an RMSD of 1.18 and 0.65 Å for the polymerase and Mtase domains, respectively. The residues V793 and G805 are located in the priming loop (aa 789–812), which is a long loop that links two α-helices in the thumb subdomain and protrudes to reach the active site. Conformational dynamics of the priming loop is believed to be necessary to form a stable initiation complex [42]. A model of the initiation complex is shown in Fig 2B; in this model a 4-mer ssRNA substrate, taken from the complex structure of the related ɸ6-RdRp (PDB 1HI0) [48], and rNTP modeled at the priming site (P-site) and catalytic site (C-site) based on the HCV RdRp complex structure (PDB 1GX5) [40] can be accommodated in the WNV NS5 RdRp active site with minimal steric clashes with the protein atoms. To form a stable initiation complex, the active-site residue Trp-800 would alter its sidechain conformation to be able to stack against the priming nucleotide. The priming loop maintains numerous interactions with residues from the thumb, fingers and palm subdomains and substitutions similar to V793I and G806R could potentially disrupt these interactions, impacting the dynamics of the priming loop and subsequently affecting the initiation process of the RdRp. It is not difficult to conceive that V793I and G806R may restrict the dynamics of the priming loop, leading to a higher fidelity mutant in a scenario similar to what has recently been shown for the G64S high-fidelity mutant of PV RdRp [49,50]. In these studies of PV RdRp, remote site mutations altered the polymerase fidelity by causing changes to the dynamics of conserved structural elements and motifs including residues at the active site. Although the interactions of the flavivirus RdRp and Mtase have now been well-documented [51,52] the finding that modifications to allosteric interactions resulting from mutation of a single residue of the Mtase can significantly alter replication fidelity is novel. The T248I mutation is located at the C-terminal loop (aa 245–267) of the MTase domain; which is expected to be at the interface between the two domains of the WNV NS5 (Fig 2C), similar to what is observed in the homologous DENV NS5 structure. The loop harboring T248 is predicted to interact with the region of the fingers in the polymerase domain (aa 350–365). It is very likely that amino acid substitution of T248 by an isoleucine could affect the interactions between the two domains and the inter-domain dynamics, eventually affecting the polymerase active site and altering fidelity. Findings with WNV are therefore consistent with previous data demonstrating that RdRp fidelity is determined by a complex network of interactions and checkpoints by which remote site mutations may alter the dynamics of conserved structural elements and motifs including residues at the active site [9,49,53,54].
Although selection for ribavirin resistance did, as predicted, result in the isolation of high fidelity WNV variants, the fact that a mutator variant also displayed resistance could be explained by antiviral mechanisms independent of lethal mutagenesis for WNV in this system. Similar results were attained with FMDV, for which a low fidelity RdRp was found to have a decreased capacity for ribavirin incorporation [55]. In addition, previous studies with another flavivirus, yellow fever virus, demonstrate that the antiviral actions of ribavirin are conferred primarily by the depletion of intracellular GTP pools [56]. Additional antiviral mechanisms of ribavirin have also been proposed, including inhibition of virus transcription [57] and inhibition of both guanyltranferase and Mtase activity [58,59]. The flavivirus Mtase is required for RNA capping [60], a process partially enabled by GTP binding [61] and competitively inhibited by ribavirin with DENV NS5 [58]. Although T248 is not within the nucleotide binding site it is possible that this mutation could perturb these interactions and subsequently interfere with antiviral susceptibility in this manner. On the other hand, given that WNV T248I also displays resistance to 5-fluorouracil, it is possible that the strain-dependent mutational biases could result in unique evolutionary trajectories and, subsequently, strain-specific differences in mutational robustness and susceptibility to lethal mutagenesis.
This sequence-dependent nature of the fidelity alterations also demonstrates that broad assumptions about fidelity and mutagen susceptibility likely discount the specificity of interactions of individual nucleotides and/or base analogs with the replication complex. Although others have demonstrated that modifications to fidelity are attainable, the possibility that unique strains may possess unique mutational biases has novel functional and evolutionary implications. Specifically, if mutational landscapes are strain-specific, so too are fitness landscapes of viral swarms and therefore evolutionary pressures acting on them. Such biases could be exploited by evolution as a means of increasing the probability of producing favorable mutant swarms following genetic bottlenecks or could have the opposite effect of constraining deleterious strains by not permitting adequate exploration of sequence space to escape unfit landscapes.
Consistent with previous studies with CHIKV [24] in vitro kinetics were generally similar for the high fidelity WNV V793I/G806R relative to WNV-IC, with modest attenuation measured in mosquito but not vertebrate cell culture. Although fitness differences were only measurable with direct competition of CHIKV and not individual growth assays, the decreases in mutation rate measured for WNV V793I/G806R were also more substantial than those measured for CHIKV, likely due to combining two RdRp mutations which appear to have an additive effect on fidelity. These host-specific effects are consistent with previous studies demonstrating increased swarm diversity in the mosquito for both WNV and its close relative St. Louis encephalitis virus [31,62], but further suggest that the invertebrate environment is not simply a more robust environment which tolerates diversity, but one in which diversity itself likely provides a fitness benefit [30]. It is possible that this fitness benefit results from an inherent need to escape RNAi or other innate invertebrate immune responses [63], or that enhancements in fitness could result from cooperative interactions among distinct genotypes and viral proteins [64]. Despite this, previous passage studies in Cx. pipiens suggest that this need for diversity may be overcome by individual variants with highly superior fitness [28] and results presented here demonstrating attenuation of the low fidelity mutant WNV T248I suggest, not surprisingly, that there is a limit to the benefit of diversity. The association of mutator phenotypes with either similar or attenuated viral growth kinetics is consistent with what has been observed in other systems [17–19], and gives further credence to the idea that replication and mutation rate are not necessarily inextricably bound phenotypes. Given that vertebrate environments have been found to be more restrictive both in vitro and in vivo [32,65], it is somewhat surprising that a virus with a mutator phenotype would not also be attenuated in vertebrate cell culture, yet even if WNV T248I is more mutationally robust than WNV-IC, attenuation may be observed if this strain were repeatedly passaged, therefore accumulating diversity and, presumably, deleterious mutants [17,18]. Consistent with this is the fact that the Mtase mutant was also found to be less infectious in vertebrate cell culture. In addition, competition assays with increased sensitivity for detecting more subtle fitness differences [24,66] or in vivo models that more accurately represent natural infections could reveal important phenotypic differences in vertebrate systems [67]. Although in the current studies results confirm that inherent biochemical differences account for differences in mutation rate independent of cell type, it is also feasible that fidelity itself could be host-dependent, as a recent study with vesicular stomatitis virus demonstrates slower mutation rates in insect cells as compared to mammalian cells [27]. Although few have investigated this concept [68], it is not necessarily surprising that the biophysical and biochemical properties of the replication complex might differ significantly in environments with variable temperature, pH, and nucleotide availability. Future studies exploiting new sequencing technologies to evaluate mutation rates in a range of systems will help to clarify these differences [17,69,70].
Although the modest attenuation in mosquito cell culture may be explained by the modest alterations to fidelity, it is much more surprising that an approximately 2.5 fold alteration to mutation rate could almost entirely eliminate the capacity for infection and/or sustainable WNV replication in mosquitoes. Although studies with CHIKV also demonstrate that fidelity variants are associated with decreased infectivity in mosquitoes, differences measured for WNV here are much more profound. These results suggest either that WNV replication in gut epithelial cells is uniquely sensitive to alterations in fidelity or that alternative mechanisms of attenuation related to host interaction with the flavivirus NS5 exist. Regardless, these variants provide powerful tools to elucidate the determinants of flavivirus mosquito competence and novel targets for viral attenuation.
|
10.1371/journal.pgen.1003339 | A Role for the Malignant Brain Tumour (MBT) Domain Protein LIN-61 in DNA Double-Strand Break Repair by Homologous Recombination | Malignant brain tumour (MBT) domain proteins are transcriptional repressors that function within Polycomb complexes. Some MBT genes are tumour suppressors, but how they prevent tumourigenesis is unknown. The Caenorhabditis elegans MBT protein LIN-61 is a member of the synMuvB chromatin-remodelling proteins that control vulval development. Here we report a new role for LIN-61: it protects the genome by promoting homologous recombination (HR) for the repair of DNA double-strand breaks (DSBs). lin-61 mutants manifest numerous problems associated with defective HR in germ and somatic cells but remain proficient in meiotic recombination. They are hypersensitive to ionizing radiation and interstrand crosslinks but not UV light. Using a novel reporter system that monitors repair of a defined DSB in C. elegans somatic cells, we show that LIN-61 contributes to HR. The involvement of this MBT protein in HR raises the possibility that MBT–deficient tumours may also have defective DSB repair.
| The genome is continually under threat from exogenous sources of DNA damage, as well as from sources that originate within the cell. DNA double-strand breaks (DSBs) are arguably the most problematic type of damage as they can cause dangerous chromosome rearrangements, which can lead to cancer, as well as mutation at the break site and/or cell death. A complex network of molecular pathways, collectively referred to as the DNA damage response (DDR), have evolved to protect the cell from these threats. We have discovered a new DDR factor, LIN-61, that promotes the repair of DSBs. This is a novel and unexpected role for LIN-61, which was previously known to act as a regulator of gene transcription during development.
| DNA is maintained in the cell as chromatin: double-stranded DNA wrapped around core histone octomers to form nucleosome subunits. Chromatin folds into higher order structures depending on how tightly DNA is wrapped around the histones and how closely the nucleosomes interact [1]. Condensed chromatin acts as a physical barrier that restricts DNA access and therefore must be remodelled to enable various cellular processes such as gene transcription, DNA replication and DNA repair [2]. This is principally achieved by post-translational modification to the N-terminal tails of histones. One example of this is the methylation of lysine residues, which alters the degree of chromatin compaction and provides a binding site for the recruitment of non-histone proteins such as malignant brain tumour (MBT) domain proteins [2]. Once bound to histones, MBT domain proteins condense chromatin and repress transcription of target genes [3]. The MBT domain is a highly conserved motif of approximately 100 amino acids in length found throughout metazoans from C. elegans to humans [4].
Some MBT domain proteins act together with Polycomb group (PcG) repressor complexes that are best known for establishing and maintaining gene expression patterns during development [4]. The C. elegans MBT protein LIN-61 is also implicated in transcriptional regulation. It is a member of the synthetic multivulva (synMuv) class B group of proteins that act redundantly with synMuvA proteins to repress transcription of lin-3 EGF and lin-60 Ras [5]–[7]. Separate to its role within the synMuvB pathway, we found lin-61 is also involved in maintaining genome stability. Worms depleted of lin-61 have elevated rates of germline and somatic mutation, including small DNA insertions and deletions, but how LIN-61 maintains the genome fidelity was unknown [8]. Intriguingly, other MBT proteins have been shown to act as tumour suppressors: lethal(3)malignant brain tumour [l(3)mbt)] mutants of Drosophila develop malignant transformations of the adult optic neuroblast and ganglion mother cells of the larval brain [9]; furthermore, the human MBT domain genes L3MBTL2, L3MBTL3 and SCML2 are mutated in rare cases of medulloblastoma [10]. Also, depletion of L3MBTL1 (another LIN-61-related protein) causes genome instability [11]. Therefore it appears MBT proteins may have a general role in genome stability. It is not known how these proteins prevent tumourigenesis or protect the genome, but their ability to repress transcription likely plays a central role considering that the l(3)mbt malignancies of Drosophila ectopically express germline genes, the expression of which is required for tumour growth [12]. Preventing the expression of germline genes in somatic tissues may be a conserved function of MBT proteins because lin-61 mutants also express germline genes in the soma in a temperature-dependent manner [13].
As well as regulating transcription, an increasing number of chromatin-remodelling proteins (including PcG proteins) have been found to act within the DNA damage response (DDR). These proteins accumulate at sites of DNA damage where they locally modify chromatin to allow the recruitment of DNA repair proteins [14]. In the present study we investigate the cause of genomic instability in lin-61 mutants. We show that LIN-61 acts within the DDR where it is needed for efficient double-strand break (DSB) repair in both the germline and somatic cells of C. elegans. LIN-61 promotes DSB repair by homologous recombination (HR), but not the competing pathways, non-homologous end joining (NHEJ) or single-strand annealing (SSA). Despite the requirement for LIN-61 in HR, it is dispensable for meiotic recombination and the DNA damage checkpoints (cell cycle arrest and apoptosis) in the germline. We also use a novel GFP-based HR reporter assay that confirms LIN-61 is needed for HR. This reporter monitors the repair of a single defined DSB and is a new tool for measuring HR in C. elegans somatic cells. This is the first report demonstrating that an MBT protein promotes DNA repair and provides an explanation for why MBT-deficient cells have genomic instability.
To investigate how LIN-61 contributes to genomic stability, we obtained three independently generated null alleles of lin-61 (n3809, pk2225 and tm2649; Figure 1A and Text S1). The fourth MBT domain [essential for binding H3K9me2/3; [15]] is truncated or deleted in each of the mutant LIN-61 proteins. Moreover, lin-61 mRNA is reduced approximately four-fold in n3809 and pk2225, likely due to nonsense-mediated decay (Figure 1B). Each of the three mutants produced small broods (17–24% fewer progeny than wild types; Figure 1C). This can be symptomatic of genomic instability as DNA repair mutants such as brc-1, rfs-1, blm-1 and smc-5/-6 also have small broods [16]–[19]. In accordance with their reduced fecundity, lin-61 mutants had considerably smaller germlines than wild types and contained fewer nuclei in the mitotic compartment (Figure 1D–1E). What is more, there were signs of DNA damage in these cells: their mitotic nuclei contained considerably more spontaneous RAD-51 foci than those of wild types (Figure 1F). RAD-51 is the DNA strand exchange protein, which accumulates at DSBs and blocked replication forks, and therefore is a marker for DNA damage [20]–[22].
Since lin-61 mutant germ cells displayed genomic instability and signs of persistent spontaneous DSBs, we wondered whether lin-61 mutants were sensitive to ectopically induced DSBs. We found that the germ cells of lin-61 mutants were hypersensitive to ionizing radiation (IR), which is a potent inducer of DSBs (Figure 2A). Also primordial germ cells that are arrested in the G2 stage of the cell cycle in L1 stage larvae, are hypersensitive to IR in lin-61 mutants animals (Figure S1).
The LIN-61 paralog, called MBTR-1 (Malignant Brain Tumour Repeat containing protein 1), shares a high degree of sequence conservation with LIN-61 and both proteins are comprised almost entirely of four MBT domains (Figure S2A). We wondered whether MBTR-1 too might be needed for resistance to IR-induced DSBs. To test this, we challenged mbtr-1(n4775) mutants with IR but found that they were not more sensitive than wild type controls (Figure S2B). Therefore LIN-61, but not the closely related MBT domain protein MBTR-1, is required for resistance to IR-induced DSBs in germ cells.
The IR-hypersensitivity of lin-61 mutant germlines suggested that LIN-61 might be required for DSB repair during gametogenesis. We therefore investigated if LIN-61 also had a role in the repair of programmed DSBs that arise during meiosis. Meiotic DSB repair is required for the proper segregation of chromosomes to gametes and involves the repair of programmed DSBs introduced by the topoisomerase-like protein SPO-11 [23]. These DSBs are repaired by HR using the homologous chromosome as the repair template (interhomolog HR). The progression of DSB repair can be monitored in meiosis by following RAD-51 foci, which first appear at prophase, peak at early/mid-pachytene, and are resolved by late pachytene once DSB repair is completed [24]. The distribution of RAD-51 foci in lin-61 meiotic cells was indistinguishable from those of wild types (Figure S3). This indicated that repair of SPO-11-introduced DSBs was unperturbed in lin-61 mutants. Interhomolog HR enables crossover (CO) formation, which establishes the physical connection (chiasmata) that holds homologs together until their separation at the first meiotic cell division. Diakinesis stage oocytes of lin-61 mutants contained the correct complement of six bivalents (paired homologs), which indicated that CO formation was competent in these mutants. Furthermore, lin-61 mutants produced mostly viable progeny and did not display an increased incidence of males (Him) phenotype (Figure 1C). Failed meiotic recombination causes nondisjunction and aneuploidy due to the uncontrolled segregation of chromosomes to gametes, which manifests as embryonic lethality and the Him phenotype [25]. We conclude that LIN-61 is necessary for the repair of IR-induced DSBs but dispensable for CO formation and meiotic recombination. This phenotype is paralleled by the HR mutant brc-1 and the cohesin-like mutants smc-5/-6. These mutants are IR hypersensitive due to defective DSB repair by HR that uses the sister chromatid (intersister HR) [19], [26], [27]. Our observation that lin-61 mutants were hypersensitivity to IR suggested that LIN-61 might also contribute to intersister HR.
In addition to repairing IR-induced DSBs, intersister HR is needed for repair of interstrand crosslinks (ICLs). ICLs are particularly cytotoxic lesions that block the replication fork by covalently linking opposing strands of double-stranded DNA [28]. During ICL repair, the crosslinked lesion is excised, thus producing a DSB substrate for intersister HR [29]. HR-deficient mutants like brc-1, or the rad-51 paralog rfs-1 are therefore hypersensitive to ICLs [21]. Consistent with LIN-61 having a possible role in intersister HR, we found that lin-61 mutants were hypersensitive to nitrogen mustard (HN2), which is a potent inducer of ICLs (Figure 2B).
Other DNA lesions that block replication forks (such as bulky photoadducts made by UV light) do not cause a DSB and do not require HR for repair. Instead, translesion synthesis (TLS) DNA polymerases such as POLH-1 bypass these lesions to allow replication to proceed [30]. polh-1 mutants are therefore hypersensitive to UV-C [31] but HR-deficient mutants such as rfs-1 are not [21]. We found that lin-61 mutants were not hypersensitive to UV-C (Figure 2C). The sensitivity of lin-61 mutants to IR and HN2, but not UV-C, suggested that LIN-61 may promote DNA repair through HR, but is not required for the repair of other replication-blocking lesions such as photoadducts.
LIN-61 is broadly expressed in somatic and germ cells throughout development [6]. To determine if LIN-61 contributes to DSB repair in somatic cells, as it does in germ cells, we used established assays that test the proficiency of HR, as well as the other major DSB repair route, NHEJ [32]. Somatic cells use either HR or NHEJ depending on developmental context and phase of the cell cycle. HR is active during S and G2 phases (when sister chromatids are closely aligned), whereas NHEJ can be performed throughout the duration of the cell cycle, but is especially important during G1 when HR is unavailable [33]. Early stage embryonic cells (<6 hours post fertilisation) rapidly transition between S phase and M phase, without G1 and G2 gap phases [34], [35] and are particularly reliant on HR for DSB repair [32] (Figure 3A). Accordingly, early stage embryos of HR-deficient mutants are very sensitive to IR, while those of NHEJ-deficient mutants are not [32]. To test whether lin-61 promotes HR in somatic cells, we scored the viability of γ-irradiated early stage lin-61 embryos. These embryos were indeed hypersensitive to IR, which was indicative of an HR defect (Figure 3B). Their degree of IR sensitivity was similar to that of HR-deficient brc-1 embryos. While HR is the dominant DSB repair route in early embryos, NHEJ is the major repair pathway in late stage embryos and arrested L1 larvae because most of their cells are arrested in G1 [32] (Figure 3C). NHEJ-deficient L1 larvae have delayed or arrested growth in response to IR [32]. We found that wild type, lin-61(n3809) and lin-61(pk2225) L1 larvae did not display substantial growth delay following IR, whereas most NHEJ-deficient cku-80 mutants failed to develop to the L4 stage 48 hours after irradiation (Figure 3D). L1 larvae of the HR-deficient mutant, brc-1, were also not hypersensitive to IR (Figure S4). Taken together, these results suggest that LIN-61 has a role in repairing DSBs by HR, but not NHEJ, in somatic cells.
Although lin-61 mutants phenocopy brc-1 mutants in many aspects of genome stability, they also differ in some important aspects. For example, brc-1 mutants display the Him phenotype, while lin-61 mutants do not. Him is an indication of problems with chromosome segregation at meiosis. Like brc-1 mutants, lin-61 mutants are able to successfully complete meiosis, indicating that their interhomolog HR is proficient. However, by genetically disrupting the synaptonemal complex (SC), and thereby preventing interhomolog HR, it has been possible to demonstrate that BRC-1 contributes to meiotic intersister HR [27]. Adamo and colleagues observed that chromosomal fragments appear in the diakinesis stage nuclei of brc-1 mutants that were depleted of key SC components [27]. Using this approach we tested whether LIN-61 also has a role in meiotic intersister HR. In contrast to brc-1 mutants, neither the oocytes of lin-61(pk2225) or lin-61(n3809) contained chromosomal fragmentation after depletion of the core SC component, SYP-2 (Figure 4A). These data, together with those showing normal RAD-51 kinetics and successful chiasmata formation in lin-61 mutants (Figure S3 and Figure 4A), indicate that LIN-61 is dispensable for HR in meiotic cells.
lin-61 mutants are proficient in the repair, at meiosis, of SPO-11-introduced DSBs (using both intersister and interhomolog repair) but are hypersensitive to IR. To confirm that LIN-61 is required for DSB repair specifically in mitotic germ cells we used an assay that directly tests whether DSBs are adequately repaired in irradiated germ cells. Completion of DSB repair can be determined in germ cells by observing chromosomes at diakinesis because chromosome fragments are present if DSBs are unrepaired [36]. In the absence of exogenous damage, the diakinesis stage oocytes of lin-61 mutants contained six bivalents and were not fragmented (Figure 4B). This demonstrated that DSBs induced by SPO-11 were efficiently repaired in lin-61 mutants, as discussed earlier. Strikingly however, both lin-61 mutants and the HR-deficient mutant brc-1 had severely fragmented chromosomes 48 hours after γ-irradiation (Figure 4B–4C). We anticipated that these nuclei could have been located within the mitotic zone at the time of irradiation, having subsequently migrated to the diakinesis stage 48 hours later. Failure to repair the introduced DSBs could therefore be due to defective HR whilst in the mitotic zone, or later whilst in the meiotic zone, or both. To distinguish between these possibilities we analysed earlier time points following irradiation (7 h and 24 h). For these time points, the nuclei being analysed were in meiosis when DSBs were introduced. We found that brc-1 mutants had fragmented chromosomes at these earlier time points (7 h and 24 h) (Figure 4B–4C), which is consistent with BRC-1 acting in meiotic DSB repair [27]. In contrast, lin-61 mutants, like wild types, rarely had fragmented chromosomes at early time points following irradiation (Figure 4B–4C). Thus while BRC-1 contributes to DSB repair in both mitotic and meiotic cells, LIN-61 seems to promote DSB repair only in mitotic cells. In accordance with that notion, we found that brc-1 mutants were more sensitive to IR than lin-61 mutants (Figure 4D). Moreover, lin-61 brc-1 double mutants were no more sensitive to IR than brc-1 single mutants suggesting that lin-61 acts within the brc-1 genetic pathway (Figure 4D).
Having established that LIN-61 promotes DSB repair via HR, we looked to address which step of HR fails in lin-61 mutants. The first stages of HR involve the nucleolytic processing at the DSB to expose single stranded 3′ overhangs (DNA end resection) and subsequent coating of these overhangs with RAD-51. RAD-51 foci rapidly formed in the γ-irradiated mitotic germ cells of both wild types and lin-61 mutants (Figure 5A). Foci were detected at a very early time point after γ-irradiation (10 minutes), which showed that DNA end resection was unperturbed in these cells (Figure 5A). The loading of RAD-51 at SPO-11-induced DSBs was also normal in lin-61 meiotic cells, as discussed earlier (Figure S3). Together this showed that DNA end resection at IR-induced and SPO-11-induced DSBs, as well as the loading of RAD-51 on resected DNA, was normal in lin-61 mutants. The number of RAD-51 foci that formed in γ-irradiated germ cells was similar between wild types and lin-61 mutants (4–5 foci per nucleus) (Figure 5B). Since the DNA in wild type and lin-61 nuclei were equally susceptible to IR, the hypersensitivity of these mutants was not due to an elevated damage load.
While IR is a potent source of DSBs, it also causes oxidative damage to proteins and cell membranes [37]. To confirm that the hypersensitivity displayed by lin-61 mutants was due to defective DSB repair (and not other types of damage), we developed an assay that specifically measures HR-mediated repair of a defined DSB. This assay was based on the DR-GFP reporter system, which has been used extensively to measure HR proficiency in cultured human cells [38]. Such an assay was previously unavailable to the C. elegans researcher. The new C. elegans reporter consisted of a gfp gene in which part of the open reading frame had been deleted and replaced by an I-SceI endonuclease recognition site, which rendered the GFP non-functional, and provided the defined location where the DSB could be introduced (Figure 6A). A fragment of gfp containing the sequences disrupted by the I-SceI site (but by itself non-functional) was located downstream of the reporter and served as a template for synthesis-dependent strand annealing (SDSA) (Figure 6A). SDSA is a sub-pathway of HR that results in gene conversion rather than a CO and is the most common HR pathway used to repair two-sided DSBs [39]. The reporter was designed such that repair of the DSB by SDSA (but not a CO pathway) would be able to restore expression to the corrupted gfp gene. Non-HR pathways such as NHEJ or SSA are unable to produce functional GFP (Figure 6B).
We created a transgenic strain that carried both the HR reporter and heat-shock inducible I-SceI endonuclease. I-SceI was fused to mCherry so that its expression could be easily monitored by epifluorescence. Since it is thought HR does not occur in postmitotic cells (i.e. G1/G0 stage cells), we chose to express the reporter in intestinal cells using the elt-2 promoter as their nuclei undergo endoreplication (S phase without mitosis) at several points during post-embryonic development [40]. We first confirmed that induction of mCherry::I-SceI resulted in GFP expression. 60–80% of wild type worms expressed GFP in intestinal nuclei 24 hours after mCherry::I-SceI expression. Importantly, reporter activation was dependent upon DSB induction because non-heat shocked worms did not express GFP (data not shown). Also, GFP expression was dependent upon the donor gfp sequences since a disabled version of the HR reporter, which lacked these sequences, was not able to express GFP (Figure S5). To confirm that GFP expression depended on HR, we tested the effect brc-1 mutation had on the reporter. BRC-1 promotes intersister HR in meiotic cells [27], and likely in somatic cells as well [41]. Indeed, brc-1 mutants had significantly reduced frequency of HR reporter activation (Figure 6C–6D). This confirmed that the assay provided a measure of HR proficiency. We also used an rtel-1 mutation to test whether reporter activation was dependent on the SDSA pathway. RTEL-1 is thought to influence HR pathway choice by removing the invaded DNA strand from its homologous template, which has the effect of promoting SDSA at the expense of CO outcomes [42]. The role of rtel-1 in somatic cells was previously untested but we found that rtel-1 mutants also had significantly reduced rates of HR reporter activation (Figure 6D). Therefore RTEL-1 likely promotes SDSA in somatic cells as it does in meiotic cells. A previous study showed that DSB repair pathways are dynamic and are in competition in C. elegans somatic cells such that the inhibition of one pathway caused increased activity in the others [41]. We therefore reasoned that inhibiting NHEJ should increase the frequency of HR reporter activation. As predicted, blocking NHEJ by cku-80 mutation resulted in substantial elevation of HR activity. More cku-80 animals expressed GFP than wild types (Figure 6D). This increase was likely an underestimation of HR activity as the GFP was also expressed much more brightly in cku-80 mutants than wild types. Brighter GFP likely results from multiple HR reporter genes being activated within a single cell. These experiments demonstrated that the HR reporter is able to measure relative changes in HR activity, in both HR-deficient and HR-hyperactive mutants.
Importantly, we found that both lin-61(n3809) and lin-61(pk2225) mutants showed a substantial reduction in the frequency of HR reporter activation compared with wild types (Figure 6D). In fact HR activation in lin-61 mutants was reduced to brc-1 levels. This confirmed LIN-61 is needed for DSB repair by the HR pathway. Further, it indicated that IR hypersensitivity of lin-61 mutants was likely due to defective DSB repair rather than other types of IR-induced cellular damage. While HR repairs DSBs in an error-free way, other DSB repair pathways such as NHEJ and SSA are error-prone processes. To test whether LIN-61 contributes to mutagenic DSB repair routes, we constructed a second reporter gene that specifically monitored SSA. This SSA reporter was similar to the HR reporter as both were expressed in intestinal nuclei and both received a single DSB from the mCherry::I-SceI enzyme, however the SSA reporter could only become active following an SSA event, and not an HR event (Figure S6A). We found that lin-61 mutants did not have reduced SSA activity but actually had increased SSA reporter activation compared to wild types (Figure S6B–S6C), in line with lin-61 mutants being HR-defective. A similar shift towards SSA has previously been found for DSB repair in brc-1 mutant animals [41]. We conclude that LIN-61 is necessary for efficient HR in somatic cells but is dispensable for SSA in somatic intestinal cells. Assays that measure sensitivity to DNA-damaging agents revealed that embryonic and germline cells of lin-61 mutants are defective for DSB repair (Figure 2 and Figure 3). The data generated using the HR and SSA reporters demonstrated that cell types other than those of the germline and embryo are defective for DSB repair in lin-61 mutants. Together, these complementary experiments suggested that lin-61 mutants have a systemic defect in DSB repair.
Sensitivity to DNA damage can be caused by failure to activate DNA damage checkpoints [43]. The G2/M checkpoint is triggered in response to DNA damage and keeps mitotic germ cells in G2 phase to provide sufficient time for DNA repair (Figure 7A) [44]. Arrested cells do not divide, but continue to grow, making them readily identifiable by their enlarged size [43]. Following exposure to IR, all three lin-61 mutants displayed proficient cell cycle arrest. Like wild type worms (and mbtr-1 mutants that are not IR sensitive), the lin-61 mutants had enlarged mitotic nuclei and a reduced number of germ cells 24 hours after γ-irradiation (Figure 7B–7C).
In addition to the G2/M checkpoint, DNA damage also triggers apoptosis in pachytene stage meiotic cells via a process dependent upon the p53 homologue, CEP-1 [43], [45]. Upon challenge with IR, apoptotic corpses accumulated in the germlines of wild type, lin-61(n3809) and lin-61(pk2225) animals, while cep-1 mutants failed to undergo DNA damage-dependent apoptosis (Figure 6D–6E). CEP-1 drives the apoptotic programme by up-regulating egl-1/BH3-only transcription [43], [45], [46]. In response to IR, egl-1 expression was increased in wild type and lin-61 worms, but not cep-1 mutants, as determined by qRT-PCR (Figure 7F). Together these results indicated that the activation of DNA damage checkpoints (cell cycle arrest and apoptosis) was normal in lin-61 mutants. The hypersensitivity of lin-61 mutants to IR could therefore not be attributed to defective checkpoint activation.
Since LIN-61 is a transcriptional repressor, we checked whether DDR genes were appropriately expressed in lin-61 mutants, as this could be the underlying cause of their HR defect. Using microarrays, we compared the expression profiles of wild types and lin-61 animals. Young adult worms (24 hours post L4) were analysed in order to increase the proportion of germ cells present in the samples, considering LIN-61 is needed for repair of DSBs in both somatic and germ cells. Microarrays were performed on two different lin-61 alleles (n3809 and pk2225) in order to control for changes in gene expression that were due by background mutations present within only one of the single strains. 58 genes were identified that, in both mutants, had a 1.5-fold or greater change in expression level (p-value<0.01) (Table S1). Most of these alternatively expressed genes were upregulated in lin-61 mutants (52 genes, 90%), with only 6 genes (10%) downregulated. This is consistent with LIN-61 acting as a transcriptional repressor. Importantly, none of the genes alternatively expressed in lin-61 mutants were implicated in DNA repair. The lin-61 transcript served as a positive control in the microarray analysis as we had previously shown, using qRT-PCR, that this transcript was reduced approximately 4-fold in lin-61 mutants, likely due to nonsense-mediated decay (Figure 1B). According to the microarray data, lin-61 mRNA was reduced 3.25-fold, which in good agreement with the qRT-PCR data. The expression analysis showed that while LIN-61 does indeed act as a transcriptional repressor, lin-61 mutation by itself (in the absence of an additional synMuvA mutation) has only a minor effect on global gene transcription. Finally, since these experiments indicated that DNA repair genes are expressed at normal levels in lin-61 mutants, it is likely that LIN-61 influences DSB repair directly and not by ensuring that other DDR genes are appropriately expressed.
In this study we have identified the underlying cause of genomic instability in lin-61 mutants: DSBs are not adequately repaired due to defective HR. Accordingly, these animals are hypersensitive to IR and nitrogen mustard and DSBs remain unrepaired in diakinesis oocytes of γ-irradiated lin-61 mutants. LIN-61 contributes to HR in mitotic cells but it is dispensable for DSB repair during meiosis. Sensitivity of lin-61 germ cells to DSBs is not due to faulty DNA damage checkpoints as both cell cycle arrest and apoptosis are functional. Moreover, DNA repair genes are not inappropriately expressed in lin-61 mutants. The role of LIN-61 in HR is not restricted to germ cells because the somatic cells of early stage embryo are also very sensitive to IR. Also, later in development, intestinal cells are HR defective, as determined by the GFP-based HR reporter system. HR is essential for genome stability, as it is the principal DSB repair route in germ cells. It is also an error-free repair pathway. Blocking HR enables mutagenic and toxic repair routes to become active, which likely contributes to genomic instability in lin-61 mutants.
LIN-61 is expressed in all nuclei, both in the germline and somatic tissues [6]. Despite this, several observations suggests that LIN-61 contributes to HR only in mitotic cells and is dispensable for both meiotic interhomolog and intersister HR. Meiotic cells rely on interhomolog HR to repair at least one programmed DSBs per chromosome pair so that the obligate CO will be established [47]. Meiotic recombination is not defective in lin-61 mutants as they form chiasmata normally and produce nearly completely viable broods. What is more, RAD-51 foci that appear in prophase are resolved by late pachytene in both wild type and lin-61 mutants, indicative of the successful repair of programmed DSBs. The proficiency of intersister HR can be tested in meiotic cells by disrupting the SC in order to prevent interhomolog HR. In this situation, DSBs remain unrepaired if intersister HR too is defective, which manifests as chromosomal fragmentation at diakinesis. Unlike brc-1 and smc-5/-6 mutants [19], [27,27], lin-61 mutants depleted of the SC component SYP-2 do not have fragmented diakinesis chromosomes, indicating that intersister HR is proficient in the meiotic cells of these mutants. Moreover, DSBs introduced by IR into lin-61 meiotic cells, but not brc-1 meiotic cells, are efficiently repaired.
While lin-61 mutants are proficient in meiotic HR, their mitotic cells are defective in HR. These cells display signs of persistent and spontaneous DNA damage. Further, γ-irradiation of mitotic germ cells causes severe chromosome fragmentation in lin-61 mutants. Finally, lin-61 mutants are also hypersensitive to ICLs and the repair of these lesions occurs in S/G2 phase using the newly synthesised sister chromatid as the HR repair template [29]. The somatic (mitotic) cells of lin-61 are also hypersensitive to IR and mitotic cells exclusively use the sister chromatid for HR [39]. Together, these observations indicate that LIN-61 contributes to DSB repair via intersister HR in mitotic cells but does not participate in meiotic HR.
Considering that the transcriptional profile of lin-61 mutants cannot explain their HR defect, LIN-61 likely acts directly at sites of DNA damage to promote DSB repair. This is an attractive hypothesis considering that chromatin can act as a physical barrier that must be remodelled to allow access of DDR factors to sites of damage. In addition, many proteins that alter chromatin structure have recently been implicated in the DDR including NuRD components MTA1, MTA2, CHD4, HDAC1 and HDAC2 [48]–[50]; and PcG proteins BMI1, RING1, RING2 and HP1 [51]–[55]. Each of these proteins is rapidly recruited to DNA damage and is necessary for DNA repair. The C. elegans counterparts of these proteins are also synMuvB proteins like LIN-61. Intriguingly, L3MBTL2, the putative human orthologue of LIN-61, is part of a PcG-like complex (PRC1L4) that shares RING1, RING2 and HP1γ as partner members [56]. Moreover, human cells depleted of RING2 [55], and C. elegans hlp-2 HP1 mutants [53], are radiosensitive like lin-61 mutants. PRC1L4, or a related L3MBTL2-containing PcG complex, may therefore act in DSB repair like LIN-61. Using immunofluorescence, we were not able to detect a change in LIN-61 intracellular localisation upon IR (data not shown). However LIN-61 is abundantly present and localised at chromatin in all cells, which may conceal its relocalisation around sites of DNA damage. Recruitment to sites of DNA damage has also not been observed for any other C. elegans synMuvB proteins, likely for similar reasons.
It is unknown how PcG activity promotes DSB repair but it is argued that inhibiting transcription locally at the DSB may be important as the transcriptional machinery could interfere with repair proteins or with DNA repair intermediates [50], [57]. PRC1L4 represses transcription of target genes by monoubiquitinating lysine 119 of histone H2A via its E3 ubiquitin ligase activity [56]. This histone mark is also implicated in the DDR as it was recently shown to rapidly accumulate at DSBs [52], [58]. It will be of interest to determine whether L3MBTL2 and the other members of PRC1L4 are involved in DSB repair in human cells.
One possible explanation we considered for why lin-61 mutants were HR-defective was that they might have altered expression of DDR genes. But contrary to this, microarray expression analysis did not reveal any alternatively expressed DDR genes in these mutants. Some alternatively expressed genes were identified but none are implicated in DNA repair. The vast majority of the alternatively expressed genes were upregulated rather than downregulated, which is in accordance with LIN-61 being a transcriptional repressor. A previous study found that germline genes were ectopically expressed in the somatic tissues of lin-61 mutants, but only when maintained at the relatively high temperature of 26°C [13]. In line with this, we found that lin-61 mutants grown at the normal laboratory temperature of 20°C had only minor changes in gene expression and did not overexpress germline genes. Importantly, lin-61 mutants grown at 20°C displayed a profound HR defect, which further indicated that altered gene expression was not the cause of defective DNA repair. The microarrays were performed using RNA from a mixed population of germ and somatic cells. We cannot strictly exclude the possibility that a distinct population of cells had altered DDR gene expression that went undetected. This is unlikely though, as the defect in DSB repair was systemic, occurring in multiple tissues and at various stages of development, and not isolated to a small number of cells.
In this study we introduce a novel reporter system for monitoring HR in C. elegans somatic cells. The reporter confirmed that LIN-61 is needed for HR. This tool was previously unavailable for C. elegans researchers. We propose it as a method for testing candidate HR genes, for example it confirmed that both BRC-1 and RTEL-1 have roles in HR in somatic cells, analogous to their functions previously only described in meiotic germ cells. Our experiments with the HR reporter also supported previous findings that suggested DSB repair pathways are dynamic and are in competition in somatic cells [41] since mutations that blocked NHEJ, increased HR reporter activity.
Though this system is a new tool that provides for the readout of repair, probably by an SDSA mechanism, of a defined DSB, it does have limitations. For example, the HR reporter does not easily allow for dissection of the biochemical processes that underpin HR pathways. These approaches are not well suited to C. elegans. Also, in its current form the HR reporter is expressed only in intestinal cells, which in contrast to most C. elegans somatic cells still cycle postembryoniccally. This choice of cell type was largely motivated by the likely need for S- and G2 phase dependent DNA end resection at DSBs for HR type of repair to occur. However, when interpreting the data it must be considered that these cells are atypical because they progress and grow through cycles of endoreduplication and not via canonical cell cycle stages including mitosis. It is thus possible that the response to the HR reporter is cell type-dependent. Finally, since formation of the DSB relies on expression of the I-SceI transgene using the heatshock promoter, any possible differences in heatshock response must be carefully controlled for as these differences may affect the level of DSB induction.
This is the first report showing that an MBT protein is needed for DSB repair. Genes encoding MBT proteins have previously been linked with tumourigenesis and can act as tumour suppressor genes. However, their contribution to DNA repair and genome stability is unknown. Our finding that LIN-61 is required for efficient HR may have implications for the treatment of MBT-deficient tumours, which may also be HR defective. HR-deficient tumours, such as those with BRCA1 or 2 hypomorphic mutations, are very susceptible to poly(ADP ribose) polymerase (PARP) inhibitors [39]. It will be important to determine whether the role of LIN-61 in DSB repair is conserved in human MBT proteins and whether MBT mutated tumours, such as medulloblastomas with mutations in L3MBTL2, L3MBTL3 or SCML2 [10], are HR deficient as they too may prove responsive to treatment with PARP inhibitors.
The Bristol N2 strain was used as the wild type strain and maintained at 20°C according to standard protocols [59]. Alleles used in the study include LG I: lin-61(n3809) [6], lin-61(pk2225) (this study), lin-61(tm2649) [15], mbtr-1(n4775) [6], cep-1(gk138) [60] and rtel-1(tm1866) [61]; LG III: brc-1(tm1145) [62], cku-80(ok861) [63], polh-1(lf31) [31], lfIs129 [elt-2::HR-reporter; hsp16-41::mCherry::I-SceI] (this study); and LG X: lfIs82 [elt-2::SSA-reporter; hsp16-41::mCherry::I-SceI] (this study). To determine brood sizes, L4 larvae were singled on 6 cm plates with OP50 E. coli and transferred each day for three days. The number of viable progeny and unhatched eggs were counted, as well as the number of males in the brood.
All γ-irradiation was performed with a dose rate of 15 Gy/minute using an electronic X-ray generator set to 200 kV 12 mA (XYLON International). For L4 larval IR sensitivity, three L4 animals per plate (three plates per condition) were treated with various doses of γ-irradiation. For UV-C sensitivity, young adult (24 post L4 stage) worms were exposed to UV (254 nm lamp, Philips). HN2 sensitivity assays were performed as described [64]. γ-irradiation of embryos and L1 larvae was preformed as described [32]. Apoptosis assays were performed in as [45]. Cell cycle arrest and fragmentation assays were as in [36]. syp-2 RNAi was performed as in [19]. For cell cycle arrest, 4–5 germlines were analysed per condition, except for irradiated lin-61(tm649) for which a single germ line was scored.
Germlines were dissected in egg salts, Tween, levamisole and fixed in 2% paraformaldehyde for 5 minutes at room temperature, and snap frozen on dry ice, then placed in methanol at −20°C for 10 minutes, washed three times for 10 minutes in PBS with 1% Triton X-100 and blocked in PBST (PBS with 0.1% Tween 20) and 1% BSA for 30 minutes at room temperature. Samples were incubated overnight at 4°C with rabbit anti-RAD-51 antibodies (Novus Biologicals) diluted 1∶200 in PBST 1% BSA and detected with Alexa488 goat anti-rabbit antibodies (Invitrogen) diluted 1∶1000. DNA was counterstained with 0.5 µg/ml DAPI and samples were mounted with VectaShield. RAD-51 foci were imaged with a Leica DM6000 deconvolution microscope collecting 0.5 µm Z-sections. The number of foci per nucleus was counted for each of the seven zones of the germline as described [64]. Three to five germlines were quantified per condition.
Worms were synchronised as L1 larvae by bleaching and grown to the L4 stage. Total RNA was isolated with Trizol reagent (Invitrogen), and cleaned with RNeasy kit (Qiagen). Service XS (Leiden, NL) performed the Affymetrix expression analysis according to standard protocols. Data was analysed with the MAS 5.0 algorithm using Tukey's biweight estimator. Significance (p-value) was determined using Wilcoxon's rank test. Sequence of qRT-PCR primers is available in Text S1.
Details on construction of the Pelt-2::HR and Pelt-2::SSA reporter strains are provided in Text S1. For HR reporter assays, expression of mCherry::ISce-I was induced in L4 larvae by heatshock twice at 34°C for 1 hour (with 30 min rest at 20°C). 24 hours after induction, worms were mounted on agarose pads and their intestinal nuclei were scored for GFP expression using a Leica DM6000 microscope with 63× objective. Experiments were performed in triplicate with 50–100 animals tested for each condition. Statistical significance was tested using the Cochran-Mantel-Haenszel test.
|
10.1371/journal.pgen.1001033 | Inter- and Intra-Individual Variation in Allele-Specific DNA Methylation and Gene Expression in Children Conceived using Assisted Reproductive Technology | Epidemiological studies have reported a higher incidence of rare disorders involving imprinted genes among children conceived using assisted reproductive technology (ART), suggesting that ART procedures may be disruptive to imprinted gene methylation patterns. We examined intra- and inter-individual variation in DNA methylation at the differentially methylated regions (DMRs) of the IGF2/H19 and IGF2R loci in a population of children conceived in vitro or in vivo. We found substantial variation in allele-specific methylation at both loci in both groups. Aberrant methylation of the maternal IGF2/H19 DMR was more common in the in vitro group, and the overall variance was also significantly greater in the in vitro group. We estimated the number of trophoblast stem cells in each group based on approximation of the variance of the binomial distribution of IGF2/H19 methylation ratios, as well as the distribution of X chromosome inactivation scores in placenta. Both of these independent measures indicated that placentas of the in vitro group were derived from fewer stem cells than the in vivo conceived group. Both IGF2 and H19 mRNAs were significantly lower in placenta from the in vitro group. Although average birth weight was lower in the in vitro group, we found no correlation between birth weight and IGF2 or IGF2R transcript levels or the ratio of IGF2/IGF2R transcript levels. Our results show that in vitro conception is associated with aberrant methylation patterns at the IGF2/H19 locus. However, very little of the inter- or intra-individual variation in H19 or IGF2 mRNA levels can be explained by differences in maternal DMR DNA methylation, in contrast to the expectations of current transcriptional imprinting models. Extraembryonic tissues of embryos cultured in vitro appear to be derived from fewer trophoblast stem cells. It is possible that this developmental difference has an effect on placental and fetal growth.
| We have screened a population of children conceived in vitro for epigenetic alterations at two loci that carry parent-of-origin specific methylation marks. We made the observation that epigenetic variability was greater in extraembryonic tissues than embryonic tissues in both groups, as has also been demonstrated in the mouse. The greater level of intra-individual variation in extraembryonic tissues of the in vitro group appears to result from these embryos having fewer trophoblast stem cells. We also made the unexpected observation that variability in parental origin-dependent epigenetic marking was poorly correlated with gene expression. In fact, there is such a high level of inter-individual variation in IGF2 transcript level that the presumed half-fold reduction in IGF2 mRNA accounted for by proper transcriptional imprinting versus complete loss of imprinting would account for less than 5% of the total population variance. Given this level of variability in the expression of an imprinted gene, the presumed operation of “parental conflict” as the selective force acting to maintain imprinted gene expression at the IGF2/H19 locus in the human should be revisited.
| Several epidemiological studies have reported a higher incidence of rare disorders involving imprinted genes (Angelman syndrome [1]–[3] and Beckwith-Wiedemann syndrome [4]–[8]) among children conceived using assisted reproductive technologies (ART). Studies on imprinted gene expression and parental allele-specific DNA methylation in animal models have also suggested that epigenetic marks may be altered by treatments and procedures commonly employed in ART (ovarian stimulation, egg retrieval, in vitro fertilization, intracytoplasmic sperm injection, preimplantation embryo culture, embryo transfer) [9]–[13].
CpG sites in differentially methylated regions (DMRs) of imprinted genes are methylated on the allele contributed by one parent and unmethylated on the allele contributed by the other. This pattern of differential allelic methylation is established during male and female gametogenesis [14], [15] and the differences are maintained after fertilization such that cells from most somatic tissues are expected to exhibit the same parental allele-specific methylation pattern [16], [17]. However, if ART treatments and procedures in the human are disruptive to imprinted gene methylation patterns, as they are in the mouse [18], one might predict that alterations could occur in some cells of the early embryo but not in others, resulting in individuals who were mosaic to varying degrees for loss or relaxation of proper imprinted allelic methylation. In addition, different degrees of relaxation of allele-specific methylation could be observed between different tissues of the same individual, depending on when the disruption occurred during development.
One of the loci shown to be susceptible to alteration of epigenetic modifications by in vitro culture and ovarian stimulation in the mouse is Igf2/H19 [11], [13], [19]–[23]. Because IGF2 is an important placental growth factor and one of the phenotypes most strongly associated with human ART procedures is low birth weight, we reasoned that the human IGF2/H19 locus might also be susceptible during ART treatments and procedures. We compared parental allele-specific methylation between children conceived in vitro or in vivo at the DMR that functions as an imprint control region (ICR) at IGF2/H19 [24], [25] and also at an IGF2 receptor (IGF2R) DMR [26], [27]. We examined a sample of cord blood, a section of umbilical cord and five sections of placenta in each child for abnormal methylation of maternal alleles at the IGF2/H19 ICR/DMR and for abnormal methylation of paternal alleles at the IGF2R DMR [28]. Under our null hypothesis, little variation in parental allele-specific methylation was expected within an individual, or between individuals, because the methylation status of each CpG site in the DMR is set in the gametes [29] and faithful replication of this status during development is expected to result in the same allelic methylation ratio in each individual and in each tissue. We also measured steady-state IGF2, H19 and IGF2R transcript levels to determine whether mRNA levels were correlated with abnormal allelic methylation ratio or birth weight. To our knowledge, this is the first study to examine intra-individual variation in epigenetic markings in children conceived using assisted reproduction.
We investigated intra- and inter-individual variation in allele-specific methylation at the IGF2/H19 DMR. We measured the relative level of CpG methylation on maternal and paternal alleles at this locus in cord blood, cord and five sections of placenta taken from a population of children conceived either in vitro or in vivo.
The imprinted IGF2/H19 DMR is located between the IGF2 and H19 genes and is normally methylated on only the paternal allele [24], [30]. We used a single nucleotide polymorphism (a C/T SNP at a CfoI site) to identify informative (heterozygous) individuals and a methylation-sensitive restriction endonuclease (MluI) to determine the methylation status of a specific CpG site within the DMR, as described previously [28]. Methylation at the MluI site, and an adjacent MaeII site, have been shown previously to be characteristic of the methylation at surrounding CpGs by bisulfite sequencing (Fig. 7 in Sandovici et al., 2003) [28]. We identified 45 in vitro and 56 in vivo individuals who were informative and for whom DNA was available from cord blood, cord and five sections of placenta.
Because previous studies have indicated that loss or relaxation of imprinting is a quantitative trait [31], [32], we measured the ratio between DNA methylation levels on maternal and paternal (M/P) alleles as an indicator of imprinting status. A ratio of zero corresponds to exclusive methylation of the MluI site on the paternal allele, while a ratio of one signifies methylation of this site on an equal number of maternal and paternal alleles (n.b.: Although we have not determined the parental origin of each allele in the present study because DNA was not available from parents, we have shown previously that the less methylated allele was maternal in all 163 individuals for whom we were able to determine parental origin by pedigree analysis [28]. We therefore assume that the less methylated allele is maternal in the population examined here, also.) In the case of controls, no uncleaved C alleles were detected in any C/C homozygous individuals, indicating that CfoI cleaved the “hot-stop” PCR products with >99% efficiency [33].
Figure 1 shows the distribution of M/P methylation ratios observed from 56 informative in vivo (Figure 1A) and 45 informative in vitro (Figure 1B) individuals. The M/P ratios were measured for cord blood, cord and five sections of placenta from each informative individual. The data are represented as a series of symbols on a vertical line, ranked from individuals showing the greatest range of variation on the left side of the graph to individuals showing the least range of variation on the right side of the graph. The distribution of individual M/P methylation ratios in cord blood (red circles in Figure 1) shows that the great majority of informative individuals have <15% methylation on the maternal allele in both groups. Approximately 8% of the population examined here has ≥15% of methylation of CpG sites on the maternal allele in cord blood. These findings are similar to those reported previously by Sandovici et al. (2003) for M/P methylation ratios measured in peripheral blood samples from the CEPH families [28]. The distribution of individual M/P methylation ratios in cord showed a similar pattern to that observed in cord blood (Figure 1), while the five sections of placenta taken from each individual showed a broad range of intra-individual variation in M/P methylation ratios.
When individual tissues are compared, both M/P ratio mean and variance are greater in the in vitro group in each tissue, although two of these comparisons (cord blood means, in which the fewest samples are compared between the two groups, and placenta variance, in which the greatest range of variation is observed) do not reach statistical significance (Table 1). However, because the a priori expectation for the mean M/P methylation ratio of the IGF2/H19 DMR is near zero, independent of the embryonic origin of the tissue (because the methylation status of maternal and paternal DMR alleles is assumed to be determined in the gametes and to escape the genome wide demethylation/remethylation that occurs in preimplantation embryos) [16], it is not inappropriate to combine all samples in each group to determine whether the two groups differ in mean M/P ratio and whether the groups have equal variance. When samples from all tissues are combined, both allele-specific methylation ratio mean and variance are significantly greater in the in vitro group (P = 0.0001 and P = 0.0006, respectively, Table 1). The fact that the intra-individual variation in M/P ratios is also greater in the in vitro group may be seen, simply, by comparing the fraction of individuals in each group in which all samples have M/P ratios below any arbitrarily chosen value. For example, only 31% of in vitro individuals (14/45) maintain M/P ratios of <0.1 in cord blood, cord and all five sections of placenta while 46% of in vivo individuals (26/56) are below this threshold.
One mechanism by which greater variance in an epigenetic character may occur is through a sampling effect that depends on the number of stem cell progenitors that give rise to a particular tissue; the fewer the number of stem cells, the greater the variance. Because much of the difference in variance observed between the two groups occurs as a result of intra-individual differences in umbilical cord and placenta samples, we estimated the number of trophoblast stem cells that give rise to the placenta in each group by comparing the distribution of X-inactivation scores in females from each comparison group [34], [35] and by comparing the distribution of M/P IGF2/H19 methylation ratios from Figure 1. We note that the assay used in each case amounts to a simple yes/no binomial trial of the form “is the CpG site being examined methylated (in which case it gives a signal) or not (in which case it does not)”, each of which is expected to yield a “success” (the DNA molecule in question is methylated) with probability “p” (p = 0.5 in the case of which X chromosome is inactivated and p = 0.1 in the case of methylation of the maternal IGF2/H19 DMR, see below) or a failure with probability “q” (which is equal to 1-p).
The number of trophoblast stem cells may be estimated from the distribution of X-inactivation scores by comparing the actual distribution of X-inactivation scores with the distribution estimated from the variance of the binomial distribution (pq/N), setting the probability that either allele is methylated (p or q) to 0.5 and generating the distribution for different values of N (number of stem cells) [34], [35]. We determined the X-inactivation score distribution for each group (using DNA samples from five sections of placenta from 50 in vitro and 54 in vivo females) by comparing allele-specific methylation at HpaII sites adjacent to the CAG trinucleotide repeat in the highly polymorphic Androgen Receptor locus (AR) [36]. The closest fit to the distribution of X-inactivation scores in placenta in children conceived in vitro corresponds to nine trophoblast stem cells and the closest fit in children conceived in vivo corresponds to 11 trophoblast stem cells (Table 2).
We additionally estimated the number of trophoblast stem cells in each group by comparing the distribution of M/P IGF2/H19 DMR methylation ratios (Figure 1) from the two comparison groups, using 0.1 and 0.9 as values for p and q (these values were selected based on the observation that ∼10% of individuals have significant methylation on the maternal DMR while ∼90% have very few cells carrying maternal DMR methylation [28]. These values are also in close agreement with the probability that any maternal DMR DNA molecule is methylated (placenta in vivo mean = 0.0801, in vitro mean = 0.1017, Table 1). Using these parameters, the closest fit to the distribution of M/P ratios in placenta in children conceived in vitro corresponds to eight trophoblast stem cells and the closest fit in children conceived in vivo corresponds to 10 trophoblast stem cells (Table 2).
Overall, these very similar independent estimates of between-group epigenetic variation (n.b.: not only are the two loci examined on different chromosomes but many of the individuals in the X-inactivation groups, composed of only females, and the IGF2/H19 groups were different) are consistent with the prediction that overall greater variance in M/P IGF2/H19 DMR methylation ratios in the in vitro group is associated with fewer trophoblast stem cells.
Because the methylation-sensitive restriction endonuclease assay used to generate the data shown in Figure 1 provides a ratio of maternal alleles at which MluI sites are methylated to paternal alleles at which MluI sites are methylated rather than an absolute fraction of all alleles, we also assayed DNA methylation at the IGF2/H19 DMR by bisulfite pyrosequencing. The assay we used quantifies the methylation status of five CpGs in the IGF2/H19 DMR that are adjacent to the CpG queried in the MluI assay. If an M/P ratio greater than zero (Figure 1) represents methylation of maternal alleles in addition to methylation of DMRs on all paternal alleles, then the CpG sites in these samples/individuals should be methylated on greater than 50% of molecules assayed by bisulfite pyrosequencing (i.e. all paternal alleles plus some fraction of maternal alleles). The placenta samples with M/P ratios greater than zero show greater than 50% methylation at all five CpGs in almost all cases (Figure 2) by bisulfite pyrosequencing, indicating that both the methylation-sensitive restriction endonuclease assay and the bisulfite pyrosequencing assay are measuring gain of methylation on maternal alleles, as has also been reported in the mouse [19], [22], [23].
The current model for imprinted transcriptional control of IGF2/H19 correlates methylation at the ICR/DMR with an inability to bind CTCF, transcriptional silencing of H19 and transcriptional activation of IGF2 [37], [38]. Because our analysis of DMR methylation indicates that all paternal and some maternal DMRs are methylated, we examined mRNA expression in individuals with varying levels of bi-allelic DMR methylation, for the presence of bi-allelic transcripts of IGF2 and H19. We assayed individuals who were informative for an ApaI polymorphism (in exon 9) for the presence of transcripts from both IGF2 alleles [32]. We also assayed individuals who were informative for an RsaI polymorphism (in exon 5) for the presence of transcripts from both H19 alleles [39], which would be expected to occur if paternal DMRs became demethylated.
Although we detected minor amounts of presumed maternal IGF2 mRNA in several individuals, there was no correlation between maternal DMR methylation and amount of transcript from the maternal allele (Table 3). We also observed only small amounts of presumed paternal allele expression of H19 mRNA, which suggested that there was no loss of methylation on paternal alleles in these samples (Table 3), also consistent with the overall greater than 50% methylation observed by pyrosequencing (Figure 2).
We investigated intra-individual variation in allele-specific methylation at an IGF2R DMR we have also examined previously [28]. The rationale for examining allele-specific methylation at this non-transcriptionally imprinted locus is that this locus may be a more sensitive reporter of any disruption in CpG site methylation by environmental factors because such changes are not predicted to affect transcription and are less likely to be selected against. The relative level of CpG methylation on paternal and maternal alleles at this locus was measured in cord blood, cord and five sections of placenta taken from populations of children conceived either in vitro or in vivo. The less methylated allele is assumed to be paternal because we found no individuals in which the paternal allele was more methylated than the maternal allele in 112 informative individuals for whom allelic inheritance could be confirmed by pedigree analysis [28].
The parental origin-specifically methylated IGF2R DMR is located in the second intron of IGF2R and is normally methylated on the maternal allele. Although the human IGF2R gene is transcribed from both alleles [40] differential methylation of maternal and paternal alleles is maintained in the human, as it is in the mouse [26] and a small fraction of the human population may have transcriptional imprinting of IGF2R [27]. We used a single nucleotide polymorphism within an MspI site to identify maternal and paternal alleles of informative (heterozygous) individuals and a methylation-sensitive restriction endonuclease (NotI) to determine the methylation status of a specific CpG site within the DMR [28]. We identified 28 in vitro and 27 in vivo individuals who were informative and assayed allele-specific methylation as described previously [28].
We calculated the ratio between the DNA methylation levels on paternal and maternal (P/M) alleles as an indicator of methylation imprint status. A ratio of zero corresponds to exclusive methylation of the NotI site on the maternal allele, while a ratio of one signifies methylation of this site on an equal number of paternal and maternal alleles. In the case of controls, no uncleaved C alleles were detected in any C/C homozygous individuals, indicating that MspI cleaved the PCR products with >99% efficiency.
Although preferential methylation of the presumed maternal allele was observed in almost all individuals (Figure 3), the distribution of paternal/maternal (P/M) methylation ratios at the IGF2R DMR in cord blood and cord showed that most individuals have an easily measurable level of methylation at the CpG within the NotI cleavage site on the presumed paternal allele (P/M>0.1, Figure 3), as has also been observed previously in peripheral blood from the CEPH families [28].
In cord blood, only 34% of the total population has low levels of methylation on the paternal allele (P/M allele ratios of <0.1) while a very small fraction of individuals (2%) have P/M allelic methylation ratios greater than 0.5. The distribution of individual P/M methylation ratios in cord (Figure 3) also showed a similar pattern to what was observed in cord blood. However, results from five placenta sections taken from the same individuals showed nearly complete loss of the methylation imprint (i.e. P/M ratios close to 1) at this locus in samples from multiple individuals in both in vitro and in vivo groups (Figure 3).
We found no difference in mean P/M ratios in cord blood, cord or placenta between the in vitro and in vivo groups, either comparing individual tissue types or combining all samples (Table 4). Cord blood allele-specific methylation ratio variance was greater in the in vitro group (P = 0.0016) but we did not attempt to calculate a cord blood stem cell number comparison because of the small number of samples on which to model the distribution. There was no significant difference in the population variance in the in vitro group in cord or placenta, although the presence of a substantial fraction of samples in both groups for which nearly complete loss of the methylation imprint (P/M>0.9) was observed is likely to affect our ability to distinguish such a difference.
We measured steady-state IGF2, H19 and IGF2R mRNA levels in cord blood and placenta from children conceived in vitro or in vivo (Table 5, Figure 4). In addition to the children who were informative for allele specific DMR methylation (Figure 1 and Figure 3), we also measured mRNA levels in the children who were not informative. Mean cord blood IGF2R mRNA levels were significantly lower in the in vitro group (fold change = 0.61, P = 0.0039). Mean placental IGF2 and H19 mRNA levels were also significantly lower in the in vitro group (fold change = 0.52, P<0.0001, and fold change = 0.72, P = 0.0193, respectively).
We have examined intra- and inter-individual variation in DNA methylation at the differentially methylated regions (DMRs) of the IGF2/H19 and IGF2R loci in cord blood, cord and five sections of placenta from a population of children conceived in vitro or in vivo. Although a significant fraction of individuals in both groups do appear to maintain the IGF2/H19 methylation imprint “correctly”, with M/P ratios of <0.1 (93.6% of peripheral blood samples are below this M/P ratio, Sandovici et al., 2003) [28], we found substantial intra-individual and inter-individual variation in allele-specific methylation in both groups in all three tissues: 8/46 individuals in the in vivo group and 7/40 individuals in the in vitro group have cord blood M/P ratios above this level (n.b. of the 56 informative in vivo and 45 informative in vitro individuals shown in Figure 1, cord blood samples were unavailable for 10 of the in vivo and five of the in vitro children). The acquisition of CpG methylation on only a fraction of maternal IGF2/H19 DMRs in so many individuals suggests that there is extensive population level variation in the time at which methylation imprints become set during development, especially in extraembryonic lineages. This assertion receives further support from the analysis of intra-individual variation in P/M methylation ratios at an IGF2R DMR. Very few individuals (one in the in vivo group, two in the in vitro group) maintained this imprint “correctly” in all samples (even if the threshold for “correctly” is reduced to P/M ratios <0.2) and the discrepancy between maintaining the imprint in embryonic and extraembryonic tissue is even more pronounced (Figure 3). The observation of greater epigenetic variation in extraembryonic than in embryonic tissues is consistent with observations at a number of imprinted loci, including IGF2/H19, in the mouse [11]. We believe this is the first time this observation has been made in the human. Parental allele-specific mean methylation ratios at the IGF2/H19 DMR were greater in the in vitro group, indicating acquisition of methylation on maternal alleles. This finding is consistent with observations made at Igf2/H19 in the mouse [23].
The total variance in M/P ratio at IGF2/H19 was significantly greater in cord blood and cord from the in vitro group, with a suggestive P-value (0.0620) for placenta. These findings indicate an association between ART and the magnitude of the variance in parental allele-specific methylation patterns. The mechanism by which greater variance might be created in the in vitro group is unclear but could be related to the number of trophoblast stem cells that give rise to the placenta in each group. If fewer trophoblast stem cells give rise to the placenta in the in vitro group, one expects greater intra-individual variation in somatically heritable epigenetic marks if the population of stem cells contains cells with more than one epigenetic state (i.e. IGF2/H19 DMRs that are methylated only on the paternal allele in some cells and methylated on both maternal and paternal alleles in others).
As an independent test of this prediction, we measured X chromosome inactivation ratios in five sections of placenta in females from the in vitro (50 individuals) and in vivo groups (54 individuals) in order to estimate the number of trophoblast stem cells that give rise to the placenta in each group. The closest fit to the in vitro X-inactivation distribution corresponds to nine trophoblast stem cells while the closest fit to the in vivo distribution corresponds to 11 trophoblast stem cells. We additionally estimated the number of trophoblast stem cells in each group by comparing the distribution of M/P IGF2/H19 DMR methylation ratios from the two comparison groups. The closest fit to the distribution of M/P ratios in placenta in children conceived in vitro corresponds to eight trophoblast stem cells and the closest fit in children conceived in vivo corresponds to 10 trophoblast stem cells. Overall, these estimates of between group intra-individual epigenetic variation are consistent with the prediction that overall greater variance in M/P IGF2/H19 DMR methylation ratios in the in vitro group is associated with fewer trophoblast stem cells.
Although our statistical estimates of the number of trophoblast stem cells is imprecise, it gives one some confidence that similar absolute numbers are obtained (Table 2) from estimates of epigenetic variance at two different loci, examining different individuals with each assay. Whether the calculated difference in number of trophoblast stem cells reflects designation of trophoblast stem cells at an earlier stage of development (when fewer cells are present in the embryos) or whether embryos from the in vitro group have fewer cells than in vivo embryos at comparable stages cannot be determined from these data, however, previous reports suggest that in vitro mouse embryos may contain fewer cells than in vivo embryos at the same developmental time [41]. Our approach of using epigenetic variance to calculate the number of trophoblast stem cells is, for obvious reasons, the only opportunity for estimating this number in the human because direct comparisons of cell numbers in in vitro and in vivo embryos is not possible.
We note that while we have uncovered locus-specific differences in the level of epigenetic variation in children conceived in vitro, we cannot distinguish whether the differences are due to some aspect of the assisted reproduction process or is related to the underlying infertility. In fact, the characteristic of epigenetic variance, itself, may be under genetic control and may also be influenced by the environment [42]. Greater variance in trait value, even without changes in trait mean, is predicted to have a substantial positive effect on fitness in a changing environment [42]. In this regard, we note that in vitro conception is associated with at least two changes of environment (hormonal stimulation, retrieval of ova from the maternal environment to fertilization and culture in vitro, followed by return to the maternal environment). A larger-scale epigenetic screen is required in order to determine whether there is a tendency for in vitro conception to be associated with overall increased variance of epigenetic marks.
Several observations are noteworthy about the steady-state mRNA levels measured for IGF2 in placenta. First, IGF2 mRNA levels in placentas from in vitro conceived children, as a group, were approximately half of what was observed in children conceived in vivo. This observation is consistent with experiments demonstrating reduced Igf2 mRNA levels in placentas from mouse embryos subject to in vitro manipulations [20]. Second, reduction in IGF2 mRNA levels in the human placentas does not occur in conjunction with loss of methylation at the paternal DMR, as expected if transcript levels are controlled by genomic imprinting alone. Furthermore, we did not observe that increased levels of methylation at the maternal DMR induced a coordinate level of transcription from the maternal IGF2 allele. In fact, given that IGF2 transcript levels vary by more than an order of magnitude between individuals (Figure 4) and almost by that much between samples within a single placenta (Figure S1), the mechanism by which natural selection might act in a population, on a process whose postulated design is to reduce transcription by half (from two alleles to one) is unclear. Along these same lines, we observed no correlation between birth weight and IGF2 transcript levels in either placenta or cord blood, whether or not birth weights were corrected for gestational age (Figure S2). This last observation was not completely unexpected, as several laboratories have failed to observe a correlation between IGF2 mRNA levels and birth weight [43]–[45]. In this regard, it is likely that epigenetic marking of genes according to parental origin plays an important role in other processes associated with reproduction and the formation of gametes, such as chromosome pairing and recombination [46]–[49]. The selective force for the maintenance of imprinting in these processes is both direct (successful recombination is required for successful gametogenesis) and related to reproductive success.
We also observed that mean steady-state levels of IGF2R mRNA were lower in cord blood from the in vitro group. This locus does not appear to be transcriptionally imprinted in most humans [40], [50], although the preferential methylation of only one parent's allele (the maternal) is conserved [26], [27]. Although not transcriptionally imprinted, we did note an inverse correlation between methylation of the paternal allele and overall transcript level, indicating that “aberrant” methylation of the “incorrect”, paternal allele does have a small effect (accounting for ∼10% of the variance, Figure S3). Lower IGF2R mRNA level in the in vitro group is, on the face of it, in contrast to expectations. If IGF2R is a receptor that acts as a “sink” for IGF2 [51], [52], one might expect children conceived in vitro to have higher levels of IGF2R because they have a higher probability of low birth weight [53]. In any case, we did not find any correlation between birth-weight, IGF2 levels, IGF2R levels or IGF2/IGF2R ratios (Figure S2).
Overall, our results indicate that epigenetic modifications at IGF2/H19 and IGF2R are subject to frequent changes during early development, especially in extraembryonic tissues. Although not all of the epigenetic changes appear to be manifested as significant differences in DNA methylation, conception in vitro is associated with gene expression differences for all three genes in some tissues. Whether the gene expression differences between in vitro and in vivo groups are also a manifestation of what appears to be a smaller number of trophoblast stem cells in children from the in vitro group is a subject for future investigation.
The cases/in vitro group are newborns conceived by assisted reproductive technology at a single infertility treatment center so that the clinical and laboratory procedures are uniform. The parents of the control/in vivo group had no prior history of infertility and the index pregnancy was achieved without medical assistance, such as the use of infertility medications or treatments. All the in vitro patients were stimulated with commercially available gonadotrophin preparations. The embryo culture media and the incubation parameters were all the same. The cases and controls were matched with regards to maternal age, race and gestational age (Table S1). Written, informed consent was obtained in advance from the mother of each newborn (University of Pennsylvania I.R.B. approved protocol no. 804530). A summary of the assays used, number of individuals studied and the tissues investigated is provided in Table S2.
Cord blood, cord and placenta samples were collected from each in vitro and in vivo newborn. All cord blood samples were collected within 20 min of delivery. Tissue samples were stored at 4°C after delivery, and samples were collected within five hours of delivery [54]. The umbilical cord was wiped with normal saline and the cord vein was punctured with a 21G needle. Whole cord blood (6–10 ml) was collected in lavender topped vacutainer tubes at room temperature. The sample was shaken thoroughly to prevent clotting as the tube contains EDTA, ethylenediaminetetraacetic acid. An aliquot (3–4 ml) of cord blood was transferred to a 15 ml Falcon tube containing RNALater RNA Stabilization Reagent (Ambion, USA), following the manufacturers guidelines, to stabilize the RNA. The remaining cord blood in the lavender topped vacutainer tubes was saved for blood DNA extraction. All cord blood DNA and RNA samples were initially stored at 4°C, and nucleic acid extractions were performed within 2–4 days of collection.
Placental tissue (1.5–2.5 cm3) was excised from the fetal surface of the placenta and rinsed extensively with sterile saline solution to minimize maternal blood contamination. Each placenta was sampled from four quadrants and from directly behind the cord insertion site (this sample was used for the RT-PCR and pyrosequencing assays, as well as for the allele-specific methylation assays). A segment of umbilical cord (2 cm) was cut and treated in a similar fashion. Placental and cord tissue for RNA extraction were chopped into small pieces (0.5 cm3) and immersed in RNALater RNA Stabilization Reagent (Ambion, USA), following the manufacturers guidelines, as soon as possible after collection. All tissue DNA and RNA samples were initially stored at 4°C, and tissue digestion and nucleic acid extractions were performed within 2–4 days of collection. Approximately 4–5 mg of tissue was used for the DNA and RNA extraction procedures, and the remaining tissue was stored at −80°C.
Cord blood DNA was isolated using the ArchivePure DNA Blood Kit (Fisher Scientific Company, USA) following the manufacturers guidelines. Tissue genomic DNA was extracted using standard phenol-chloroform extraction methods. The isolated DNA was dissolved in 10 mM TrisCl, pH 8.0, quantified using a spectrophotometer and stored at −80°C until further use. Cord blood RNA was isolated using the PerfectPure RNA Blood Kit (Fisher Scientific Company, USA) following the manufacturers guidelines.
Total cellular RNA was extracted from each tissue sample using TRIzol Reagent (Invitrogen Corporation, USA), according to the manufacturers instructions. The isolated RNA was dissolved in Milli-Q water, quantified using a spectrophotometer and stored at −80°C until further use.
There are many DMRs on chromosome 11, but the most consistent observations indicating a role in the control of transcription of the IGF2 and H19 genes involve a CpG island located in a 5 kb region centromeric to the H19 gene, known as the IGF2/H19 DMR [25]. CpG sites within this DMR on the paternal allele are normally methylated, while those on the maternal allele are normally unmethylated [30], [55]–[58]. This region also contains seven different binding sites for the CTCF protein [38] and the methylation status of the sixth binding site was found to be most consistently associated with the transcriptional status of both IGF2 and H19 [24].
The upstream H19 sequence used in this study is available from GenBank (accession number AF125183). Allele-specific methylation was investigated by screening the DNA samples for a C/T polymorphism recognized by CfoI at the IGF2/H19 DMR (near the sixth binding site for CTCF) [24]. After identifying maternal and paternal alleles of heterozygous individuals, a methylation-sensitive restriction endonuclease (MluI) was used to determine the methylation status of specific CpG sites within the DMR. If all paternal alleles are methylated and all maternal alleles are unmethylated at these sites in a sample of genomic DNA, then all maternal alleles should be cleaved by MluI while all paternal alleles will remain uncleaved. Amplification of the region by PCR using primers that flank the MluI site should amplify only paternal alleles (identified by post-PCR cleavage with CfoI). Amplification of maternal alleles indicates resistance to cleavage by MluI. This may occur as a result of methylation of the CpG site within the MluI recognition sequence (the principle upon which the assay is based), mutation of the MluI site or technical artifact. The latter two possibilities may be distinguished from the first by DNA sequencing, assay reproducibility and use of additional methylation-sensitive restriction endonucleases.
The IGF2R DMR is located in the second intron of IGF2R and is normally methylated on the maternal allele. The sequence of IGF2R is available from GenBank (accession number AF069333). Allele-specific methylation at the IGF2R DMR was investigated by screening the DNA for a C/T polymorphism recognized by MspI. After identifying maternal and paternal alleles of heterozygous individuals, a methylation-sensitive restriction endonuclease (NotI) was used to determine the methylation status of specific CpG sites within the DMR.
Genomic DNA (100 ng) from informative individuals was digested overnight at 37°C with an excess of a methyl-sensitive restriction endonuclease: MluI and NotI for the IGF2/H19 and IGF2R DMRs, respectively. Control individuals who were homozygous for C alleles and homozygous for T alleles were also analyzed in each experiment.
After digestion, the enzymes were denatured and the digested DNA was amplified in a hot-stop PCR assay using the following primers: IGF2-F 5′-GAGATGGGAGGAGATACTAGG-3′ and IGF2-R 5′-GTCAGTTCAGTAAAAGGCTGG-3′ for the IGF2/H19 DMR, and IGF2R-F 5′-GGCCGAGGCCTGGCATGTTGG -3′ and IGF2R-R 5′-TGGGGAAGCGCGAGAGGCCTAGG-3′ for the IGF2R DMR. After 30 cycles at 94°C for 30 s, 50°C for 30 s (IGF2/H19) or 63°C for 30 s (IGF2R), and 72°C for 1 min, we added 3 µCi α-32P dCTP for one additional cycle and a final elongation step (72°C for 7 min). PCR products were then digested overnight at 37°C with the enzyme used for identifying the parental origin of the alleles (CfoI for IGF2/H19 and MspI for IGF2R). The samples were separated on denaturing 5% polyacrylamide gels and the intensity of the bands (alleles) were quantified using a PhosphoImage Reader FLA 5000 (FUJIFILM Medical Systems USA, Inc.).
We used a custom pyrosequencing assay for the IGF2/H19 DMR (NCBI36:11,2019856-2019740) which included five CpGs. Genomic DNA (500 ng per sample) was bisulfite treated using EZ Gold DNA Methylation Kit (Zymo Research, USA) following the manufacturers protocol.
Bisulfite treated DNA was used for generating PCR amplified templates for pyrosequencing. The PCR primer sequences were: forward 5′- GGGGTTATTTGGGAATAGG-3′ and biotin labeled reverse, 5′- CCAAACCATAACACTAAAACCCTC-3′. The PCR reaction (30 µl) was following: 25 ng of bisulfite DNA, 0.75 U HotStar Taq Polymerase (Qiagen, USA), 1X PCR buffer, 3 mM MgCl2, 200 µM of each dNTPs, 6 pmol forward primer and 6 pmol reverse primer. Recommended PCR cycling conditions were: 95°C for 15 min; 45 cycles (95°C for 30 s; 60°C for 30 s; 72°C for 30 s); 72°C for 5 min. The biotinylated PCR product (10 µl) was used for each sequencing assay with the following sequencing primer: 5′- GAATAGGATATTTATAGGAG-3′. Pyrosequencing was done using the PSQ96HS system according to standard procedures using Pyro Gold Reagent kits (Biotage, Sweden). Methylation was quantified using Pyro Q-CpG Software (Biotage, Sweden), which calculates the ratio of converted C's (T's) to unconverted C's at each CpG and expresses this as a percentage methylation.
First-strand cDNA was obtained using SuperScript III Reverse Transcriptase (RT) (Invitrogen Corporation, USA). To produce cDNA from total RNA, a mixture containing 0.5–1 µg extracted total RNA, 0.5 µg oligo(dT)18 primer and 1 µl dNTP mix (10 mM each) in a final 13 µl reaction volume was heated to 65°C for 5 min, cooled down on ice for 1 min, and then added to a 7 µl reaction mixture containing 4 µl SuperScript III RT buffer (10), 1 µl DTT (0.1 M), 1 µl RNaseOUT Recombinant RNase inhibitor (40 U/µl; Invitrogen Corporation, USA) and 1 µl SuperScript III M-MLV reverse transcriptase (200 U/µl). The samples were mixed and incubated at 50°C for 60 min. Reactions were terminated at 70°C for 15 min and the RT products were stored at −20°C until further use.
Quantitative real-time RT-PCR assays were carried out using a 7700 Sequence Detector (Applied Biosystems, USA). GAPDH, which has previously been used as a housekeeping gene in placenta by several investigators [59]–[61], was used as the housekeeping gene. All the placental tissue samples were from the third trimester. There was a positive correlation between GAPDH expression and the expression of another commonly used housekeeping gene HPRT, when studied in the same samples (Figure S4).
Steady-state mRNA levels of IGF2, H19, IGF2R and housekeeping gene GAPDH were measured using gene-specific primers and QuantiFast SYBR Green PCR Master Mix (Qiagen, USA). The primer sequences were following: IGF2 Forward 5′-TCTGACCTCCGTGCCTA-3′, IGF2 Reverse 5′-TTGGGATTGCAAGCGTTA-3′, H19 Forward 5′-AGAAGCGGGTCTGTTTCTTTA-3′, H19 Reverse 5′-TGGGTAGCACCATTTCTTTCA-3′, IGF2R Forward 5′-ACCTCAGCCGTGTGTCCTCT-3′, IGF2R Reverse 5′-CTCCTCTCCTTCTTGTAGAGCAA-3′, GAPDH Forward 5′-GAGTCAACGGATTTGGTCGT-3′, and GAPDH Reverse 5′-TTGATTTTGGAGGGATCTCG-3′. PCR reactions were performed by mixing 1 µl of cDNA (50 ng/µl placenta, 25 ng/µl cord blood) with 24 µl of reaction mixture (12.5 µl QuantiFast SYBR Green PCR Master Mix (2X), 2.5 µl forward primer (10 µM), 2.5 µl reverse primer (10 µM), and 6.5 µl nuclease free dH2O) and amplified under the following conditions: 95°C for 5 min, followed by 40 cycles of 95°C for 10 s and 60°C for 30 s. A melting curve analysis of the PCR products was performed to verify their specificity and identity. PCR products were also run on 2% agarose gels to confirm the size of the amplified products. Relative gene expression levels were obtained using the ΔΔCT method [62].
To avoid genomic DNA contamination during imprinting analysis, PCR was done across an intron-exon boundary and the cDNA products were gel-purified. The primers used for assaying IGF2 imprinting were: primer 1, 5′−ATCGTTGAGGAGTGCTGTTTC−3′; primer 2, 5′−CGGGGATGCATAAAGTATGAG−3′; primer 3, 5′−CTTGGACTTTGAGTCAAATTGG−3′; and primer 4, 5′−GGTCGTGCCAATTACATTTCA−3′ [32]. Heterozygosity of an ApaI polymorphism in exon 9 of IGF2 was ascertained by doing PCR with genomic DNA using primers 3 and 4 across the ApaI site, and the PCR product was digested with ApaI. Imprinting status was ascertained by doing RT-PCR, using primers 1 and 2 in exons 8 and 9, respectively. The cDNA PCR product, which is shorter than any possible contaminating genomic DNA product because of intron splicing, was electrophoresed and purified from a 2% agarose gel using the QIAquick Gel Extraction Kit (Qiagen, USA) following the manufacturers protocol. Hot-stop PCR was then done using primers 3 and 4, with α-32P dCTP added before the last cycle. The PCR product was digested with ApaI and then separated on denaturing 5% polyacrylamide gels and the intensity of the bands (alleles) were quantified using a PhosphoImage Reader FLA 5000 (FUJIFILM Medical Systems USA, Inc.).
The primers used for assaying H19 imprinting were: H19-1, 5′-GGAGTTGTGGAGACGGCCTTGAGT-3′; H19-2, 5′-CCAGTCACCCGGCCCAGATGGAG-3′; and H19-3, 5′-CTTTACAACCACTGCACTACCTGAC-3′. Heterozygosity of an RsaI polymorphism in exon 5 of H19 was ascertained by doing PCR with genomic DNA using primers H19-1 and H19-2 across the RsaI site, and the PCR product was digested with RsaI. Imprinting status was ascertained by doing RT-PCR, using primers H19-3 and H19-2 in exons 4 and 5, respectively. The cDNA PCR product, which is shorter than any possible contaminating genomic DNA product because of intron splicing, was electrophoresed and purified from a 2% agarose gel using the QIAquick Gel Extraction Kit (Qiagen, USA) following the manufacturers protocol. Hot-stop PCR was then done using primers H19-1 and H19-2, with α-32P dCTP added before the last cycle. The PCR product was digested with RsaI and then separated on denaturing 5% polyacrylamide gels and the intensity of the bands (alleles) were quantified using a PhosphoImage Reader FLA 5000 (FUJIFILM Medical Systems USA, Inc.).
X-chromosome inactivation ratios were assayed using previously published modifications of a methylation-sensitive PCR assay [33], [36], [63]–[65]. We measured the methylation status of a CpG site that is correlated with the expression of alleles at the X-linked, highly polymorphic androgen receptor (AR) locus. Genomic DNA from cord blood, cord and five sections of placenta was available for 50 in vitro and 54 in vivo females who were heterozygous for AR alleles that differed by more than one CAG repeat.
Genomic DNA from females who were heterozygous (informative) at the highly polymorphic (CAG)n repeat of the X-linked AR gene was amplified after previous overnight digestion with HhaI methyl-sensitive restriction endonuclease, with primers (AR1: 5′-AGAGGCCGCGAGCGCAGCAC-3′ and AR2: 5′-ACTCCAGGGCCGACTGCGGC-3′), which flank the repeat and two HhaI sites. We added a radiolabeled nucleotide for the last cycle of ‘hot-stop’ PCR, rather than a single end-labeled primer, to increase the signal. After 27 cycles at 94°C for 1 min, 68°C for 1 min, and 72°C for 1 min, 3 µCi α-32P dCTP was added for one additional cycle. PCR products were separated on denaturing 5% polyacrylamide gels and the intensity of the alleles was quantified by using the PhosphoImage Reader FLA 5000 (FUJIFILM Medical Systems USA, Inc.) [65]. As a way of quantifying the degree of skewing, i.e., the degree to which the somatic cells of an individual female deviated from a 1∶1 ratio, the intensity of the upper allele divided by the sum of the intensities of both alleles was computed for each individual.
The statistical significance of the methylation datasets representing the in vivo and in vitro group was examined using the Wilcoxon Rank Sums Test. Data from the real time RT-PCR experiments were analyzed using Student's T-test. The number of cells was estimated using the method described by Amos-Landgraf et al. (2006) and Mclaren A (1972) [34], [35]. P-values ≤0.05 were considered significant.
|
10.1371/journal.pcbi.1003074 | “Gate-keeper” Residues and Active-Site Rearrangements in DNA Polymerase μ Help Discriminate Non-cognate Nucleotides | Incorporating the cognate instead of non-cognate substrates is crucial for DNA polymerase function. Here we analyze molecular dynamics simulations of DNA polymerase μ (pol μ) bound to different non-cognate incoming nucleotides including A:dCTP, A:dGTP, A(syn):dGTP, A:dATP, A(syn):dATP, T:dCTP, and T:dGTP to study the structure-function relationships involved with aberrant base pairs in the conformational pathway; while a pol μ complex with the A:dTTP base pair is available, no solved non-cognate structures are available. We observe distinct differences of the non-cognate systems compared to the cognate system. Specifically, the motions of active-site residue His329 and Asp330 distort the active site, and Trp436, Gln440, Glu443 and Arg444 tend to tighten the nucleotide-binding pocket when non-cognate nucleotides are bound; the latter effect may further lead to an altered electrostatic potential within the active site. That most of these “gate-keeper” residues are located farther apart from the upstream primer in pol μ, compared to other X family members, also suggests an interesting relation to pol μ's ability to incorporate nucleotides when the upstream primer is not paired. By examining the correlated motions within pol μ complexes, we also observe different patterns of correlations between non-cognate systems and the cognate system, especially decreased interactions between the incoming nucleotides and the nucleotide-binding pocket. Altered correlated motions in non-cognate systems agree with our recently proposed hybrid conformational selection/induced-fit models. Taken together, our studies propose the following order for difficulty of non-cognate system insertions by pol μ: T:dGTP<A(syn):dATP<T:dCTP<A:dGTP<A(syn):dGTP<A:dCTP<A:dATP. This sequence agrees with available kinetic data for non-cognate nucleotide insertions, with the exception of A:dGTP, which may be more sensitive to the template sequence. The structures and conformational aspects predicted here are experimentally testable.
| DNA polymerase μ (pol μ) is an enzyme that participates in DNA repair and thus has a central role in maintaining the integrity of genetic information. To efficiently repair the DNA, discriminating the cognate instead of non-cognate nucleotides (“fidelity-checking”) is required. Here we analyze molecular dynamics simulations of pol μ bound to different non-cognate nucleotides to study the structure-function relationships involved in the fidelity-checking mechanism of pol μ on the atomic level. Our results suggest that His329, Asp330, Trp436, Gln440, Glu443, and Arg444 are of great importance for pol μ's fidelity-checking mechanism. We also observe altered patterns of correlated motions within pol μ complex when non-cognate instead of cognate nucleotides are bound, which agrees with our recently proposed hybrid conformational selection/induced-fit models. Taken together, our studies help interpret the available kinetic data of various non-cognate nucleotide insertions by pol μ. We also suggest experimentally testable predictions; for example, a point mutation like E443M may reduce the ability of pol μ to insert the cognate more than of non-cognate nucleotides. Our studies suggest an interesting relation to pol μ's unique ability to incorporate nucleotides when the upstream primer is not paired.
| The integrity of genetic information depends largely on DNA polymerases that are central to DNA replication, damage repair, and recombination. DNA polymerase errors are associated with numerous diseases, including various cancers and neurological conditions [1]–[13]. One of the most basic types of errors that DNA polymerases generate is base substitution error, which means that DNA polymerase inserts an non-cognate (“non-cognate”) nucleotide opposite the DNA template base to form a nonstandard base pair (i.e., A:dATP base pair, instead of A:dTTP base pair). Although DNA polymerases conduct similar nucleotidyl transfer reaction and share a similar structure - palm, thumb and fingers subdomains [14], they can exhibit varying levels of accuracy (“fidelity”) in inserting nucleotides [15].
DNA polymerase μ (pol μ) of the X-family, like the other X-family members, participates mainly in DNA repair rather than replication [16]. Like two other X-family members polymerase β (pol β) and polymerase λ (pol λ), pol μ can bind to DNA and fill single-strand DNA gaps in a template-dependent manner with moderate fidelity (10−4–10−5) [17]–[19]. Furthermore, like another X-family member terminal deoxynucleotidyl transferase (Tdt), pol μ can also perform nucleotide insertion in a template-independent manner [20], [21]. In addition, pol μ can direct template-based DNA synthesis without requiring all upstream primer bases to be paired [17], [22]. The unique substrate flexibility of pol μ may signal a unique role in the nonhomologous DNA end joining (NHEJ) process for double-strand breaks in DNA and V(D)J recombination [22]–[28].
Structural and computational studies have uncovered important differences and similarities regarding how pol μ incorporate a cognate nucleotide into single-nucleotide gapped DNA, compared to other X-family members [29], [30]. For pol β, upon binding the cognate incoming nucleotide, the enzyme undergoes a large-scale protein motion in the thumb subdomain from open (inactive) to closed (active) conformation [31]–[33]. Such open-to-closed protein motion is also observed in pol X, another X-family polymerase [34]. Pol λ lacks such large-scale protein transitions; instead, a large shift of the DNA template from the inactive to the active state is indicated by both crystal structures [35] and simulations [36]. The large-scale protein motion in pol β and pol X or DNA motion in pol λ is crucial for the polymerization activity [31], [37], [38]. In pol μ, studies have suggested the lack of significant DNA or protein motion before chemistry [29]. Pol μ shares with pol β and pol λ the notion that subtle active-site protein residue motions help organize the conformation of the active site and prepare for the following chemical step [39], but the specific residues are different [29], [33], [36]. In pol μ, His329 and Asp330 assemble pol μ's active site, and Gln440 and Glu443 help accommodate the incoming nucleotide. See Fig. 1(a) and (b) for key residues and their motion in pol μ's cognate system.
In prior mismatch studies on various X-family DNA polymerases such as pol β [40]–[45], pol X [46], and pol λ [47], reduced large-scale protein (pol β and pol X) or DNA motions (pol λ) were observed, related to the inactivity of non-cognate systems. Varying amounts of active-site distortions are observed. Distortions of the active site are caused by the conformational changes of several key residues (“gate-keepers”) [40], [47]. For example, in pol β, structural and dynamics analyses revealed different behavior of Arg258, Asp192, and Phe272 in non-cognate systems [40]–[43]. These residues distort the active site, with the degree of active-site distortions system-dependent and in accord with the sequence of kinetic data for non-cognate base pair incorporations. The different conformational behavior between the cognate and non-cognate systems before and/or after chemistry are also observed and are related to fidelity for DNA polymerases in other families [48]–[51].
From prior results, we further demonstrated that characteristic motions recur within various 2′-deoxyribonucleoside 5′-triphosphate (dNTP) contexts. Specifically, correlated protein and dNTP motions occur within cognate dNTP complexes and are altered within non-cognate dNTP complexes. We therefore proposed a hybrid conformational selection/induced-fit model for DNA polymerases [52]. In this model, the cognate dNTP selectively binds to a near-active conformation from an ensemble of possible polymerase/DNA conformations, and then the bound dNTP induces small adjustments within the active site, driving the complex to a fully-active state ready for catalysis. Non-cognate dNTPs that are relatively efficiently handled by the polymerase would also selectively bind to a near-active conformation, but the active-site changes induced by the non-cognate dNTP binding would differ from those by cognate dNTP binding. For non-cognate dNTPs that are relatively poorly inserted by the polymerase, dNTP may bind to a variable inactive conformation. The resulting incomplete organization of the active site would reduce the efficiency for inserting an non-cognate dNTP. This proposed broader view better reflects both the intrinsic motions of polymerases and the highly specific nature of polymerase/ligand interactions, and has gained further support from additional computations [53]–[57] (Arora, Zahran, and Schlick, in preparation).
Several key experimental studies of pol μ's fidelity exist [17], [19], but no structure of an non-cognate incoming nucleotide bound to pol μ has been reported. Modeling and all-atom dynamics simulations can help study the structural and dynamic properties of non-cognate pol μ systems, which in turn can be related to specific functions of pol μ. Needless to say, all dynamics simulation data are subject to the approximations and limitations of an empirical force field, limited sampling, and large computational requirements [58]. Yet, modeling and simulation have demonstrated many successes in biomolecular structure and function problems, and can be valuable especially when few experimental data are available [59].
In this study, we investigate dynamics of pol μ bound to various mismatches (A:dCTP, A:dATP, A:dGTP, T:dCTP, and T:dGTP) to determine the factors that contribute to insertion differences of pol μ during its conformational pathway before chemistry. We also analyze simulations of the bulky purine-purine mismatches with the template base in both the anti and syn orientations to determine whether particular base pair geometry might facilitate mismatch incorporation. We find that His329 and Asp330 near the active site help discriminate cognate from non-cognate incoming nucleotides. In addition, we suggest that Trp436, Gln440, Glu443, and Arg444 play the role of “gate-keepers” in pol μ by tightening (deactivating) the nucleotide-binding pocket when non-cognate nucleotides are bound. Compared to pol β and pol λ, most of these residues are much farther from the upstream primer in pol μ. A comparison of the correlated motions in cognate and non-cognate pol μ systems indicates decreased interactions in non-cognate systems, especially those between the incoming nucleotides and the nucleotide-binding pocket, and suggests that pol μ also fits into the hybrid conformational selection/induced-fit model. As in pol β and pol λ, the degree of active-site distortion in pol μ mirrors trends in kinetic data, except for A:dGTP, which is more disordered and sequence-context dependent as indicated by kinetic data. Though the chemical step can also impact the fidelity of pol μ, the conformational pathway is a pre-requisite for chemistry [39]. Indeed, in non-cognate systems, the conformational pathway produces a deformed active site that is farther from the chemistry competent state. Thus, even if the chemical step is hindered in non-cognate systems, it is the distorted conformational pathway that leads to initial hindrance in the chemical pathway. Finally, we suggest that the ability of pol μ to incorporate nucleotides when the upstream primer is not paired may arise in part from the fact that most “gate-keeper” residues in pol μ are much farther from the upstream primer, compared to pol β and pol λ; thus, pol μ may be less sensitive to changes around the upstream primer.
Seven pol μ non-cognate models were prepared on the basis of the X-ray crystal murine pol μ cognate ternary complex (PDB entry 2IHM) [30]. In the crystal structure, two loops in the palm (Loop1, His366-Arg389; Loop2, Pro397-Cys411) are partially missing. Missing protein residues His366-Val386 and Ala403-Ala405 were inserted with the InsightII package (Accelrys Inc., San Diego, CA). A hydroxyl group was added to the 3′ carbon of the 2′,3′-dideoxythymidine 5′-triphosphate (ddTTP) sugar moiety to form 2′-deoxythymidine 5′-triphosphate (dTTP). All other missing atoms from the crystal structure were similarly added. The Na+ occupying the catalytic ion site in the crystal structure was modified to Mg2+. In our previous study on pol μ cognate system [29], we observed that different protonation states of His329 do not affect the geometry of active site or the conformation of key residues significantly. Therefore, in this study, we only modeled His329 in its default protonation state (Nδ).
In each model, the A:dTTP nascent base pair was replaced with a different non-cognate base pair; namely, A:dCTP, A:dATP, A:dGTP, T:dCTP, or T:dGTP (the template base's symbol is written first, followed by the incoming nucleotide's symbol). Purine bases can assume both anti and syn orientations. Because a crystal structure of pol β with a template base in syn conformation has been reported [60], we modeled the template adenine in the A:dATP and A:dGTP systems in both orientations. The protein residues and other DNA base sequences remain unchanged. We also built a cognate T:dATP system to discern similarities of cognate base pairs.
All models were solvated with explicit TIP3 water model in a water box using the VMD program [61]. The smallest image distance between the solute and the faces of the periodic cubic cell was 7 Å. Besides the water molecules in the crystal structure, 13,625 water molecules were added into each model using VMD program. The total number of water molecules is 13,716. To obtain a neutral system at an ionic strength of 150 mM, 46 Na+ and 28 Cl− ions were added to each system. All of the Na+ and Cl− ions were placed at least 8 Å away from both protein and DNA atoms and from each other.
All initial models contained approximately 47,621 atoms, 91 crystallographically resolved water molecules from the ternary complex, 13,625 bulk water molecules, 2 Mg2+ ions, incoming nucleotide dNTP, and 46 Na+ and 28 Cl− counter-ions.
Initial energy minimizations and equilibration simulations were performed using the CHARMM program (version c35b2) [62] with the CHARMM all-atom force field including the cross term energy correction map (CMAP) specification for proteins [63]–[65]. The system was minimized with fixed positions for all heavy atoms of protein or nucleotides, using SD for 10,000 steps followed by ABNR for 20,000 steps. Then the atoms of added residues (His366-Val386 and Ala403-Ala405) and non-cognate nucleotide base-pair were released. Another cycle of minimization was performed for 10,000 steps using SD followed by 20,000 steps of ABNR. The equilibration process was started with a 100 ps simulation at 300 K using single-time step Langevin dynamics, while keeping all the heavy atoms of protein or nucleotides fixed. The SHAKE algorithm [66] was employed to constrain the bonds involving hydrogen atoms. This was followed by unconstrained minimization consisting of 10,000 steps of SD and 20,000 steps of ABNR.
The missing loop construction was performed using the program NAMD [67] with the CHARMM force field. All protein or DNA atoms were fixed, except those from the added residues (His366-Val386 and Ala403-Ala405) and the non-cognate base-pair in order to relax the added loop, the non-cognate base-pair, and the water around our complexes. Each system was equilibrated for 1 ns at constant pressure and temperature. Pressure was maintained at 1 atm using the Langevin piston method [68] with a piston period of 100 fs, a damping time constant of 50 fs and a piston temperature of 300 K; the temperature was maintained at 300 K using weakly coupled Langevin dynamics of non-hydrogen atoms with a damping coefficient of 10 ps−1. Bonds to all hydrogen atoms were kept rigid using SHAKE, permitting a time step of 2 fs. The system was simulated in periodic boundary conditions with full electrostatics computed using the PME method [69] with grid spacing on the order of 1 Å or less. Short-range non-bonded terms were evaluated at every step using a 12 Å cutoff for van der Waals interactions and a smooth switching function. Molecular dynamics at a constant temperature and volume for 4 ns were followed, using the same constraints as above. The final dimension of each system is approximately 78.95 Å × 74.61 Å × 79.91 Å. The model of the A:dCTP system is shown in Fig. 1(c) as an example.
In prior study, we found that the conformation of the added Loop1 does not affect the behavior of pol μ system significantly [29]. In addition, Loop1 is far away from the active-site region we are interested in. Therefore, we only modeled one conformation of Loop1 for all systems.
Production dynamics were also performed using the NAMD program with the CHARMM force field. In all trajectories, all heavy atoms were free to move. Each simulation was run for 120 ns. Molecular dynamics simulations using the NAMD package were run on the IBM Blue Gene/L at Rensselaer Polytechnic Institute and the Dell computer cluster at New York University.
No substantial protein subdomain or DNA motions were captured during all our non-cognate simulations (Fig. S1). This agrees with our prior suggestion that unlike pol β or pol λ, an open-to-closed transition characterized by large-scale protein or DNA motions may not exist in pol μ [29]. Due to the larger size of dGTP and dATP than that of the cognate dTTP, the template base A5 at the gap pairing with dNTP shifts from its original position significantly (at 95% confidence level, Fig. S1(b) and Fig. S2) in A:dGTP and A:dATP non-cognate systems, to better accommodate the incoming nucleotide. In the A:dCTP system, dCTP is relatively smaller, therefore dCTP can be accommodated without the shift of A5. However, the shift of the single base A5 does not incur wide range movements in DNA backbones of pol μ complexes. This agrees with our prior work that pol μ binds to the DNA more tightly than pol λ [29].
Active sites in the non-cognate systems are significantly distorted compared to those in the cognate systems because the Watson-Crick base-pair between the incoming nucleotide and its corresponding template base no longer exist (Fig. 2). New hydrogen bonds between those two bases form (directly, or through a water molecule in T:dCTP system). However, these new hydrogen bonds are less stable than those in the Watson-Crick base pair. In addition, the steric hindrance between the two large purine bases in purine:purine non-cognate systems like A:dATP and A:dGTP further destabilize their interactions. Thus, the nucleotide fluctuates substantially within the active site, indicating a lower active-site conformational stability. In the A:dATP and A(syn):dATP systems, the non-cognate dATP interacts with both A5 and A6 in the template, without breaking the hydrogen bonds between A6 and the primer terminus T17. Thus, dATP stacks between A5 and A6 during the simulation. A similar nucleotide-stacking was also observed in pol λ's A(syn):dATP system. However, in pol λ, a positively-charged residue (Lys273) near A5 attracts A5 further away from dATP and stabilizes the DNA backbone, thereby shifting the DNA backbone [47]. In contrast, pol μ's negatively-charged Glu173 at the corresponding position “pushes” A5 back and keeps the DNA backbone near to its original position. As a result, the following shift of DNA backbone in pol λ's A(syn):dATP system does not occur in pol μ's A:dATP or A(syn):dATP systems.
The geometry of the active-site conformation in each system is shown in Fig. 3, and the critical distances in the active site are summarized in Table 1. The cognate A:dTTP and T:dATP systems share a similar active-site conformation: two water molecules coordinate with the catalytic Mg2+ ion (A). Mg2+ (A) connects with the primer terminus through two water molecules, and connects with the incoming nucleotide both directly and through a water molecule. Thus, the active site is relatively tight and appears ready for the chemical reaction. In the A:dGTP, A(syn):dATP, and T:dGTP non-cognate systems, few rearrangements in the active-site geometry occur. Mg2+ (A) connects to both the incoming nucleotide and the primer terminus through two water molecules, and the catalytic aspartate residues remain in their active conformation. The T:dCTP system has a similar active-site geometry in the beginning of simulation, but after 75 ns, O1A in the dCTP shifts away from the nucleotide-binding Mg2+ ion (B). After the shift, Mg2+ (B) coordinates with O in Asp330. Other coordination interactions in the T:dCTP system remains the same as the cognate system.
In a prior study of cognate pol μ systems, we found that His329 is the most sensitive residue to the absence or presence of incoming nucleotide. Its conformational change triggers the flip of the catalytic aspartate residue Asp330, thus contributing to the assembly of the active site [29]. In the A:dCTP system, His329 flips to an alternative conformation within 10 ns [Fig. S3(a)]. In the new conformation, His329 does not fully “open” to the inactive conformation, though still interrupts binding with dCTP. His329 further flips to its inactive “open” conformation but then flips back to the alternative conformation. Following the flip of His329, Asp330 rotates to an alternative conformation, where both OD1 and OD2 on Asp330 coordinate with Mg2+ (A). As a result, Mg2+ (A) coordinates with only one water molecule instead of two, and its connection with the primer terminus through water molecules weakens. Interestingly, the distance between and O1A on dCTP is significantly smaller than that in cognate A:dTTP system. However, O1A in the dCTP shifts away from Mg2+ (B), and Mg2+ (B) coordinates with O in Asp330, just as in the T:dCTP system.
In the A:dATP system, His329 also flips, but this is only followed by a slight rotation of Asp330 [at an 80% significance level, Fig. S3(b)]. The coordination around Mg2+ (B) remains the same. Asp420 rotates toward Mg2+ (A), and Mg2+ (A) coordinates with both OD1 and OD2 on Asp420. Due to attraction by Asp420, Mg2+ (A) shifts away from dATP, no longer able to directly coordinate with O1A on dATP, though it still interacts with O1A through a water molecule. Though Mg2+ (A) directly coordinates with the primer terminus T17, the distance between Mg2+ (A) and O3′ in T17 is significantly larger than that in the cognate system. In fact, the distance between O3′ in T17 and Pα in dATP is more than 8 Å (compared to ∼5 Å in the cognate system, Fig. S4), significantly larger than the optimal distance for chemical reaction. Again, distortion in the active site in the A:dATP system can be correlated to inactivity.
The A(syn):dGTP system also has a significantly larger O3′ - Pα distance. Like in A:dATP system, Mg2+ (A) also deviates from O1A in the incoming nucleotide, interacting with it only through a water molecule. Three water molecules coordinate with Mg2+ (A) instead of two in the cognate system. Because the third water molecule coordinate with neither the primer terminus nor the incoming nucleotide, interactions within the active site weaken overall. The three aspartate residues and His329 all remain in their active conformation.
In the A:dCTP, A:dATP, and A(syn):dGTP systems, water-mediated hydrogen bonds are generally weaker than direct hydrogen bonds in cognate systems. Therefore, active sites in those non-cognate systems have weaker internal interactions and thus may be more likely to deform.
We observe that even in the cognate system, the crucial O3′ - Pα distance (∼5 Å) appears to be longer than that is required for the chemical reaction (∼3 Å) [70], and also longer than the O3′ - Pα distance in the crystal structure (∼4 Å). Similar observations have been noted and discussed for various pol X family members [33], [36]. Such deviations likely occur because of the imperfection of force fields. For example, the energetics of divalent ions like Mg2+ are considered in the van der Waals (described by the phenomenological Lennard-Jones potential) and Coulombic interactions. Thus, while data generated for divalent ions with these force fields are generally useful and informative, ligand/ion distances may differ from those observed in high resolution x-ray crystal structures. Nonetheless, because our study focuses on general trends in Mg2+ ion coordination and involves systematic comparisons of the trends among closely-related systems, the above limitations are acceptable. Recent crystallographic studies also reveal that the O3′ - Pα distance may be much longer than the expected value [51].
In all the cognate and non-cognate systems we studied, the sugar puckers at the upstream primer and the dNTP remain in the C2′-endo state during our simulations.
Further rearrangements occur in the non-cognate systems that increase active-site disorder. A summary of residue rearrangements involved in each non-cognate system are provided in Table 2 and Fig. 4. In the cognate system of pol μ, a cognate incoming nucleotide triggers the rotation of Gln440 and Glu443, and this “loosens” the nucleotide-binding pocket and helps accommodate the incoming nucleotide. When an non-cognate nucleotide is present, Glu443 flips to an inactive state, thus “tightening” the nucleotide-binding pocket and deactivating the active site. Following the flip, Glu443 may interact with the non-cognate nucleotide through water molecules in several non-cognate systems, though the water-bridged interactions are dynamic. Interestingly, after the flip, the distance between Glu443 and the nucleotide does not decrease (Fig. S5). Thus, the deactivation effect of Glu443's flip may be due to an electrostatic effect rather than the steric hindrance. We discuss this further in the next section. We hypothesize that a mutation of Glu443 to a residue with similar length but neutral charge (for example, methionine) may reduce the fidelity of pol μ. Such an E443M substitution may reduce the catalytic ability for both cognate and non-cognate systems, but the cognate system may be affected more. Thus, the fidelity of the E443M mutant may decrease. This hypothesis may be tested by further experimental and computational studies.
The motion of Gln440 is more flexible. Without dNTP, Gln440 flips to its inactive form and binds to the primer terminus. In the non-cognate systems, Gln440 cannot bind to the primer terminus because of the hindrance, and therefore it cannot reach a stable inactive nor active state. When we plot the distance between the center of mass of Gln440 and the center of mass of dNTP in Fig. S6(a), we see that in the A(syn):dGTP and T:dGTP systems, Gln440 is significantly closer to the incoming nucleotide than in the cognate systems (computed with the data from the last 40 ns, at a confidence level of 90% and 85%, respectively). In the A:dCTP and A:dGTP systems, Gln440 also displays a tendency to shift towards dNTP. The average distance between Gln440 and dNTP decreases for 0.44 Å and 0.62 Å over the simulation in A:dCTP and A:dGTP systems, respectively. In contrast, overall change is only 0.02 Å in the cognate A:dTTP system. Therefore, Gln440 also participates in “tightening” the active site by shifting towards the non-cognate nucleotide. These two residues are not conserved in pol β or pol λ, and thus must be unique to pol μ function.
As its corresponding residue Arg514 in pol λ, Arg444 in pol μ mainly helps stabilize the template base at the gap through stacking interactions [29], [36]. It also stabilizes Gln440 in its active conformation by hydrogen bonding to it. However, unlike Arg514 in pol λ, Arg444 in pol μ does not participate in the conformational change of active site upon binding a cognate nucleotide. Because of the distortion in the mispaired bases, the stacking interactions are interrupted in non-cognate systems. Therefore, Arg444 flips away from the active site in all non-cognate systems except A(syn):dGTP and T:dGTP [Fig. S7(a)]. The flip of Arg444 increases the flexibility of Gln440 because the hydrogen bond between Arg444 and Gln440 breaks. However, the flip of Arg444 does not necessarily cause the shift of Gln440 towards the non-cognate dNTP. In addition, because Arg444 binds to the backbone of the template base A5, its conformational change may also induce the shift of A5 away from the active site.
Trp436 in pol μ is analogous to Phe272 in pol β, or Phe506 in pol λ, residues that initiate DNA or subdomain motions through a flip during the conformational transition of the polymerase complex [33],[36]. When pol μ incorporates a cognate nucleotide, no significant motion of Trp436 is observed because DNA or subdomain motion is not part of pol μ's conformational pathway. However, in the A(syn):dGTP system, Trp436 rotates its indole ring towards dGTP [Fig. S7(b)]. The rotation of Trp436 limits the space in the active site and pushes the dGTP away from the active site, thereby also “tightening” the nucleotide-binding pocket. We observe a similar rotation occasionally in the A:dCTP system toward the end of the simulation.
Arg447 in pol μ is analogous to Arg283 in pol β and Arg517 in pol λ, both important for checking the cognate base-pairing [47], [71]–[73]. Arg517 in pol λ is also crucial in pol λ's ability to accommodate frame-shifted DNA [74]. This arginine binds to both the base at the gap and the one pairing with primer terminus in the template DNA, and stabilizes DNA in the closed form of complex. However, in pol β and pol λ, this binding is sensitive to the incoming nucleotide context. When an non-cognate nucleotide is present and the active site is distorted by abnormal base pairing, fewer direct hydrogen bond interactions between Arg283 (in pol β) and Arg517 (in pol λ) with the DNA occur [41], [47]. This leads to the poor stabilization of the DNA template bases, which incurs further rearrangements of incoming nucleotides and/or shift of DNA backbone. In contrast, in pol μ, the binding of Arg447 to DNA is not affected significantly by the non-cognate nucleotides. The direct hydrogen bonding of Arg447 - A6:N3 and Arg447 - A6:O4′, as well as Arg447 - A6:O1P interaction through a water molecule, are present in all non-cognate systems. Arg447 - A5:N3 or Arg447 - T5:O2 interaction is also present in all systems except A(syn):dATP and A(syn):dGTP, where the syn conformation of A5 keeps N3 away from Arg447. In T:dGTP system, Arg447 - T5:O2 interaction is not stable. The hydrogen bond between them does not form until after 90 ns, and deforms near the end of our 120 ns simulation [Fig. S6(b)]. However, in the A(syn):dGTP and T:dGTP systems, Arg444 does not flip and stacks with A5 or T5. Thus, the mispaired bases are still stable, and further rearrangements or motions caused by the lack of Arg447 interactions are not observed. This may suggest that pol μ's active site is more flexible than those in pol β and pol λ, so it might better accommodate the non-cognate nucleotide without breaking Arg447/DNA interactions. The flexibility of active site also supports the observation that pol μ can accommodate and insert ribonucleotides in the active site [75].
Motions of “Gate-keeper” residues, namely the flip of Glu443, shifting of Gln440, flip of Arg444, and rotation of Trp436, are not observed in our modeled cognate T:dATP system. This further confirms that “gate-keeper” residues can help discriminate non-cognate nucleotides from cognate ones and thus may have a significant role in controlling the fidelity of pol μ.
Three of the four “gate-keeper” residues in pol μ (Gln440, Glu443, and Arg444) are located apart from the upstream primer (>8 Å) and near the downstream primer; Trp436, which is near the upstream primer, functions as “gate-keeper” residue in only one non-cognate system (A:dGTP). In comparison to other X-family members, both “gate-keeper” residues in pol β (Arg258 and Phe272), as well as two of three “gate-keeper” residues in pol λ (Tyr505 and Phe506), are located near the upstream primer (<6.5 Å, Fig. 5). This difference may be related to the fact that pol μ can incorporate and insert the incoming nucleotide when the upstream primer is not paired. That is, pol μ may be less sensitive to changes around the upstream primer. Residue flexibility differences when the upstream primer is not paired may be interesting to explore in future computational and experimental studies of pol μ.
Using the number of dNTP and protein residue changes in Table 1 and Table 2, we suggest the following sequence for difficulty of nucleotide incorporation by pol μ: T:dGTP<A(syn):dATP<T:dCTP<A:dGTP<A(syn):dGTP<A:dCTP<A:dATP (T:dGTP is the easiest to incorporate and A:dATP is the most difficult). This sequence agrees with the observed trends in the reaction kinetics data for nucleotide insertion [17], [18], as summarized in Table 3, with the exception of A:dGTP, which may depend sensitively on the surrounding sequence [76]. For example, kinetic data obtained with another DNA sequence [19] suggests a different trend: T:dCTP<A:dCTP<T:dGTP<A:dATP<A:dGTP. Another possible explanation for our observing greater difficulty in the A:dGTP mispair relative to T:dCTP and T:dGTP while kinetic data indicate that A:dGTP is more favorable is that T:dCTP and T:dGTP mispairs are less favorable overall for chemical reaction following the conformational changes, as discussed below.
We further examine in Fig. 6 the electrostatic potential of pol μ's active site with the cognate A:dTTP and T:dATP base pairs and various non-cognate systems. Unfavorable protein/dNTP interactions emerge in the non-cognate systems that destabilize the dNTP. Though subtle differences exist, the active site in pol μ's cognate A:dTTP and T:dATP systems have mainly negative (red) electrostatic potential, whereas the non-cognate systems have more neutral (white) or positive (blue) electrostatic potentials. Interestingly, for pol λ, changes in electrostatic potential are also observed, but in an opposite way: more neutral or positive for the cognate system, and more negative for the non-cognate system [47]. The changes in electrostatics environments suggest altered interactions within the active site, which in turn affect active-site rearrangements.
In the A:dCTP system, Arg447 (green circle in Fig. 6B) appears in a negatively charged region, thus its stabilization effect to DNA template base A5 and A6 weakens, which in turn may destabilize the dCTP and primer terminus pairing with A5 and A6. In A:dATP and A(syn):dATP system, following the flip of Glu443, the region near N1 and N3 atoms of dATP becomes more negative (pink circle in Fig. 6C and 6D), while the region near the amino group of dATP is relatively more positive (cyan circle in Fig. 6C and 6D). These two disruptive forces together destabilize dATP and drive it towards the primer terminus direction, allowing dATP to stack between A5 and A6. Moreover, in the A:dCTP, T:dCTP, and T:dGTP systems, the end of the phosphate group on dNTP falls into a mainly positive region (cyan circle in Fig. 6B, 6H, and 6I, compared to pink circle in the cognate systems, Fig. 6A and 6G), which is unfavorable for the proton transfer reaction to follow. Therefore, not only does the altered electrostatic potential around dNTP disturb the conformational rearrangements in active site, but it also may it affect the chemical step after the conformational changes.
In prior work we studied the coupled conformational changes within polymerase complex upon binding a cognate or non-cognate incoming nucleotide for pol λ, pol β, and pol X [52]. Similar coupled motions within the same subdomain, among different subdomains, and between protein and DNA/dNTP, were revealed across the X family. Even within the same subdomain, the coupled regions can be distant in space. These correlated motions together drive the polymerase towards its active form. When an non-cognate nucleotide is bound, such correlated motions decrease.
Correlated motions in pol μ system are shown in Fig. 7. The cognate system of pol μ displays a similar network of coupled motions as pol β and pol λ, but with fewer interactions. Specifically, within the palm subdomain, correlated motions are observed among three regions as follows (region A in Fig. 7): a conserved X-family loop [77] (Thr314-Thr336) containing two of the three catalytic aspartates (Asp330 and Asp332) with a region (Thr288-Val290) near the finger including Pro289 (its analogous residue Arg149 in pol β or Arg346 in pol λ binds with the incoming nucleotide, though Pro289 in pol μ does not has such binding ability); the Asp loop with the Loop 2 (Ala407-Lys417) that is apart from the active site; and the Pro289 region with the Loop 2. Gly435-Arg444 in the thumb that includes the nucleotide-binding pocket residues Gly435, Trp436, Gln440, Glu443, and Arg444, also correlate with the Asp loop and Pro289 region in the palm (region B). All these regions in the palm and thumb are further correlated to the dTTP (region C).
Non-cognate systems of pol μ generally have less correlated motions than the cognate system. In all the three non-cognate systems, the correlated motions between Gly435-Arg444 and the dTTP, and those between Gly435-Arg444 and Pro289 region are greatly reduced or missing. The A:dGTP system has the least changes of coupled motions, and is most similar to the cognate A:dTTP system. In the A:dCTP system, more intense motions correlated within the fingers are observed. The correlated motions between the polymerase fingers and 8-kDa domain, and between the fingers and DNA also increase. These motions may suggest that pol μ requires more conformational rearrangements in the finger when accommodating dCTP. With these additional conformational changes, pol μ deviates from its active conformation. The A:dATP non-cognate system is the most different of the three, compared to the cognate system. Correlated motions between the Asp loop and Gly435-Arg444 are reduced significantly. Because both the Asp loop and Gly435-Arg444 are within the active site, the reduced interactions among active-site residues largely hamper the orchestration of cooperative events to reach at an optimal active-site conformation. Almost all other correlated motions in the A:dATP system also appear reduced. The limited correlated motions suggest that A:dATP system remains in an inactive.
Overall, our correlated motion analysis suggests an order of A:dTTP≈A:dGTP>A:dCTP>A:dATP for the degree of correlated motions. This order also agrees with the trend we suggested above from active-site distortion and key residue motion. We further present the difference matrices between the A:dTTP cognate system and the A:dATP/A:dCTP/A:dGTP non-cognate systems in Fig. S8. Within the three systems, the correlated motion of A:dGTP system is most similar to that of the cognate A:dTTP system, suggesting a more favorable active site for the following chemical step. This may explain the high misincorporation rate of A:dGTP observed in experiments (Table 3). These results provide further support for the hybrid conformational selection/induced-fit model for polymerases: before substrate binding, the polymerase/DNA complex adapts a series of possible conformations, and substrate binding stabilizes specific conformation. This inherent flexibility is evident from Fig. S9, which reveals the correlated motions when the substrate (dNTP) is absent. From this ensemble of conformations, in the A:dTTP cognate system, dTTP would selectively bind to a near-active conformation and guide the system into a fully active form as well as trigger required active-site changes. In the relatively active A:dGTP system, dGTP would also selectively bind to a near-active conformation with correlated motions similar to those in A:dTTP system. However, the suboptimal fit of dGTP within the active site induces active-site changes that differ from that in A:dTTP system. In the A:dCTP and A:dATP systems, the dNTPs bind to variable conformations of pol μ that deviate from the active forms; those tailored fits, however, hamper correlated motions that are essential for preparing the enzyme for subsequent catalysis.
Our molecular dynamics simulations of pol μ cognate and non-cognate systems reveal significant differences in the active site and regarding the correlated motions upon binding an non-cognate nucleotide compared to a cognate substrate. The results suggest that, compared to pol β or pol λ, no significant changes in global motion of protein or DNA would occur for pol μ. His329 and Asp330 in the active site, as well as Trp436, Gln440, Glu443, and Arg444 in the nucleotide-binding pocket, play the role of “gate-keeper” in pol μ. These residues alter the electrostatic potential in the active site and trigger the distortion of active site when an non-cognate nucleotide is bound. Because most “gate-keeper” residues in pol μ are relatively far from the upstream primer, this fact may explain in part pol μ's ability to incorporate nucleotides when the upstream primer is not paired. Furthermore, in non-cognate systems, correlated motions within the complex are reduced. These results suggest that like other X-family polymerase, pol μ also fits in a hybrid conformational select/induced-fit model; the cognate substrate would bind to the active form and trigger active-site changes, while non-cognate substrates with relative high efficiency would bind to an active form but not trigger the following active-site changes, and non-cognate substrates with poor efficiency would bind to variable conformations. The degree of active-site geometry distortion determined from our simulations roughly parallels the kinetic data, suggesting a direct relation between active-site structural distortions and fidelity of pol μ. We also suggest experimentally testable predictions that mutation on pol μ's “gate-keeper” residues, like E443M, may reduce the fidelity of pol μ.
|
10.1371/journal.pcbi.1002762 | Phenomenological Model for Predicting the Catabolic Potential of an Arbitrary Nutrient | The ability of microbial species to consume compounds found in the environment to generate commercially-valuable products has long been exploited by humanity. The untapped, staggering diversity of microbial organisms offers a wealth of potential resources for tackling medical, environmental, and energy challenges. Understanding microbial metabolism will be crucial to many of these potential applications. Thermodynamically-feasible metabolic reconstructions can be used, under some conditions, to predict the growth rate of certain microbes using constraint-based methods. While these reconstructions are powerful, they are still cumbersome to build and, because of the complexity of metabolic networks, it is hard for researchers to gain from these reconstructions an understanding of why a certain nutrient yields a given growth rate for a given microbe. Here, we present a simple model of biomass production that accurately reproduces the predictions of thermodynamically-feasible metabolic reconstructions. Our model makes use of only: i) a nutrient's structure and function, ii) the presence of a small number of enzymes in the organism, and iii) the carbon flow in pathways that catabolize nutrients. When applied to test organisms, our model allows us to predict whether a nutrient can be a carbon source with an accuracy of about 90% with respect to in silico experiments. In addition, our model provides excellent predictions of whether a medium will produce more or less growth than another () and good predictions of the actual value of the in silico biomass production.
| The ability of microbial species to consume compounds found in the environment to generate commercially-valuable products has long been exploited by humanity. The vast untapped diversity of microbial species offers a wealth of potential resources. However, little is known about most microbial species. While the metabolic network of an organism can be studied to find its nutritional requirements, we lack a reliable metabolic reconstruction for most species. We use in silico organisms to systematically explore whether an arbitrary nutrient can stimulate growth as a single source of carbon, and how effectively it can be used by the organism. We find that we can predict whether a nutrient is a source of carbon and the biomass yield of that nutrient with a simple model that transcends the diversity of species and their environments. Our model for catabolic potential can therefore be used as a baseline model for any microbe for which we lack a metabolic reconstruction.
| Predicting microbial metabolism under a broad range of conditions would enable us to leverage microbes for applications in critical areas such as energy production [1], pollution amelioration [2], [3], bioengineering [4], physiology or medicine [5], [6] to name a few. While systematic in vivo growth experiments could in principle fill the gaps in our current knowledge, those experiments are time consuming and contingent on the ability to grow the microbial species of interest in the laboratory [7]. As a consequence, only a small number of microbes have been studied using these techniques. For example, Biolog (http://www.biolog.com) provides Phenotype Microarrays that have been used on species such as Escherichia coli [8] and Bacillus subtilis [9] but to date there are only 200 publications listed in Biolog.
To circumvent experimental limitations, a number of mathematical models have been developed aiming to predict microbial growth rates [10]–[14]. However, these models are only valid for a limited number of specific nutrients and are not easily generalizable because of the need to determine many parameters empirically. Indeed, developing such a theory seems an insurmountable challenge given the combinatorial number of possible growth media and the large number of unknown parameters such as reaction constants and enzyme affinities that control metabolic reactions [15]–[17]. For instance, the in silico reconstruction of E. coli contains more than 2,000 reactions [8]; taking into account that each reaction has at least two and up to tens of kinetic parameters [18], a detailed kinetic model would have on the order of 10,000 parameters, unless some approximations valid under certain conditions are made to reduce this number [19].
An alternative approach that is gaining in popularity is the development of thermodynamically-feasible metabolic reconstructions that can be used to predict the growth of individual organisms using constraint-based methods [8], [9], [20]–[27]. Researchers fine-tune these reconstructions to match the conditions of specific growth rate experiments, such as nutrient availability and ATP maintenance. These reconstructions, built by literature mining, can, under certain conditions, accurately predict the impact of individual nutrients on growth as sources of carbon, nitrogen, phosphorus and sulfur, thus allowing researchers to evaluate the demands of biomass production and to investigate how individual nutrients can meet that demand. While building metabolic reconstructions for new organisms is quite challenging, nowadays the process is becoming more and more automated [28]. Nonetheless, building a reconstruction still requires manual and/or experimental tuning, which hinders the generalization of these models. An additional caveat of these metabolic reconstructions is that the complexity of metabolic networks prevents researchers from obtaining an understanding of why a certain nutrient yields a given growth rate for a given microbe.
In order to formalize the intuition used to build and fine-tune the constraints used to predict growth rates using metabolic reconstructions, we present here a systems-level phenomenological theory of microbial metabolism, that is, a theory that yields a mathematical relationship between the maximal biomass production of a microbe and the set of available nutrients acting as carbon sources, without taking into account any microscopic details of the processes occurring inside the cell. The biomass production predictions of our model depend exclusively on the characteristics of the available carbon sources and the set of metabolic pathways that can catabolize them. Our model is able to match the predictions of flux balance analysis with no significant computational effort and providing insight into the determinants of catabolic efficiency.
Our phenomenological model expresses the impact that different carbon sources (or nutrients) have on a microbe's ability to grow using only information on the chemical structure of these carbon sources and on the ability of a microbe to catabolize these nutrients. Our model is built to reproduce the predictions of flux balance analysis calculations (FBA) on metabolic reconstructions, thus it will suffer from the same limitations. Indeed, while FBA is a powerful tool to investigate microbial metabolism and microbial growth in particular, it has a number of limitations when estimating growth rates and the effect of media and environmental conditions on growth.
It is well-known that microbes need a minimal medium and a carbon source in order to grow. Minimal media have been described for a number of species and contain essential chemical species without which the species would not be able to grow [20], [29]. The growth rate of a microbe, however, depends on many other factors including the uptake rate of nutrients, temperature, regulation, the availability of oxygen, etc.
FBA is a linear optimization method that predicts the maximal conversion of a set of carbon sources into biomass with a fixed minimal medium. In order for FBA conversion rates to reproduce empirical growth rates, one needs to consider an additional ATP maintenance flux which is obtained by fitting FBA results to empirical growth rates obtained for a certain temperature, minimal medium, and carbon source uptake. While ATP maintenance rates obtained for a specific minimal medium have been shown to give accurate predictions of growth rates in different minimal media for some organisms [8], in principle one cannot assume that they are valid for predicting growth rates for arbitrary minimal media.
Additionally, because metabolic reconstructions do not consider regulatory constraints, FBA will predict that a microbe is capable of uptaking two different sugars simultaneously, while it is well-known that if there is more than one sugar carbon source at high enough concentrations, the organism will exhaust the preferred one before consuming the others [30]. A microbe will, however, consume multiple sources of carbon other than sugars simultaneously and as a consequence grow faster [31]. It has been shown that steric constraints can already reproduce diauxic growth in some organisms in the presence of multiple sugars [32], however, there is no general framework that is able to deal with this issue when using FBA.
To develop a phenomenological model that is able to reproduce maximal biomass conversion rates per carbon source under aerobic conditions, we investigate the maximum amount of biomass that can be produced by an organism in the presence of a minimal medium, oxygen and one or more carbon sources including at most one sugar. To this end, we run FBA on metabolic reconstructions in which we remove ATP maintenance [27] (see Methods for specific details). We concentrate on a training set of four species for which there are high-quality metabolic reconstructions available, and which cover a wide range of microbial phylogeny: a gram-negative bacterium (E. coli [8]), a gram-positive bacterium (B.subtilis [9]), an eukaryote (Saccharomyces cerevisiae [21]), and an archaeon (Methanosarcina barkeri [24]). We validate our model on a test set of three species for which we also have high-quality metabolic reconstructions—Helicobacter pylori, a gram-negative bacterium [20], Staphylococcus aureus, a gram-positive bacterium [22], and Mycobacterium tuberculosis, an acid-fast gram-positive bacterium [23].
The rationale for choosing a small number of species for model building and validation is that lower quality reconstructions are likely to have significant gaps that could incorrectly bias the determination of the model. Additionally, the aggregate set of nutrients available for these reconstructions is of 352 nutrients, which cover all nutrient types and 90 out of 97 pathways available in KEGG [33]–[35], thus ensuring that our model is comprehensive.
We believe that the development of a systems-level model of microbial metabolism is not only complementary to current approaches but offers some advantages with respect to them. Specifically, while FBA run on metabolic reconstructions already has the capability of predicting maximum biomass yield, our model has the advantage that since it does not consider microscopic details of the metabolism of a species it is directly applicable to any other organism growing under the conditions of our analysis. In fact, we show that solving over a thousand linear equations under constraints can be well approximated by a simpler model whose principles are easier to understand. As such, our model offers the possibility of uncovering universal features of the metabolism of organisms that other computational approaches are not capable of. The mathematical model we develop is thus a valuable tool from both fundamental and applied perspectives, since it can help understand the metabolism of organisms for which a metabolic reconstruction is not available, or guide the process of validating metabolic reconstructions.
In order to develop a mathematical model that relates nutrient (carbon source) uptake in complex media to biomass production, we need to address three different questions (see Fig. 1). The first question we need to answer is whether a nutrient can be a source of carbon or not. This is to say, first we need to develop a mathematical model that determines whether a nutrient can produce growth or not based on the information available for that nutrient and organism. Note that because we are interested in a binary output (growth/no growth), our model will depend exclusively on the composition of the nutrient and the pathways that can catabolize this nutrient.
Then, we need to assess what is the maximum amount of biomass a nutrient can produce when acting as the sole source of carbon. Our specific aim will be to find a mathematical relationship between nutrient composition (in our case, carbon content) and the biomass produced per unit of nutrient uptaken.
In third place, we need to assess how maximal biomass production is affected when several nutrients are present in a medium. That is to say, we need to determine the relationship between biomass produced and the characteristics of the nutrients available in the medium.
Finally, to prompt and aid experimental studies, we use our model to predict which nutrients can be a source of carbon for four species lacking a metabolic reconstruction, and predict the biomass production of these species in complex media. We show that the carbons available in the 20 natural amino acids in a medium provide the best boost to biomass production regardless of the species.
The first step is to build a model that predicts whether an individual nutrient can be used as a source of carbon by species (see Fig. 1). If it can, we say that belongs to the group of nutrients that contribute to growth in ; otherwise we say that belongs to the group of nutrients that do not contribute to growth in species . We use flux balance analysis on the species in the training set to empirically determine which nutrients belong to and which ones to (see Materials and Methods). We then use these data to build the model. In this section we describe the model (which we summarize in Fig. 2) and validate it with the species in the test set.
The next step is to determine for those nutrients that produce growth what is the maximal production of biomass when that nutrient is the only source of carbon. In fact, if a nutrient is the carbon-limiting source in a given medium, the biomass yield of the nutrient must be related to the number of carbons in the nutrient. In Fig. 6, we display the in silico biomass production and the biomass yield of all nutrients as a function of the number of carbon atoms they contain, for the species in the training and test sets (we do not consider M. barkeri in our model for biomass yield because M. barkeri can only grow anaerobically, and this has a significant effect on the energy used to polymerize the proteins and nucleic acids in the biomass).
It is visually apparent from Fig. 6 that there is a strong correlation between the number of carbons in a nutrient and the biomass production. We thus model the biomass production induced by nutrient as(3)where is the (nutrient-independent) average biomass yield of the nutrients in The data suggests the existence of an upper bound for the biomass yield () as a function of the number of carbons (blue line in Figs. 6, 7, and 8). The nutrients frequently found at or close to this upper bound are sugars, alcohols, and other compounds with hydroxyl groups. Many of the hydroxyl groups on these compounds are typically oxidized in order to reduce NAD to NADH, which is an important source of energy, and will in turn increase the biomass yield. For simplicity, we obtain as the average biomass yield for the sugars uptaken by an organism. The average nutrient has of the efficiency of these high-efficiency sugars, but nutrients vary wildly in their yield (Fig. 7). We consider one main factor that contributes to this variation in biomass yield: the number of carbons in a nutrient that are effectively catabolized.
Finally, we want to model the maximal biomass production when there is more than a single source of carbon present in the medium, or in other words when organisms grow in a complex medium. We consider complex media in which nutrients are restricted to five classes: sugars, fatty acids, amino acids, purines, and pyrimidines, partly because they are commonly used in growth rate/biomass yield experiments, and partly because it simplifies the analysis of the results. In addition, complex purines and pyrimidines are a mixture of sugars and nucleobases, and as we are already considering sugars, we only use the simple nucleobases.
The simplest plausible model for the contribution of nutrients to the biomass production is one in which each nutrient has an independent contribution to the biomass production. For each species , we calculate the biomass production on a complex medium containing nutrients using(5)We estimate for a nutrient from class using the average yield for that class in the organisms in the training set(6)where is the number of nutrients of class that are taken up by species , and represents the average over species in the training set. For tryptophan and for purines, we use the effective number of carbons when calculating yield, as described previously.
To test this model, we randomly generate ensembles of complex media as described in the Data and Methods section. For each medium and for each species, we calculate ; we use FBA to find the actual in silico biomass production of the organism on the complex media . In Fig. 9 we show how compares to for each species, and for each of the complex media we generate. In these comparisons we also include predictions for the species in the test set (with the exception of S. aureus, for whose reconstruction it is not clear what units were used for the biomass production).
We find a very strong correlation ( using the Spearman rank correlation coefficient) between the predictions of our model and in silico experimental results for training set species (E. coli, B. subtilis, and S. cerevisiae) and test set species (H. pylori and M. tuberculosis). This indicates that the model accurately captures which media will result in faster/slower production of biomass. For each species, however, the model systematically under or over-predicts growth. A regression of the predicted biomass production versus experimental in silico production indicates that , with: for E. coli, for B. subtilis, for S. cerevisiae, for H. pylori, and for M. tuberculosis.
The parameters used to make the predictions for the individual nutrients in the linear model were trained on three species. The linear model consistently under-predicted the in silico biomass production for these three species, and more so for media containing more nutrients. This is a strong indicator that the nutrients that are uptaken are used synergistically by the in silico organism to produce more biomass than expected, showcasing the effect of catabolic pathways being highly connected in the metabolic network. A more complete model for predicting biomass production in complex media will therefore need to take into account synergistic interactions among catabolic pathways.
Our model sheds light on several questions related to the impact of nutrients on the biomass production of microbes. Our approach treats microbial metabolism as a “black box” that uses nutrients to reach optimal biomass production. Because the model does not take species-specific details into consideration, it is useful for generating predictions for any microbe, something that would be impossible with any existing modeling approach. To illustrate how one would proceed to extrapolate to “new” organisms and to show what kind of insights one could obtain, we generate predictions for four organisms that lack a metabolic reconstruction: R. palustris (a gram-negative bacterium), L. monocytogenes (a gram-positive bacterium), D. discoideum (an eukaryote), and T. acidophilum (an archaeon).
We find that, overall, these species are predicted to take up fewer nutrients than the species in the test set (Fig. 10A). This is a consequence of the limited annotation that the authors of TransportDB could use for the predicted protein transporters. One example of this limitation is that for all four species, there were many transporters predicted to take up amino acids, but there was no indication of which amino acids the transporters were specific for. Therefore, for the sake of prediction, we consider that all four species uptake all twenty natural amino acids.
We then use the model of catabolic potential to predict whether each of these nutrients could be a source of carbon. In Fig. 10B we show the number of nutrients that belong to one of the nutrient classes previously described and whether these nutrients are or . For the fatty acids, none of which were predicted to be uptaken by any of the four species, we examined the enzymes available and found that only D. discoideum contains the enzymes for -oxidation, and could therefore catabolize fatty acids.
Finally, we combine our models of biomass yield of nutrients, and of biomass production on complex media (Fig. 11). We choose four complex media which contain a different number of the same nutrients we used for the randomly-generated complex media, namely: sugars, fatty acids, bases, and amino acids. The four media contain: 1) glucose; 2) glucose and hexanoic acid; 3) glucose, hexanoic acid, adenine, guanosine, cytosine, and thymine; 4) glucose, hexanoic acid, adenine, guanosine, cytosine, and thymine and the twenty natural amino acids. We estimate the biomass production in the same manner that we described earlier for the randomly-generated complex media. We find that the biggest influence on the biomass production is the number of carbons available in the nutrients present in the medium; because we are now adding 20 amino acids to medium 4, the biomass production increases almost -fold, on average, in that case.
Note that these predictions for biomass production are based on the biomass yield per carbon available in the nutrients present in the medium. Biomass yield is a time independent quantity that cannot be directly associated to growth rate. However, because biomass yield gives an upper limit for growth rate [48], our model can be used as a baseline for researchers to explore and model growth rates.
Microbes use nutrients found in their environments to grow. Understanding and developing quantitative models of this process is of fundamental importance in cell biology, physiology, medicine, evolution, synthetic biology and bioengineering, not to mention of practical importance to those that need to grow microbes in laboratories. Given the diversity of the microbial world and the number of combinations of nutrients available in their environments, this seems too difficult a problem. In this study, we focused on how microbes might catabolize nutrients to obtain carbon for biomass production.
Our model comprises three levels, with each level building up on the results from the previous one. The first level concerns whether a nutrient will be catabolized. The second level concerns whether all of the carbon in a catabolized nutrient is available for biomass production. The final level incorporates the biomass yield for selected classes of nutrients and enables us to make a prediction on the biomass production of a microbe in a complex medium. To validate our approach, we compare the predictions of the complete model with in silico predictions of growth in complex media of species that are not part of the training set. Our results on these species are excellent predictions of which media will produce more/less growth.
Finally, we looked at the biomass yield of sugars, amino acids, purines, pyrimidines, and fatty acids. We found little variation in the yield of these nutrients amongst different species, and were able to postulate a model for the biomass production of a microbe on a complex medium containing any number of these important nutrients. All of these nutrients have separate catabolic pathways, with the exception of some groups of amino acids which share a catabolic pathway. The fact that the in silico microbial biomass production was more than our model predicts indicates that each of these catabolic pathways can work together synergistically to improve the biomass production of the microbe.
The ability of our model to predict biomass production on complex media must be balanced by an understanding of its limitations. First, we model microbial metabolism as a black box, and therefore we do not account for absent pathways that biosynthesize some biomass components. The absence of such pathways would require that the corresponding biomass components be made available in the medium used, but our model cannot be used to predict which of these are indeed required. This means we cannot use the model in its current form to predict the minimal medium that is needed for a microbe to grow.
Second, microbial species are sometimes identified by the nutrients they take up and excrete. The ability of a microbe to transport such compounds in and out of the cell is largely dependent on the specific protein transporters available. A separate body of work exists, largely in the form of TransportDB [39], and TCDB [49], [50], which will enable the researchers to predict the proteins that transport specific nutrients. We incorporate such predictions of nutrient transport into our model to make predictions for specific species that lack a metabolic reconstruction. Importantly, while knowledge of transporters can enable us to predict which nutrients in the medium can be taken up, we are currently unable to predict which nutrients are excreted as that would in principle require knowledge of the full metabolic network.
Third, the stoichiometric method we describe for generating the biomass data cannot be used to predict the growth rate of a microbe because kinetic information is not included. However, our model provides a baseline to which one can add kinetic information. For example, the rate-limiting step in poor growth media is likely to be the rate at which a microbe takes up nutrients. Therefore one can use the kinetics of nutrient transport with our prediction of biomass yield to predict growth rate.
All this notwithstanding, we believe that our approach and models open the door to significant advances in the quantitative modeling of microbial metabolism, and eventually of the metabolism of more complex organisms. In particular, our models could be extended to consider whether a nutrient acts, not only as a source of carbon, but also as a source of nitrogen or energy, or directly as a component of the biomass. Our models could also be extended to include more detailed information about pathways, or to consider functional metabolic modules [51]–[53] instead of pathways.
Stoichiometric methods have been widely used in metabolic engineering for over 20 years [59], the most used of which is Flux Balance Analysis (FBA) [26], [27], [60]. FBA aims at determining the fluxes through each one of the metabolic reactions in an organism. Thus, FBA relies on the determination of a stoichiometric matrix that represents all the reactions and metabolites in an in silico organism , and a vector of uptake fluxes.
In the matrix , each row corresponds to a reaction, and each column to a metabolite. is the stoichiometric coefficient of metabolite in reaction . The vector of uptake fluxes can have a non-zero entry for every transport reaction that moves a nutrient into the in silico organism.
As a simple example, consider a metabolic network with three reactions:(7)where represents the transport reaction for uptaking Glucose from the environment. The stoichiometric matrix for metabolic network (7) is:(8)Assuming a steady-state concentration of every metabolite and requiring mass-conservation we must impose that(9)where is the set of fluxes through each metabolic reaction. In order to solve Eq. (9), we need to provide the vector . The default flux for each uptake reaction is , meaning that the nutrient is not in the medium or that it cannot be uptaken by the organism. We set if nutrient is present in the medium and could be taken up by the organism.
The system in Eq. (9) has many solutions. In our analysis, the biologically relevant metabolic state is the one that maximizes biomass production [59]. The problem of finding the fluxes through the reactions with a cost function that have to satisfy a number of constraints is a standard linear optimization problem that can be numerically solved using the subroutines provided in GLPK [61].
In order for us to model whether a nutrient can be a source of carbon and the biomass yield of the nutrient, we control the data in two ways. First, for some organisms the uptake of specific nutrients can lead to without being considered significant [8]. Secondly, ATP hydrolysis is integrated into the biomass of the in silico organisms, and is typically trained to better match empirical results based on growth rates. We are not exploring growth rates and we thus adjust the ATP hydrolysis in the biomass of the in silico organisms in order to better support the conclusions we reach. These controls are explained in detail in Text S1.
We tested our model on a large number of complex media generated using the nutrients available for uptake in the in silico organisms. Because the number of possible combinations of these nutrients was too large for us to test each one computationally, we considered an ensemble of 1000 randomly generated complex media . For each complex medium, every nutrient was made available for uptake with the probability , and was excluded with the probability . We generated ensembles for , , and (for a total of 3000 random media).
Sugars present an unusual case because a microbe such as E. coli has been known to exhibit diauxic growth [65]–[67]. This means that microbes regulate sugar uptake so that, despite having various sugars present in the medium, they will only take up one sugar at a time. Our model does not take gene regulation into account, and therefore we manually limited the number of sugars presents in each complex medium to one, specifically glucose.
|
10.1371/journal.pntd.0000183 | T Helper 1–Inducing Adjuvant Protects against Experimental Paracoccidioidomycosis | Immunostimulatory therapy is a promising approach to improving the treatment of systemic fungal infections such as paracoccidioidomycosis (PCM), whose drug therapy is usually prolonged and associated with toxic side effects and relapses. The current study was undertaken to determine if the injection of a T helper (Th) 1–stimulating adjuvant in P. brasiliensis–infected mice could have a beneficial effect on the course of experimental PCM. For this purpose, mice were infected and treated with complete Freund's adjuvant (CFA), a well-established Th1 experimental inductor, or incomplete Freund's adjuvant (IFA - control group) on day 20 postinfection. Four weeks after treatment, the CFA-treated mice presented a mild infection in the lungs characterized by absence of epithelioid cell granulomas and yeast cells, whereas the control mice presented multiple sites of focal epithelioid granulomas with lymphomonocytic halos circumscribing a high number of viable and nonviable yeast cells. In addition, CFA administration induced a 2.4 log reduction (>99%) in the fungal burden when compared to the control group, and led to an improvement of immune response, reversing the immunosuppression observed in the control group. The immunotherapy with Th1-inducing adjuvant, approved to be used in humans, might be a valuable tool in the treatment of PCM and potentially useful to improve the clinical cure rate in humans.
| P. brasiliensis is a thermally dimorphic human pathogenic fungus that causes paracoccidioidomycosis (PCM), the most prevalent human systemic mycosis in Latin America, whose drug therapy is usually prolonged and associated with toxic side effects and relapses. Although immunostimulatory therapy is a promising approach to improving the treatment of fungal infections as PCM, few studies have been reported. In the current study, we verified that a single-dose administration of an adjuvant that induces T helper (Th) 1 immune response (complete Freund's adjuvant [CFA]) in P. brasiliensis–infected mice was sufficient to break the lack of immune response to the fungus observed in infected mice. Four weeks after treatment, the CFA-treated mice presented a mild infection in the lungs characterized by preserved lung structure and small fungal burden, whereas control mice that had been treated with incomplete Freund's adjuvant presented many granulomatous lesions and high fungal burden. The immunotherapy with Th1-inducing adjuvant might be a valuable tool in the treatment of PCM and potentially useful for faster and efficient cure of PCM in humans.
| Paracoccidioides brasiliensis is a thermally dimorphic human pathogenic fungus that causes paracoccidioidomycosis (PCM), the most prevalent human systemic mycosis in Latin America, being endemic in Brazil, Argentina, Venezuela and Colombia. This infection is acquired by inhalation of airborne propagules found in nature, which reach the lungs and are converted to the yeast form [1],[2]. The yeasts can either be eliminated by immune-competent cells or disseminated into tissues through lymphatic or hematogenous routes. PCM is characterized by granulomatous inflammation, intense immunological involvement with suppression of cellular immunity and high levels of non-protective antibodies in serum [3]. The disease may present a broad spectrum of clinical and pathological manifestations ranging from asymptomatic pulmonary infection to severe and disseminated forms [4],[5]. The chronic progressive form of the disease (CF) is the most common clinical presentation and predominantly affects adult males, with frequent pulmonary, mucosal, cutaneous and adrenal involvement. Although the outcome of the infection can be due to several factors, it is especially dependent on the protective capacity of the host immune system. The cell-mediated immune response represents the main mechanism of defense in PCM [1]. Conversely, it has been reported that a high level of humoral immune response is associated with increased disease dissemination [6].
The mechanisms underlying resistance or susceptibility to PCM remain to be elucidated. The development of the appropriate CD4+ T helper (Th) subset is important for PCM resolution and several studies have shown that different disease outcomes can be derived from the commitment of precursors to either Th1 or Th2 lineage [7],[8]. Resistance to P. brasiliensis infection has been related to interferon-γ (IFN-γ) and other Th1-type cytokines [9]–[11], while susceptibility has been linked to the preferential production of the Th2-type cytokines, i.e., interleukin (IL)-4, IL-5, and IL-10 [12]–[14]. Several investigators have suggested that progressive disseminated forms of PCM in humans are associated with various degrees of suppressed cell-mediated immunity [1],[15],[16]. This anergy can be reversed after successful therapy, when normal levels of T cell function are partially or completely restored [17].
The prognosis of PCM has been improved through antimycotic drugs, however treatment regimens require an extended period of time often associated with relapses. P. brasiliensis has the peculiarity of responding to treatment with sulpha drugs. Nevertheless, regimens with these agents often require extended period of maintenance therapy that may range from months to years. Clinically, the antifungal drugs most commonly used for PCM include amphotericin B, sulpha derivatives and azoles, but their toxicity can be a limiting factor in treatment [18],[19]. These concerns, together with the elucidation of the protective immune response against PCM have renewed interest in the development of alternative therapeutic strategies such as immunotherapeutic procedures, which can be useful for controlling PCM. The present study was designed to verify if immunomodulation with CFA could play a protective role in experimental PCM leading to a less severe infection with decreased fungal burdens in the lungs.
Yeast cells of virulent Pb 18 strain of P. brasiliensis were cultured at 37°C in YPD (Yeast Extract/Peptone/Dextrose) Medium (Difco Laboratories, Detroit, USA) for 7 days and washed three times in 0.01 M phosphate-buffered saline (PBS), pH 7·2. Viability of yeast cells was determined by the fluorescein diacetate-ethidium bromide treatment [20].
BALB/c mice, aged 6–8 wk, were bred and maintained under standard conditions in the animal house of the Medical School of Ribeirão Preto, University of São Paulo, Ribeirão Preto, SP, Brazil. All animal experiments were performed in accordance with protocols approved by the School of Medicine of Ribeirão Preto Institutional Animal Care and Use Committee. Mice were inoculated intravenously with 1×106 viable yeast cells in 100 µl of PBS. On day 20 postinfection, mice were injected subcutaneously with 100 µl of CFA or IFA (Sigma Chemical Co., St. Louis, USA), both emulsified in PBS in a ratio of 1∶1. Mice were killed on day 30 after treatment and their lungs were aseptically removed. One lung from each mouse was used for histopathology analyses and the other for quantification of fungal burden and cytokines.
The lungs were fixed in 10% neutral buffered formalin for 24 hours and embedded in paraffin. Tissue sections (5 µm) were stained with hematoxylin and eosin (H&E) or silver methenamine (Grocott) to detect the mycotic structures using standard protocols. Samples were analyzed by light microscopy in an Axiophot photomicroscope (Carl Zeiss, Jena, Germany) coupled with a JVC TK-1270 camera (Victor Company of Japan Ltd, Tokyo, Japan). The area of individual granulomas, as well as the total area of the lung sections and the area taken by granulomas per slide, was measured by computer-aided image analysis (ImageJ 1.37v, National Institutes of Health, Bethesda, USA). The following data were thus generated: granuloma area (mean area of all granulomas in each lung section), granuloma relative area (% represented by total granuloma area/total area of the lung sections) and number of granuloma cells per area (total number of cells from a granuloma section/the area of the respective granuloma section) of each mouse.
The lungs were weighed and homogenized in 1 ml of sterile PBS using tissue homogenizer (Ultra-Turrax T25 Basic, IKA Works, Inc., Wilmington, USA). To determine the number of CFU, lung homogenates were diluted 1∶10 in PBS. Aliquots of 100 µl of each sample were dispensed into Petri dishes containing brain heart infusion agar (BHI, Difco) supplemented with 4% (v/v) of heat-inactivated fetal calf serum (FCS, Gibco BRL, Gaithersburg, USA). The plates were incubated at 37°C, the colonies were counted 14 days later, and then, the number of CFU per gram of tissue was calculated. For cytokine determination, remaining lung homogenates were centrifuged at 5,000×g for 10 minutes and the supernatants stored at −20°C until cytokine determination. Supernatants were analyzed as duplicate samples from replicate wells. A sandwich-type ELISA was used to determine IL-12, IFN-γ, TNF-α, IL-4, IL-10, and TGF-β levels, using OptEIA ELISA kits (BD PharMingen, San Diego, USA), according to the manufacturer's recommendations.
Inh-ELISA was performed as previously described [21]. Briefly, inhibition standard curve was constructed by adding different concentrations of P. brasiliensis gp43 (from 1 ng to 30 µg/ml) in 100 µl of normal serum and then adding 100 µl of the standardized concentration of monoclonal antibody (MAb) anti-gp43 (10 µg/ml). Serum samples (100 µl) were added to 100 µl of MAb anti-gp43. Normal serum was used as a negative control. Polystyrene plates (Corning Costar Co., Corning, USA) were coated with 500 ng of gp43 in 0.06 M carbonate buffer (pH 9.6) per well (100 µl/well) overnight at 4°C. After, the plates were blocked by incubation with 200 µl of 1% bovine serum albumin in PBS per well for 1 h at 37°C; washed 3 times and 100 µl from inhibition standard curve, samples and controls were added per well and allowed to stand for 2 h at 37°C. After being washed 3 times, 100 µl of goat anti-mouse immunoglobulin G-peroxidase (Sigma) was added, and the plates were incubated for 1 h at 37°C. After further washings, the reaction was developed with a solution of o-phenylenediamine (0.5 mg/ml; Sigma) and 0.005% H2O2. The reaction was stopped with 4 N H2SO4 after 8 to 10 min of incubation in the dark. Optical densities were measured at 490 nm on a PowerWave X microplate reader (Bio-Tek Instruments, Inc., Winooski, USA). The degree of inhibition in MAb binding was shown to be reciprocal to the concentration of circulating antigen in the sample. The cutoff point was established as the receiver operator characteristic (ROC) curve.
Statistical determinations of the difference between means of experimental groups were performed using two-tailed Mann-Whitney U-test. Differences which provided P<0.05 were considered to be statistically significant. All experiments were performed at least three times.
The depression of cell-mediated immune responses has been associated with severe PCM in humans and in the experimental host [1],[15],[16],[22]. However, the propensity for persistence of the fungus in infected tissues appears to be consequence of cell-mediated immune dysregulation with suppression of Th1 and overexpression of Th2 responses [12]–[14].To evaluate whether therapeutic immunostimulation is able to interfere in experimental murine PCM and restore the host immune response, we selected immunomodulators for therapy strategy based on the induction of Th1 or Th2 immune response. Since CFA supports a Th1 status, while incomplete Freund's adjuvant (IFA) promotes a Th2 status [23], BALB/c mice were divided into two groups and treated with CFA or IFA on day 20 after infection with P. brasiliensis. The progression of P. brasiliensis infection was determined by lung histopathology and analysis of colony-forming unity (CFU), parameters that are considered trustworthy to discriminate susceptible and resistant mice to systemic fungal infection [9],[12],[19],[24]. At 20 days of infection the mice presented 5.8×104 CFU/g of lung tissue (Figure 1A, dashed line) and compact granulomas (data not shown), for this reason, this time was chosen for the treatment regimens. On day 30 after treatment (50 days postinfection), the lungs from IFA-treated mice presented multiple sites of focal and confluent epithelioid granulomas with lymphomonocytic halos circumscribing a high number of viable and nonviable yeast cells (Figure 2A, C and E). Morphometric analysis of the lungs from IFA-treated mice revealed a number of granulomas of 41±5.2, with a relative area of 40.7±6.2%. These granulomas presented 12.2±1.8% of yeast cells and 6±0.6% collagen (data not shown). In contrast, in the P. brasiliensis-infected mice treated with CFA, no granulomas or yeast cells were seen in the pulmonary sections examined and a well-preserved alveolar architecture was observed on day 30 after treatment (Figure 2B, D and F). Most importantly, the treatment with CFA induced a 2.4 log reduction in the fungal burden when compared to the IFA-treated mice, corresponding to 99% less CFU (Figure 1A). The CFU data are in agreement with the histopathology analyses, pointing out that therapeutic immunostimulation led to an increased clearance of fungal burden from lungs.
In order to evaluate the impact of the treatment with adjuvant, the animals were weighed weekly until the study end point. We observe that the animals of therapy group gained more weight (20%) than the control group (data not shown). These results can be correlated with a good prognostic in the PCM.
When we analyzed the production of pro and anti-inflammatory cytokines in the supernatants of lung homogenates from the P. brasiliensis-infected BALB/c mice treated with CFA or IFA, we observed that the IFA-treated group produced low levels of IFN-γ, IL-4, IL-12, TNF-α, IL-10 and TGF-β (Figure 1B–G), suggesting a suppression of the immune response in these animals. In contrast, CFA-treated mice produced high levels of these pro and anti-inflammatory cytokines (Figure 1B–G). Although many reports have demonstrated that the Th2 pattern is associated with a severe disease, whereas a Th1-biased immune response is linked to the asymptomatic and mild forms of PCM [9]–[14], others have shown that the induction of inflammatory cytokines, such as IFN-γ and TNF-α, can lead to overproduction of nitric oxide that has been associated with suppression of cell immunity [25]–[27]. Recently, it was demonstrated that the anti-inflammatory Th2 cytokine IL-4 has a dual role in PCM, leading to a protective or a disease-promoting effect depending on the genetic background of the host [28]. Regarding TGF-β, we observed that this cytokine is produced by pulmonary epithelium, so we hypothesized that it might contribute to the lung tissue renewal (unpublished data). In this study we obtained an effective protection against P. brasiliensis infection even in the presence of anti-inflammatory cytokines, suggesting that, in this therapy model, the protective effect against PCM seems to be dependent on the induction of a mixed Th1/Th2 immune response pattern. The production of both inflammatory and anti-inflammatory cytokines is extremely helpful to balance the immune response, since anti-inflammatory cytokines can control the inflammatory responses, which can result in local pathology and systemic and centrally controlled adverse events. CD4+ T cells also play a role in the regulation of inflammation [29]. On the basis of the differences between the groups treated with CFA or IFA, we suggest that the protection induced by CFA injection was due to a noticeable increase in the pulmonary levels of cytokines, which probably broke the immunosuppression status observed in the infected mice treated with IFA. Nonetheless, we cannot exclude the involvement of other mechanisms, such as modulation by regulatory T cells [30], apoptosis in the antigen-specific T cells [31], and Fas-FasL and CTLA-4 expression [32].
The levels of circulating antigen in the mice infected and treated with IFA were two-fold higher than those treated with CFA (Figure 1H). These results supported by other reports that showed that the depression of cell-mediated immunity is associated with the high levels of specific circulating antibodies or soluble antigens in disseminated disease [13],[15],[21].
Although many studies on protection against PCM have been performed, only few of them have reported the efficacy of the immunostimulatory therapy. In one of these studies, the therapy with peptide p10 from gp43, emulsified in CFA, and chemotherapy was used in an attempt to improve the treatment of PCM [33]. The combined treatment showed a beneficial effect when administered at 48 h or 30 days after challenge. However, the control mice that received only CFA and the non-immunized mice presented similar lung fungal burden. These data are in contrast to those observed herein. This difference might be due to distinct experimental protocols used, such as challenge route, dose, and treatment regimen. Nevertheless, other reports have demonstrated that the use of immunostimulatory therapy can lead to a positive prognostic in fungal diseases [34]–[36]. Basically, therapeutic immunostimulation can be used by reinforcing or broadening defenses when specific immune responses are unable to do this during the natural course of the PCM.
The present study demonstrated that a single-dose administration of the Th1-inducing adjuvant (CFA) in P. brasiliensis-infected mice was sufficient to break the anergy observed in these animals restoring their ability to mount an effective immune response to the fungus. While the control mice presented large amount of yeasts and extensive sites of parenchymal lung injury, the CFA-treated mice were capable to control not only the fungal systemic dissemination but also its growth, leading to a noticeable fungal clearance without apparent lung injury. Our results indicate that Th1-inducing adjuvant proved to be a valuable tool in the treatment of PCM. Overall, these data open new possibilities for the potential use of Th1-inducing adjuvant not only as a sole therapy but also as an adjunct to conventional antifungal therapy against PCM, improving the regular chemotherapy and reducing the time of treatment. |
10.1371/journal.pcbi.1001095 | Identifying Causal Genes and Dysregulated Pathways in Complex Diseases | In complex diseases, various combinations of genomic perturbations often lead to the same phenotype. On a molecular level, combinations of genomic perturbations are assumed to dys-regulate the same cellular pathways. Such a pathway-centric perspective is fundamental to understanding the mechanisms of complex diseases and the identification of potential drug targets. In order to provide an integrated perspective on complex disease mechanisms, we developed a novel computational method to simultaneously identify causal genes and dys-regulated pathways. First, we identified a representative set of genes that are differentially expressed in cancer compared to non-tumor control cases. Assuming that disease-associated gene expression changes are caused by genomic alterations, we determined potential paths from such genomic causes to target genes through a network of molecular interactions. Applying our method to sets of genomic alterations and gene expression profiles of 158 Glioblastoma multiforme (GBM) patients we uncovered candidate causal genes and causal paths that are potentially responsible for the altered expression of disease genes. We discovered a set of putative causal genes that potentially play a role in the disease. Combining an expression Quantitative Trait Loci (eQTL) analysis with pathway information, our approach allowed us not only to identify potential causal genes but also to find intermediate nodes and pathways mediating the information flow between causal and target genes. Our results indicate that different genomic perturbations indeed dys-regulate the same functional pathways, supporting a pathway-centric perspective of cancer. While copy number alterations and gene expression data of glioblastoma patients provided opportunities to test our approach, our method can be applied to any disease system where genetic variations play a fundamental causal role.
| It is now being recognized that complex diseases should be studied from the perspective of dys-regulated pathways and processes rather than individual genes. Indeed, various combinations of molecular perturbations might lead to the same disease. In such cases, responses to these perturbations are expected to converge to common pathways. In addition, signals that are associated with each individual perturbation might be weak, rendering studies of complex diseases particularly challenging. Aiming to provide an integrated perspective on complex disease mechanisms we developed a novel computational method to simultaneously identify causal genes and dys-regulated pathways. Starting with an identification of a disease-associated set of genes and their statistical associations with genomic alterations, we utilized graph-theoretical techniques and combinatorial algorithms to determine potential paths from the genomic causes through a network of molecular interactions. We applied our method to sets of genomic alterations and gene expression profiles of Glioblastoma multiforme (GBM) patients, uncovering candidate causal genes and causal paths that are potentially responsible for the altered expression of disease associated target genes. While copy number alterations and gene expression data of GBM patients provided opportunities to test our approach, our method can be applied to any disease system where genetic alterations play a fundamental causal role, and provides an important step toward the understanding of complex diseases.
| Complex diseases are typically caused by combinations of molecular perturbations that might vary strongly in different patients, yet dys-regulate the same component of a cellular system [1]. In recent years, whole-genome gene expression sets have increasingly been used to search for markers, allowing an improved diagnosis of diseases or classification of their subtypes [2], [3], [4], [5], [6], [7], [8]. Several approaches combined expression measurements with various types of direct or indirect pathway information, leading to improved disease classification [9], [10], [11], [12], prioritization of disease associated genes [13], [14], [15] and identification of disease specific dysregulated pathways [16]. Furthermore, considerable efforts towards integrated approaches for uncovering disease causing genes [17], [18] and elucidation of relations between variability in gene expression and genotype [19] have recently been made. In particular, Tu et al. developed a random walk approach to infer regulatory pathways [13], [14], [20] in yeast. Suthram et al. [21] further improved this approach by using the analogy between random walks and current flow in electric circuits. Recently, Yeger-Lotem et al. developed a min-cost flow based algorithm, uncovering cellular pathways that are implicated in several neurodegenerative disorders [22].
Studying associations between individual disease genes and genotype alterations allowed us to uncover potential causative factors and affected molecular entities. While previous methods provided valuable insights into the modular nature of diseases by elucidating groups of differentially expressed genes, the flow of information from potential causes to effected genes in the molecular interaction network hasn't been investigated. In this paper, we present a genome-wide approach to simultaneously determine dys-regulated pathways and their putative causes/factors. We utilized gene expression and genomic alteration profiles of 158 glioblastoma multiforme (GBM) patients. We started by selecting a set of differentially expressed target genes, and then identified pathways connecting genes that are located in areas of genomic alterations. Then, we selected target genes, choosing pathways that are likely to explain the expression abnormalities of target genes. Consistent with the general strategy of eQTL analysis, we assumed that expression variations of the target genes are, at least in part, caused by genomic alterations. Specifically, we first used association analysis to identify possible cause-target gene pairs. Then, we modeled the propagation of information from a potentially causal gene to a target gene as the flow of electric current through a network of molecular interactions. To assess the significance of identified pathways we carefully designed a permutation test. Finally, we used a graph-theoretical approach to further narrow down the selected set of putative causal genes. We validated our approach by testing the enrichment of selected causal genes with known GBM/Glioma disease genes and literature searches. We also examined the subnetworks, connecting causal and target genes and identified cancer hub genes and sets of functionally related genes which indicate involvement of specific cellular pathways. Among these pathways we found several expected key players such as EGFR and Insulin Receptor signaling pathways, RAS signaling, as well as a glioma-associated regulation of transforming growth factor-β2 production and SMAD pathway. Importantly, such pathways can be considered as “GO biological process hubs” or “highways”, connecting many different causal genes with their targets. Such an observation supports the hypothesis that many different genomic alterations potentially dys-regulate the same pathways in complex diseases. In addition, we analyzed the global properties of identified associations and found a cluster of causal/disease gene activities on chromosomes that are strongly affected by genomic alterations. Such results allowed us to identify candidate causal genes for prominent signaling and regulation proteins that putatively play a role in GBM. Comparing our method to a basic genome-wide association approach, we demonstrated the increased predictive power of our model.
We developed a novel computational method to identify causal genes and associated dys-regulated pathways by an integration of several layers of data, including profiles of gene expression and genomic alterations (Fig. 1). Specifically, we assembled an interaction network, utilizing molecular interaction data such as protein-protein interactions, phosphorylation events and protein-transcription factor interactions. Briefly, our algorithm consists of four main steps (Figures 1A–D): (i) selection of a set of differentially expressed target genes, (ii) identification of possible causal loci of each target gene by an eQTL-analysis, (iii) identification of a set of putative causal genes by determining pathways between causal and target genes through the network of molecular interactions, and (iv) determination of a subset of causal genes that best explain the underlying disease cases. In the following, we present a more detailed description of these four steps. Further details are described in the corresponding sections of Materials & Methods.
Since a complex disease may manifest itself differently in patients, we first developed a method that selects a set of genes that are differentially expressed in the disease cases and cover individual patient alterations. To identify such representative genes, we modeled the selection of target genes as a multi-set cover problem (Fig. 1A). Specifically, we determined a set of genes that were differentially expressed in 158 glioblastoma cases compared to 32 non-tumor control cases (see selection of target genes section in Materials & Methods). We defined that a differentially expressed gene covers a particular disease case if the gene was differentially expressed in the underlying case. Clearly, genes that cover many cases are expected to represent genes and pathways commonly dys-regulated in the disease. To capture disease heterogeneities we also demanded that each disease case was covered by at least a certain number of target genes, a key parameter of our approach. Intuitively, with very small coverage we can identify only the most commonly differentially expressed genes. By increasing coverage we can capture genes that are specific to smaller subgroups of patients. Thus, we required a certain level of coverage and simultaneously demanded that each gene covers as many cases as possible. To achieve this goal, we formulated the problem as a minimum multi-set cover (see selection of target genes section in Materials & Methods) and solved it using a greedy algorithm. We tested several combinations of coverage and the number of outliers (a second, less prominent parameter of the algorithm) and observed that obtained gene sets strongly overlapped, demonstrating the robustness of our approach (see Text S1 for details of the algorithm and parameter settings). Demanding coverage of 55 and allowing 3 outliers, we selected 74 target genes as presented in Fig. 2 (see Table S1 for an annotated list of target genes).
The goal of this step is to identify an initial set of possible associations between copy number variations and target genes for further analysis (Fig. 1B). Since genomic variations in neighboring regions tend to be highly correlated, we first chose a subset of 911 representative loci (i.e. tag loci), significantly lowering computational costs (see eQTL mapping section in Materials and Methods). We observed that the number of genes that a tag locus can harbor varied strongly and found on average 27 genes per tag locus. Applying a standard eQTL approach [19], [23] we performed a linear regression analysis, allowing us to determine genome-wide associations between the expression of target genes and copy number alterations of tag loci. Specifically, we calculated p-values for each gene-locus pair under the null hypothesis that the slope of the linear regression is 0. This way we selected, for further analysis, 3,091 associated gene-locus pairs (p<0.01), amounting to <5% of all 67,414 (911×74) tested pairs. On average, we selected 41 associated tag loci per target gene, while 776 tag loci had at least one target gene (see Text S1 for algorithmic details).
The relatively liberal p-value threshold used in the previous step allowed us to retain most of potentially interesting relationships. Although this step filtered out the least promising pairs, a large number of false positives are expected to pass this threshold. Reducing false discovery rate by simply decreasing the p-value threshold would retain extremely well correlated loci-target gene pairs only, therefore missing a large spectrum of potentially interesting pairs. In fact, the correlation between copy number variation in the causal gene and the gene expression of its target gene doesn't have to be strong since such a signal might have been affected by varying degradation rates and posttranslational modifications. Furthermore, genotypic alterations in several loci might lead to the dys-regulation of the same pathway and therefore changing the expression of a target gene in potentially non-additive, epistatic ways. Since each genetically altered region might harbor a large number of genes, we also aimed to identify the most likely causal genes within each region.
In order to account for such effects we utilized protein-protein, protein-DNA and phosphorylation networks (Fig. 1C). Existence of statistically significant paths through an interaction network, connecting putative causal and target genes not only provides additional support for the relationship but also helps to identify genes that participate in propagating the signal. This approach also allows identifying the gene(s) within the altered regions which were most likely the cause of the observed expression changes of the selected target genes (Fig. 1C). Motivated by the results of Suthram et al. [21], we adopted a variant of a circuit flow algorithm and modeled the problem of finding a pathway through an interaction network as current flow in an electric circuit. We defined the conductance of each interaction as a function of the expression correlation of the genes at the endpoints of edges and the target gene. Such a model allows the current to preferentially use interactions that more likely mediate information from a causal to a target gene. Stipulating that only transcription factors can change the expression of genes, we required that a causal path ended with a link between a transcription factor and the target gene. The flow of current from the target to its potential causal genes was computed by solving a system of linear equations, allowing us to find a set of candidate causal genes for each target gene. Importantly, we considered edges corresponding to phosphorylation events and protein-DNA interactions as directed, prompting a computational problem that theoretically can be tackled with a linear programming approach [21]. However, the large size of the underlying human interaction network imposed considerable computational costs, prompting us to develop a heuristic that preserved the directions of such molecular interactions. As a null-model, we utilized a permutation test to estimate the statistical significance of the current flow. After obtaining empirical p-values we selected candidate causal genes for each target gene if the empirical, gene specific p-value was <0.05 (for algorithmic details and parameter settings please see solution of the electric circuit problem section in Materials & Methods and Supplement Text S1). We obtained 1,763 pairs, consisting of 74 target and 701 potential causal genes that included a significant number of GBM and glioma-specific genes (Table 1). Since we identified associated gene-locus pairs with p<0.01 and found target-causal gene pairs with p<0.05, all 1,763 pairs had an estimated nominal p-value <5×10−4.
While the electric circuit approach reduced the number of putative causal genes significantly, the size of this gene set was still considerably large. In the final step, we applied another filter by considering two approaches – a statistical method and a hypothesis driven optimization approach. In the statistical approach, we accounted for multiple hypothesis testing and used a p-value cut-off of 5×10−8, producing 280 candidate causal genes. In the optimization-based approach, we identified relevant causal genes by selecting the set of genes that best explained all disease cases. We defined that a putative causal gene explains a disease case if its corresponding tag locus has a copy number alteration and its affected target genes (i.e., genes sending a significant amount of current to the causal gene) were differentially expressed in the underlying disease case. In other words, if a link between a causal gene and a disease case existed, we expected to observe both a genomic alteration of a causal gene and differential expression of its target gene in the same disease case. Since a causal gene may potentially affect one or more target genes, we defined the weight of the explanation as the number of such target genes. Therefore, a gene that explained a disease by perturbing a larger number of target genes had a higher weight, increasing the likelihood to be chosen as a final causal gene (Fig. 1D). To choose a set of causal genes explaining all cases except a few outliers with a minimum number of causal genes, we formulated the problem as a variant of the minimum weighted multi-set cover problem (please see selecting a final set of causal genes section in Materials & Methods and Text S1 for algorithmic details). Utilizing a greedy algorithm, we determined a set of 128 putative, final causal genes that were involved in 625 causal and target gene pairs. Using a permutation test, we found that the random selection of a gene set of at most this size occurred with p<3.1×10−4.
In the following, we provide a quantitative validation of the set of putative causal genes, pathway hubs and target genes. Where applicable, we also compared our results to previous approaches. Subsequently, we established the robustness of our method with respect to parameter settings. Finally, we analyzed individual genes and pathways.
To assess the significance of our set of causal genes, we determined the overlap with sets of GBM/glioma specific genes. In particular, AceView [24] provided a list of 93 GBM specific genes. In the first step of the algorithm, we determined associations between copy number variations and expression of target genes, yielding 16,056 associated genes that had a large, but statistically insignificant overlap with the set of glioblastoma specific genes (p<0.56, Table 1). The application of the electric circuit algorithm reduced this set to 701 candidate causal genes with a significant enrichment of 10 GBM specific and 25 Glioma related genes (p<0.05, Table 1). We also checked the advantage of using the current flow approach instead of simply selecting pairs based on more stringent p-value cut-offs. Namely, given our eQTL results, we used a Bonferroni-corrected threshold of 1.5×10−7, providing 24 pairs between 4 target genes and 22 loci that harbor a total of 1,026 genes, including 12 GBM relevant genes from AceView (p<0.003, Table 1). However, this approach failed to find any significant associations for most of the target genes. For the 4 target genes, we obtained a rather big set of candidate causal genes, which was not enriched with glioma genes in DAVID.
Next, we focused on the last step of the algorithm. As a result of the current flow step we obtained 1,763 pairs with a nominal p-value <5×10−4, involving 701 causal genes. Using the weighted set cover approach, we identified 128 causal genes that harbored 6 GBM relevant genes (Table 1). Specifically, we found that both sets shared CDKN2A, EGFR, ERBB4, PTEN, RB1 and TP53 (p<4.7×10−4). Utilizing a set of glioma relevant genes from DAVID database [25], [26], we obtained consistent results (Table 1). In contrast, by Bonferroni-correcting causal-target gene pairs we obtained 280 causal genes, including only 4 GBM related genes according to AceView (p<0.17, Table 1).
To test an alternative approach, we greedily chose loci with smallest p-values until we pooled at least 128 putative causal genes. The obtained set of putative causal genes included only 2 GBM genes (p<0.3), suggesting that the current flow algorithm and the subsequent filtering step with a set-cover allowed us to uncover more cancer relevant genes than the simple association approach.
Focusing on the final set of 128 causal genes, we utilized canonical pathway data from DAVID and found that the final set of 128 causal genes was significantly enriched with glioma, cell cycle genes, p53 signaling pathway and proteasomal genes (p<0.05). In Table 2 we listed the most enriched annotated pathways, their genes and p-values. The complete list of 128 final causal genes is shown in Fig. 2, and an annotated list is provided in Table S2.
We also assessed the importance of genes in the paths from putative causal genes to their target genes. As described in identifying dysregulated pathways section in Materials &Methods, we identified causal paths between a target and a causal gene by finding a maximum current path through the network of molecular interactions. In particular, we demanded that the genes in causal paths have significant p-values while the current passing through all genes in the path is maximized (please see identifying dysregulated pathways section in Materials & Methods and also Text S1 for algorithmic details), allowing us to identify 461 genes in 995 interactions. Using a threshold of more than 10 occurrences in causal paths (corresponding to 20% of most frequently appearing genes), we observed the emergence of hubs, genes that appeared in a disproportionally large number of pathways (Fig. 2). Such a set of hubs contained important transcription factors such as MYC and E2F1 and oncogenes such as JUN and RELA and was enriched with genes that appeared in cancer pathways (p<2.2×10−8), the cell cycle (p<3.5×10−6) and several important signaling pathways from DAVID. While such hub genes were clearly related to cancer, we hardly would have identified them by analyzing differentially expressed genes or copy number alterations alone, demonstrating that the pathway-based approach considerably helped us to uncover these important players.
Utilizing DAVID, we also found that our target gene set was enriched with genes in the cell cycle (p<7.6×10−4), p53 signaling pathway (p<9.1×10−4), and RB Tumor Suppressor/Checkpoint Signaling in response to DNA damage (p<4.8×10−3). Among target genes, we also found up-regulated WEE1, a tyrosine kinase that phosphorylates CDK1 [27], a signaling event that is crucial for the cyclin-dependent passage of various cell cycle checkpoints. Previous reports suggested that overexpression of WEE1 is critical for the viability of some cancer types, and cell lines displaying higher expression levels of WEE1 are sensitive to WEE1 inhibition [28].
In an additional test, we eliminated the requirement that the last node on a path leading to a target gene must be a transcription factor. With this change, we selected parameters in our multiset-cover approach to obtain an alternative set with approximately the same number of target genes and we found that it was almost disjoint from our original set of 74 target genes (Fig. 3A). Despite these differences, the final sets of causal genes had a strong overlap (Fig. 3B) of 58 genes that we found in both sets. Such a level of robustness is consistent with a pathway-centric view of complex diseases: different sets of target genes are bundled within dys-regulated pathways that are influenced by specific combinations of causal genes. Even though the two target gene sets looked largely different, both sets include genes that are differentially expressed in the disease cases. In addition, we found that the genes are close relatives in the network: the average distance between the two sets of target genes is 1.7 (p = 1.7×10−12), suggesting that the genes were selected from the same dysregulated pathways.
In Fig. 4A, we show the profile of genomic alterations in GBM where we observed large areas of genomic amplification on chromosome 7 and deletions on chromosome 10 (upper panel), alterations that coincided with the genomic locations of EGFR and PTEN. We located the genomic position of our 128 causal genes and counted the number of corresponding target genes. We largely observed that causal genes on chromosome 7 and 10 were strongly connected to target genes, a pattern that strongly coincided with the signature alterations of GBMs.
Since a target and a causal gene might be located on different chromosomes, we determined the occurrences of such chromosome combinations using all target-causal pairs. Constructing such a matrix (Fig. 4B) we found that strong causal signals emerged from chromosomes 7 and 10. In turn, we observed that target genes fell into three large clusters. In particular, target genes on chromosomes 2, 3, 6, 10, 11, 12, 19 and 20 appeared to have numerous links to causal genes located on chromosomes 7 and 10. Focusing on target and causal genes in these areas, we found a large cluster (box, Fig. 4C) of up-regulated genes that were connected to an array of largely down-regulated causal genes.
In addition, we also looked for literature-based validation of other causal genes. In particular we found RHOBOTB2, a recently discovered tumor suppressor gene [29], in our set of 128 causal genes. We observed that this gene lacked a strong genomic alteration signal, suggesting that our approach was also capable of discovering a subtle causal signature that may have been otherwise missed with a simple disease association analysis. We also found some causal genes with strong genomic alterations that, although not included in AceView nor in DAVID, are well known to be associated with cancer. For example, our final causal gene set included GBAS (for its causal network, see Text S1), a gene that was reported amplified in more than 40% of glioblastomas [30], [31] and CEBPA (enhancer binding protein) that was amplified in about 10% of leukemia cases [32].
We obtained 128 causal subnetworks from causal genes to their target genes (see identifying dysregulated pathways section in Materials and Methods). For each causal subnetwork, we performed an enrichment analysis of GO-annotated biological processes. Due to the hierarchical structure of GO terms, results included many redundant terms, and general terms tend to have more hits. In Table 3, we listed the most specific GO-annotated biological processes with which more than one subnetworks are enriched. For the full list, see Table S3. In Supplementary Dataset S1 we provided a cytoscape file that allows an interactive exploration of enrichment in the GO hierarchy. The frequently enriched GO processes included several classical cancer-related pathways. For example, 9 causal subnetworks are enriched with epidermal growth factor receptor signaling pathway that has anti-apoptotic properties and may enhance proliferation, invasion, and migration of glioma cells [33], [34], [35]. Similarly, 6 causal-target relationships affected the Insulin signaling pathway. Indeed, recent reports provide an additional evidence for the role of this pathway in glioblastoma [36], supporting the hypothesis that alterations in different genes may dysregulate the same pathways and cause the same disease. Other less frequent pathways were positive regulation of MAP kinase activity, regulation of nitric-oxide synthase activity, estrogen receptor signaling pathway, JAK-STAT cascade and the regulation of transforming growth factor-beta2 production. In particular, transforming growth factor-beta2 (TGFB2) is known to be an important modulator of glioma invasion [37], [38]. Of particular interest is also a related SMAD pathway that occurred in two of our causal subnetworks. While it is debated if this pathway plays a role in TGF β-promoted oncogenesis, a recent study indicated that SMAD-dependent signaling through the induction of PDGF-B has a proliferative and oncogenic role in glioma [39], which is in line with the presence of SMAD genes in our causal subnetworks.
Testing if these GO-processes were enriched in the set of target genes, we only found an enrichment of a small number of very general, mostly cell-cycle related pathways (see Table S4 for the complete list). Only one term “G1/S transition of mitotic cell cycle” overlapped with the list of most specific terms discovered through the analysis with flow-based causal paths. The lack of specific terms in the GO analysis using target genes was expected since target genes were sampled from multiple dys-regulated pathways, therefore not leading to significant enrichment of specific pathways.
We took a closer look at paths involving PTEN and EGFR. In Fig. 5, we show a subnet of dysregulated pathways with PTEN as a causal gene. We observed that the influence PTEN might exert on target genes was largely mediated by prominent transcription factors, such as TP53, MYC and MYB. Compared to pathways from DAVID [25], [26], this small network of causal paths was enriched with cell cycle genes (p<0.003) and glioma genes (p<0.02) as well as various types of cancer genes. As their causal roles are indicated in Fig. 4C, we observed that PTEN and CDC2 (see Text S1) might exert their influence on the expression of WEE1 through transcription factors TP53 and E2F4. Since CDC2 codes for CDK1, which is phosphorylated by WEE1 [27] , our results suggest a feedback loop that might be important for cancer.
EGFR is highly expressed in disease cases and was selected as both a target and causal gene. The considerable amplifications of chromosome 7 make EGFR a strong candidate for a causal gene. Indeed, we found causal paths that connected EGFR to a few target genes (Fig. 6A). However, we also found a rather large number of causal genes that regulated the expression of EGFR as a target gene (Fig. 6B). Such observations suggest that EGFR might play a dual role as a driver of changed gene expression as well as integrator of causal molecular information from other genomic sites. Indeed, we found numerous disease cases where EGFR was over-expressed without alterations in its genomic location. Instead, we observed that there exist a number of potential causal genes of EGFR with copy number alterations such as ANXA11, CDKN2A, CHUK, PTEN, IFNA4 and ZNF107 among others. Utilizing pathway information from DAVID, we found that the subnet with EGFR as a target gene was highly enriched with glioma genes (p<0.004), the MAPK signaling pathway (p<0.02), and pathways in cancer in general (p<8×10−8).
Integrating phenotypic, genomic and interaction data, we introduced a novel approach for the simultaneous identification of causal disease genes and dys-regulated pathways. Such causal genes may include potential drivers of a tumor's emergence as well as potential drug targets. After selecting target genes that covered the underlying disease cases, we determined associations between altered genomic loci and changed expression levels of target genes by a simple eQTL analysis. The key idea of our approach is to combine evidence from association analysis with evidence from pathway analysis. We also demonstrated the power of graph-theoretical approaches in the selection of gene sets and determination of cause-target relationships. Indeed, set cover approaches are increasingly recognized as appropriate tools for selecting disease genes [16], [40], while current flow approaches or equivalent random walk models have been successfully used for modeling of information flow in biological and social networks [41], [42], [43].
Adopting a current flow algorithm, we combined gene expression and molecular interaction data to determine causal paths through interaction networks. This approach allowed for preferential use of network paths supported by expression data, bypassing potential problems of pure topology based methods such as shortest paths that treat all edges equally. Namely, the assignment of resistance to network edges pushed electric current preferentially through nodes that were expression-correlated with the target genes. However, our method also tolerates a fraction of non-correlated nodes, balancing the impact of network connections and a strongly varying degree of gene expression correlation of nodes in the paths.
Current networks of protein interactions, protein-DNA interactions and phosphorylation events are incomplete and noisy. In addition, transcription factors for many genes are unknown, a shortcoming that certainly affected the completeness of our results. However, the problem is alleviated by the fact that cancer is considered as a disease of pathways, suggesting that there exist many ways of selecting a representative set of target genes that represent dys-regulated pathways. Considering a cluster of neighboring genes that participate in the same pathway, any member of the cluster might serve as a target gene to uncover causal genes dys-regulating the underlying pathway. We found that the choice of different target genes provided robust results, diminishing the effects of incomplete data.
We used linear regression for associations to take advantage of its simplicity. To capture the complex relationship of copy number and gene expression more accurately, other non-linear methods can also be considered. However, little is currently known about the precise impact of gene copy number variations on gene expression levels in model organisms, a problem that might even be aggravated by the presence of potential epistatic interactions between loci. In our approach, we alleviated such problems by adopting a relatively liberal p-value cut-off in the initial step of the algorithm. To compensate for this choice, we augmented genome-wide associations with putative paths through a network of molecular interactions. This step allowed us to filter spurious associations and simultaneously uncover other molecules that participate- in the propagation of the perturbation.
Being based on high-throughput interaction data, our approach does not allow us to propose specific molecular mechanisms of signal propagation at this point. Although our method provides an important step forward suggesting potential intermediate nodes for observed associations, uncovered pathways should be considered testable hypotheses rather than ultimate and mechanistic proofs of causal relationships.
The augmentation of associated gene-loci pairs with pathway information resulted in a very powerful strategy, allowing us to not only uncover potential causal genes, but also find intermediate nodes on molecular network paths that mediated information between causal and target genes. Using this method, we also identified functional GO-pathways that mediate many genotype - phenotype associations in GBM. In addition to identifying putative causal genes and dys-regulated functional pathways, our approach provided evidences for the pathway-centric perspective of complex diseases. Firstly, we showed that various genetic perturbations lead to dys-regulation of the same functional pathways. Furthermore, consistent with the hypothesis that genotypic variations dys-regulate whole pathways rather than target individual genes, we found that different sets of target genes sampled from the same pathways lead to uncovering the same causal genotypic variations.
Our method consists of multiple steps of analyses. However, each individual step can be used separately, depending on a specific application. For example, in the first step we selected a set of differentially expressed genes in cancer as target genes. However, this set can be replaced with other user selected set of interest, therefore facilitating targeted studies of particular pathways.
To our best knowledge, our method is the first genome-wide computational approach that reached beyond a simple association analysis. In addition, our method supported genome-wide associations by paths through interaction networks that can, in principle, propagate the information flow from causal genes to target genes. While copy number variation and gene expression data of glioblastoma patients provided an opportunity to test our approach, our method can be applied to any disease system where genetic variations play a fundamental, causal role.
We utilized 158 patient and 32 non-tumor control samples collected from the NCI-sponsored Glioma Molecular Diagnostic Initiative (GMDI) [44], [45] which were profiled using HG-U133 Plus 2.0 arrays. Arrays were normalized at the PM and MM probe level with dChip [44], [46]. Using the average difference model to compute expression values, model-based expression levels were calculated with normalized probe level data. Negative average differences (MM > PM) were set to 0 after log-transforming expression values [44]. Accounting for weak signal intensities, all probesets with more than 10% of zero log-transformed expression values were removed. To represent a gene, we chose the corresponding probeset with the highest mean intensity in the tumor and control samples. Gene expression profiles are available through the Rembrandt database (http://rembrandt.nci.nih.gov/).
All patient and non-tumor control samples were hybridized on the Genechip Human Mapping 100K arrays, and copy numbers were calculated using Affymetrix Copy Number Analysis Tool (CNAT 4). After probe-level normalization and summarization, calculated log2-tranformed ratios were used to estimate raw copy numbers. Using a Gaussian approach, raw SNP profiles were smoothed (>500 kb window by default) and segmented using a Hidden Markov Model approach [45], [47], [48]. Genomic alteration profiles are available through the Rembrandt database (http://rembrandt.nci.nih.gov/).
Considering alterations of copy numbers (CN), we defined an amplification if log2 CN - 1>0.1 and a deletion if log2 CN - 1<−0.1.
We utilized human protein-protein interaction data from large-scale high-throughput screens [49], [50], [51] and several interaction databases [52], [53], [54], [55] totaling 93,178 interactions among 11,691 genes. As a reliable source of experimentally confirmed protein-DNA interactions, we used 6,669 interactions between 2,822 transcription factors and structural genes from the TRED database [56]. As for phosphorylation events between kinases and other proteins we used 5,462 interactions between 1,707 human proteins from the networKIN [57], [58] and phosphoELM database [59]. Pooling all interactions we obtained a network of 11,969 human proteins that are connected by 103,966 links.
We identified genes that are differentially expressed in the disease cases compared to the non-disease controls in each case. Specifically, we normalized gene expression values as a Z-score, utilizing mean and standard deviation of gene expression values in the non-disease control cases. We considered a gene differentially expressed if the normalized gene expression value of the gene had a p-value <0.01 in the given case using a Z-test.
We chose a representative set of target genes by formulating the problem as a minimum multi-set cover. First, we defined a bipartite graph B(T, S) between genes T and disease cases S by adding edges between genes g and cases s if and only if gene g was differentially expressed in case s. We constructed a multi-set cover instance SC = {B(T, S),α, β} where α represented the number of times that a case needed to be covered, and β was the maximum number of outliers. In other words, all but β cases needed to be covered at least α times in the output cover. The problem to choose a minimum number of genes, satisfying the constraints is NP-hard (i.e., computationally not feasible), prompting us to design a greedy algorithm. The pseudocode of the corresponding algorithm is shown in the Text S1. We demanded that a case needed to be covered at least α = 55 times with a maximum of β = 3 outliers, obtaining 74 target genes.
We utilized a set of loci L = {l1, l2,…, lm} where each locus li was characterized by the corresponding copy number cni,j in each case j, CNi = {cni,1, cni,2,…, cni,n}. Since copy numbers of nearby loci tend to be highly correlated we significantly reduced the number of loci by a local clustering. Specifically, for a potential tag locus tlk, we greedily accumulated all consecutive loci, ensuring that the Pearson's correlation coefficient of CNk and CNi at any locus li in the region was > θTL = 0.9. Tag loci and associated regions can be computed in time linear to the number of loci. Note, that adjacent regions may overlap and a gene may belong to more than one region. Given a set of tag loci TL = {tl1, tl2,…, tlm}, we identified candidate causal loci by associating copy number alterations with expression profiles of target genes. Given a set of target genes TG and tag loci TL, we calculated significant associations by a linear regression between the normalized expression values of gene tgi, E(tgi), and copy numbers of tag locus tlj, CN(tlj). For each target gene tgi, included all tag loci with p<0.01. We considered a tag loci tlj associated with tgi if tlj TL(i).The pseudocode for selecting tag loci and eQTL mapping is presented in the Text S1.
The circuit flow algorithm is based on the well-known analogy between random walks and electronic networks where the amount of current entering a node or an edge in the network is proportional to the expected number of times a random walker will visit the node or edge. Let G = (N, E) represent a gene network where N is a set of genes and E is a set of molecular interactions. Let vector I = [I(e) for e ∈ E] denote current passing through the edges, and vector V = [V(n) for n ∈ N] holds variables of voltage at the nodes. For a given tag locus, let C be the set of candidate genes located in its genomic region. Vector X = [X[c] for c ∈ C] denotes the current leaving the candidate genes. For an edge e = (u,v) connecting genes u and v, we calculated the gene expression correlations corr(u, tg) and corr(v, tg) between both genes and target gene tg. We defined the conductance of edge e, w(e) as the mean of corr(u, tg) and corr(v, tg). As such, we ensured that a single non-correlated node reduced but not completely interrupted the current flow, while a cluster of non-correlated nodes put a considerable resistance to the current flow. Ohm's law is defined as(1)where Id is an |E|×|E| identity matrix, and O is a zero matrix. P is an |E| ×|N| matrix and P(e, n) = w(e) if n = v, -w(e) if n = u, and 0 otherwise. Kirchhoff's current law is(2)where Q is an |N| ×|E| matrix, and Q(n, e) = 1 if n = u, −1 if n = v, and 0 otherwise. R is an |N|×|C| matrix where R(n, c) = 1 if n = c, and 0 otherwise. T is an |N|×1 vector where T(n) = 1 if n is the target gene tg, and 0 otherwise.
Finally, we set the voltage of all genes in C to be 0 so that all current flowed into the candidate genes and there is no current flow between candidate genes, defined as(3)where S is a |C|×|N| matrix and S(c, n) = 1 if n = c, and 0 otherwise.
The set-up of such a linear system implicitly considered all interactions undirected and stipulated that each interaction can have a regulatory effect on the expression of a target gene. In order to obtain more biologically meaningful results, we demanded that direct regulation activity on the expression of target genes is mediated by transcription factors. Therefore, we determined paths where target genes interacted with transcription factors only. In addition, we also accounted for directions of protein-DNA interactions and phosphorylation events. Since linear programming approaches to solve such a directed model [21] required extreme computational resources, we implemented a simple heuristic: after solving the linear system, we removed edges that were used in the wrong direction. We repeated this procedure until only a small number of directed edges were used in the wrong direction (see Text S1 for details). We chose a threshold of 100, which was approximately 0.1% of the total number of edges and found that this heuristic provided a reasonable approximation to the linear programming approach.
Since the number of genes located in each region varied from 0 to several hundreds, the amount of current that flows to genes cannot be compared directly among different loci to prioritize genes. Given the results of the circuit flow algorithm, an empirical p-value for each pair of a target and a causal gene was estimated, utilizing 30 random networks. Random networks were generated by swapping edges while preserving node degrees to avoid potential biases toward hub nodes. Assuming that each edge had a unit conductance, we ran the circuit flow algorithm in each random network for the same set of genes and computed the amount of current flowing into each gene located in the tag locus. A normal distribution was fitted to the current values in the random networks, and empirical p-values were computed using a Z-test.
For each locus and a set of genes in the associated region, we only considered genes receiving current of at least 70% of the maximum current among all genes in the region. Utilizing the permutation method, we selected candidate causal genes for each target gene if the empirical, gene specific p<0.05. On average, we found a total of 701 causal genes for all 74 target genes (for details of parameter settings, please see Text S1).
Let region(cg) be the region that contains a causal gene cg. Recall that regions may overlap, and therefore a gene can be part of more than one region. Let regionmax(cg, tg) and tlmax(cg, tg) be the region and tag locus that harbored causal gene cg and have the most significant p-value among all the current flow solutions from a target gene tg to regions in region(cg). Utilizing a current flow solution Sol(tg, tlmax(cg, tg)) from tg to tlmax(cg, tg), we first removed any nodes with empirical p-value >0.05 from the network. Subsequently, we determined a maximum current path from tg to cg which was defined as a simple path P (tg, cg) = (tg, g1, g2,…, cg) such that was maximized where I(gi) was the total current passing through the gene gi (please see Text S1 for algorithmic details). We computed a path for each pair of a final causal gene and a target gene affected by the causal gene.
One of our primary goals was to identify a set of causal genes that explains (almost) all disease cases. Given a set of candidate causal genes and their corresponding copy number variations we identified a subset of common causal genes that explains the disease cases. Specifically, a causal gene cgk explains a case si if (i) the tag locus including the gene has copy number alterations in case si and (ii) there exists a nonempty set of target gene(s), TG(cgk, si), which are affected by cgk (i.e., with P<0.05) and differentially expressed in case si. The weight between a causal gene and a case, w(k,j) is defined as w(k,j) = |TG(cgk, si)|.
A weighted bipartite graph WB(C, S) between a set of candidate causal genes C and disease cases S can be constructed by adding edges between gene cgk and case si if and only if gene cgk explains a case si. For a subset of candidate causal genes C0 and a case s, let W(C0, s) be the total number of target genes covering s by the genes in C0, . We considered a case as explained if the total weight covering the case exceeds a certain threshold. As in the preprocessing in the first step, we wanted to explain all cases (allowing a few outliers) with minimum number of causal genes (Fig. 1D). The problem can be formulated as a variant of minimum weighted multi-set cover problem. Consider an instance WSC = {WB(C, S), γ, δ} where WB(C, S) is a weighted bipartite graph between causal genes C and cases S. We wanted to choose a subset of genes C' from C such that for each case s except δ cases, W(C', s) ≥ γ. Since a very simple version of the multi-set cover problem (unweighted without outliers) is NP-hard, we designed an algorithm, using a greedy approach to choose a subset of causal genes. Repeatedly, we computed the total weight that can be covered by choosing a gene and selected a gene with maximum additional total weight until the constraints are satisfied (See Text S1 for algorithmic details). Recall that target genes were chosen so that each disease case (except 3 cases) had at least 55 target genes in the first step. As some target genes may not cover the same disease case due to the stricter definition in this step, we found that δ = 21 disease cases had less than 50 target genes covering the cases. Therefore, we required an accumulated weight between the set of causal genes and cases W(C', s) ≥ γ = 50 in all but δ = 21 cases and selected 128 final causal genes.
The computationally most expensive component in our algorithm was the circuit flow algorithm. Due to the large size of the human molecular interaction network and the large number of potential causal loci per target gene, the approach required significant computational resources to find a solution to the circuit flow problem and calculate empirical p-values using a permutation method. On average, it took approximately 60-80 hours per target gene to compute solutions for all associated loci (including permutation tests). We used the computing cluster at the NCBI for our computations, allowing us to run several dozens of computations in parallel. In addition, we adapted various optimization techniques to expedite the procedure [60].
|
10.1371/journal.pcbi.1006842 | Analyzing the symmetrical arrangement of structural repeats in proteins with CE-Symm | Many proteins fold into highly regular and repetitive three dimensional structures. The analysis of structural patterns and repeated elements is fundamental to understand protein function and evolution. We present recent improvements to the CE-Symm tool for systematically detecting and analyzing the internal symmetry and structural repeats in proteins. In addition to the accurate detection of internal symmetry, the tool is now capable of i) reporting the type of symmetry, ii) identifying the smallest repeating unit, iii) describing the arrangement of repeats with transformation operations and symmetry axes, and iv) comparing the similarity of all the internal repeats at the residue level. CE-Symm 2.0 helps the user investigate proteins with a robust and intuitive sequence-to-structure analysis, with many applications in protein classification, functional annotation and evolutionary studies. We describe the algorithmic extensions of the method and demonstrate its applications to the study of interesting cases of protein evolution.
| Many protein structures show a great deal of regularity. Even within single polypeptide chains, about 25% of proteins contain self-similar repeating structures, which can be organized in ring-like symmetric arrangements or linear open repeats. The repeats are often related, and thus comparing the sequence and structure of repeats can give an idea as to the early evolutionary history of a protein family. Additionally, the conservation and divergence of repeats can lead to insights about the function of the proteins. This work describes CE-Symm 2.0, a tool for the analysis of protein symmetry. The method automatically detects internal symmetry in protein structures and produces a multiple alignment of structural repeats. The algorithm is able to detect the geometric relationships between the repeats, including cyclic, dihedral, and polyhedral symmetries, translational repeats, and cases where multiple symmetry operators are applicable in a hierarchical manner. These complex relationships can then be visualized in a graphical interface as a complete structure, as a superposition of repeats, or as a multiple alignment of the protein sequence. CE-Symm 2.0 can be systematically used for the automatic detection of internal symmetry in protein structures, or as an interactive tool for the analysis of structural repeats.
| François Jacob described molecular evolution as a “tinkering” process, where pre-existing elements are combined and repurposed to solve new biological problems [1]. Traces of this “tinkerer evolution” can be seen in the widespread reuse of structural elements in proteins at different scales: small motifs [2], functional domains [3], and protein oligomerization [4]. One example is the repetition of structural elements within a protein chain, thought to arise from gene duplication and fusion events [5].
It is common for structural repeats in proteins to maintain a symmetric arrangement [6], which has been associated with many biological functions [7]. The internal symmetry of proteins is thought to arise from ancestral quaternary structures fused into a single polypeptide chain [8–10]. However, since symmetric protein folds theoretically have a folding thermodynamic advantage, their symmetry could also have arisen by evolutionary convergence [11]. On the other hand, the evolution of functional patches is often symmetry breaking [12]. High-quality alignments of structural repeats are essential to resolve these opposing evolutionary explanations and understand the tension between conservation and divergence.
There are a number of computational methods and tools to detect and analyze structural repeats in proteins. Some methods focus on the detection of patterns and periodicities in protein structures and are better suited for the prediction of solenoids repeats [13–16]. Other methods, including the CE-Symm tool presented here, make use of structural alignments and generally perform better in larger regular repeats [6, 17–21]. These two approaches have also been combined to improve the repeat detection performance [22]. In addition, there are tools that use existing libraries of protein structural repeats [23, 24]. The most comprehensive database of known structural repeats is RepeatsDB [25].
Two of the repeat detection methods primarily focus on the detection of internal symmetry. Both SymD [20] and CE-Symm [21] start with a self-alignment of the structure against itself to identify significant self-similarities. The extraction of repeats from the self-alignments, however, is a nontrivial task, so initial versions of both methods were concerned only with internal symmetry detection (binary decision) and estimation of the number of repeats. Here we present an extension of CE-Symm (version 2.0) that, apart from accurately detecting internal symmetry in proteins, defines the repeat boundaries, reports the type of symmetry and describes the arrangement of repeats using symmetry axes. The similarity of the structural repeats can be further compared at the residue level in a multiple structure alignment.
Several definitions of internal symmetry and repeats are possible, depending on the biological question of interest. For the purposes of this paper, we define it as the regular arrangement of a common repeating structural unit within a protein chain. Therefore, a repeat is an asymmetric structural motif present multiple times in the same structure. We restrict our consideration of repeats to cases where the orientation between adjacent structural units is regular; that is, where a consistent geometric transformation can be applied to superimpose each repeat onto the next. In other words, CE-Symm focuses on identifying repeats which conserve both the structure and the interface between repeats.
Several types of internal symmetry can be derived from this broad definition. The most basic division is between closed symmetry and open symmetry. In proteins with closed symmetry, the repeats are arranged in a point group symmetry. This can be defined mathematically as a set of rotations that superimpose equivalent repeats while keeping at least one point at the center of rotation fixed. In contrast, repeats in proteins with open symmetry are related by transformations with a translational component. Examples of closed and open symmetry can be found in Fig 1a–1d and 1e–1h, respectively.
Closed symmetries can be further characterized according to the possible chiral point groups: cyclic (Cn), generated by a single n-fold rotational operator (Fig 1a–1b); dihedral (Dn), which requires an n-fold rotation and n perpendicular 2-fold operators (Fig 1c–1d); and polyhedral point-groups (T, O, and I), which feature non-perpendicular rotation operators. Both cyclic and dihedral internal symmetries are common in proteins, but, although common at the quaternary structure level, polyhedral symmetries have not yet been observed within a single polypeptide chain.
Open symmetry can be further subdivided into special cases of helical, translational, and superhelical repeats. Helical symmetry consists of repeats arranged around a screw axis, where each repeat is related to the next by a fixed linear translation combined by a rotation around the central axis (Fig 1e). In cases where the rotation angle is close to an fraction of a turn, we indicate the approximate number of subunits needed per turn (Hn). Proteins with open symmetry that have negligible translation are called rotational repeats (Fig 1f), and those with negligible rotation between repeats are called translational repeats (Fig 1g), both annotated as R. Superhelical symmetry (SH) provides the most general description of repeats with open symmetry, and is reserved for cases which cannot be expressed as a single fixed operator relating each repeat to the next. Instead, the rotation axis between adjacent repeats precesses along a helical path (Fig 1h). Proteins with open symmetry are sometimes referred to as solenoid proteins [28].
CE-Symm analyzes the symmetry in a protein structure and produces a multiple alignment of all repeats, as well as ancillary information about the type and order of symmetry in the structure. An overview of CE-Symm 2.0 alignment steps is shown in Fig 2. These are described in detail below, but consist of (1) structural self-alignment, (2) order detection, (3) refinement to a multiple alignment, (4) Monte Carlo optimization of the multiple alignment, and (5) point group symmetry detection. These steps are repeated iteratively to detect multiple levels of symmetry (hierarchical symmetry) and higher-order point groups.
CE-Symm begins with a structural self-alignment (other than the identity alignment) of the input protein structure using the Combinatorial Extension (CE) algorithm [29] (Fig 2b and 2c). Identifying significant self-alignments was the primary focus of the first version of the algorithm [21]. In the self-alignment of structures with closed symmetry the first and last repeats are aligned, forming a circular permutation (CP) of the structure. This is why the structure alignment method used in CE-Symm shares algorithmic primitives with CE-CP [30]. For proteins with open symmetry, the initial self-alignment will always be missing one of the repeats due to the translation component of the symmetry operator.
The alignment quality is quantified using TM-Score [31]. Both irregularly arranged repeats and large asymmetric regions in a structure will reduce the score of the self-alignment. In addition, open symmetry will generally have lower scores than closed symmetry, because the terminal repeats are unaligned in the initial self-alignment.
The order of symmetry is defined as the number of symmetric units (repeats) in a structure. Extracting the order of symmetry is a key part of symmetry detection, and subsequent steps of the CE-Symm method depend on its correctness.
Two methods to automatically determine the order of symmetry in closed structures were described in the previous CE-Symm publication [21]: DeltaPosition and RotationAngle. An error in the distance formula was corrected in CE-Symm 2.0 (see S1 Text, Supplemental Methods), but the DeltaPosition method still gives better overall performance and is used by default for closed symmetry.
A third method for order detection which is able to handle open symmetry has been introduced in the new version of the tool and is named GraphComponent. Conceptually, the self-alignment is treated as a directed graph over the set of aligned residues (Fig 2d). Residues that are aligned in all k repeats will form a path with k nodes. For open symmetry these paths tend to be disjoint, so simply finding the most frequent size of the connected components in the graph can accurately determine the order for open symmetry. For well-aligned cases of closed symmetry, the aligned residues form a cycle of k nodes, so the same method can also work in the general case. Those residues which participate in a path or cycle of the most frequent size form the refined alignment discussed in the following section.
For cases of closed symmetry, small alignment errors can lead to a situation where paths of k residues do not form a closed cycle, but rather lead to a different residue at a small offset in the sequence. This can lead to failures of the GraphComponent order detector due to the merging of multiple alignment paths. This case can be handled by the DeltaPosition order detector.
Whether a protein is open or closed can be easily determined by looking for a circular permutation in the self-alignment. CE-Symm uses DeltaPosition for closed cases where a permutation is found and GraphComponent for open cases. This can also be overridden by the user when running CE-Symm.
The refinement procedure takes as input the self-alignment of the structure and the order of symmetry (k) and returns a multiple alignment of the repeats (Fig 2e). CE-Symm has two implementations of the refinement procedure: GraphComponent and DeltaPosition, which are closely related to their respective order detectors.
The GraphComponent refiner combines all connected components of the self-alignment graph with size equal to the order of symmetry. Each connected component contributes one column to the refined alignment of repeats. Care must be taken that the repeat sequences preserve the sequence order of the polypeptide chain; where some pairs of connected components would violate this property, some are discarded in a way that maximizes the total length of the resulting alignment.
The DeltaPosition refiner takes the self-alignment graph and modifies it until all remaining nodes are part of k-cycles. The modification heuristic is described in detail in the Supplemental Methods (S1 Text). These cycles each contribute one column of the multiple alignment of the symmetric repeats, as in the GraphComponent refiner. Note that, like the GraphComponent refiner, the multiple alignments obtained at the end of this stage consist of ungapped columns, so all repeats are of the same size.
The multiple alignment obtained from the refinement is sometimes far from optimal, and depends very much on the quality of the self-alignment. In addition, the refinement process prioritizes precision over coverage, which means that only the best residue equivalencies will be included, resulting in a shorter multiple alignment. The goal of the optimization is to increase the multiple alignment length while keeping the RMSD low (Fig 2f). Furthermore, the optimization procedure can improve parts of the alignment that were not fully represented in the self-alignment, and thus not captured in the refinement result.
The optimization process uses a similar approach to the Combinatorial Extension Monte Carlo (CEMC) multiple structure alignment algorithm [32]. The multiple alignment can be described by a matrix, where the rows represent aligned structures and the columns represent aligned positions (residue equivalencies). Rearranging and modifying the entries of the matrix results in changes of the multiple alignment. There are four possible moves (changes in the multiple alignment):
The insertion of gaps allows for partial repeat similarities in the alignment. All moves take into consideration that rows of the alignment occur sequentially in the protein sequence, so unaligned residues between repeats can be considered either at the end of a repeat or the beginning of the following one. In addition, the shrink and insert gap moves have been biased, so that the probability of choosing an alignment column or an equivalent residue, respectively, is proportional to the average inter-residue distance of the given column or the given residue, respectively. A geometric distribution with parameter 0.5 is chosen to allocate the probability among alignment columns. A schematic representation of the steps and how they affect the multiple alignment is provided in S1 Fig.
After each optimization step, an alignment score is calculated. The score function to be optimized has also been smoothed with respect to the original CEMC score to remove discontinuities:
S = ∑ i = 0 N ∑ j = 0 L[ C1+(dijd0)2−A ]−G (1)
N is the number of structures (rows) in the alignment; L is the number of equivalent positions (columns) in the alignment, including gaps; C is the maximum score of an alignment position (by default set to 20); dij is the average distance from aligned residue j in structure i to all its equivalent residues; d0 is the structural similarity function parameter, as defined by the TM-score [31]; A is the distance cutoff penalization, which shifts the function to negative values when the maximum allowed average distance of an aligned position (dc) is higher than dij; and G is a linear gap penalty term. Calculation of A using a distance cutoff parameter dc (by default set to 7Å) is straightforward from the condition that the score S has to be 0 when dij = dc. The shape of the score function for different values of dc is shown in S2 Fig.
Moves with positive score changes are always accepted. The acceptance probability of a negative scoring move is proportional to the score difference and decreases proportional to the number of optimization steps as follows:
p = [ C - Δ S C m ] ( 1 - m M ) (2)
m is the current iteration number, ΔS is the change in alignment score and M is the maximum number of iterations. The maximum number of optimization iterations is proportional to the length of the protein, by default a hundred times the number of residues in the protein. Optimization finishes either because it reaches the maximum number of iterations or in the case that no moves are accepted for a fraction of the total number of iterations (by default M divided by 50).
So far, the procedure described can only identify symmetry operations that require a single axis. However, some structures present symmetries represented with more than one axis. This is the case for point groups other than cyclic, like dihedral symmetry, or structures with more than one level of symmetry, what we define as hierarchical symmetries. Multiple CE-Symm iterations are run in a recursive manner, i.e. repeats found in previous rounds are recursively fed into the next run until a non-significant result (no symmetry) is found. The goal is to find all the significant symmetry levels of a structure.
At the end of an iteration, repeats are extracted from the internal symmetry result and one of them is chosen as the representative, by default the N-terminal repeat. Results of successive iterations are merged by combining the symmetry axes and multiple alignments, generating a unique result for the query structure.
The recursive symmetry detection allows better order determination for difficult cases (e.g., TIM barrels), because usually fractions of the order of symmetry are initially found (e.g., 2-fold instead of 8-fold). Continuing the analysis recursively breaks the structure down to the true asymmetric repeating units (e.g., with three levels of symmetry: 2-fold, 4-fold and finally 8-fold).
There are three decision checkpoints in the algorithm flowchart in Fig 2. The first significance criterion for a symmetry result is the self-alignment TM-score. Like in the previous version, the default threshold value is set to 0.4. The second significance criterion is the order of symmetry. A symmetric structure must have symmetry order greater than one and the refinement of the self-alignment into a multiple repeat alignment has to be successful. The third significance criterion is the average TM-score of the multiple alignment of repeats, defined as the average TM-score of all pairwise repeat alignments. The default threshold value for the average TM-score is set to 0.36, because a 10% decrease from the original TM-score is allowed after refinement due to the restrictive conditions imposed on it. In addition the number secondary structure elements (SSE) of the final asymmetric repeating unit is considered. If the the number of SSE of each repeat is lower than the threshold, the result will be considered non-significant. For many applications it may be desirable to exclude simple repeat units (e.g. helical bundle proteins), but these are included in CE-Symm analysis by default in order to find the highest possible symmetry in a structure.
The recursive symmetry detection identifies a collection of symmetry axes that describe the arrangement of repeats in the query structure. In many cases, several of these axes can be combined to form higher-order symmetries. For example, a two-fold rotation axis can be combined with another orthogonal axis to form dihedral symmetry. Near-identical rotation axes can also be combined to form higher-order rotational symmetry.
To determine the point group symmetry, we build on the algorithm described by Levy et al. [33]. The symmetry axes can be found efficiently by first considering only the centroids of each repeat, since they must be in a symmetric configuration if the entire complex is symmetric. To find all possible symmetry axes, the centroids are rotated around axes that go through the centroid of the whole structure using an orientation grid search in quaternion space [34]. For each orientation, the RMSD of the aligned centroids is calculated. If the centroids align within a threshold, then all Cα atoms are superimposed. The symmetry axis is then defined by the rotation matrix of this superposition. If the RMSD is less than a threshold value (i.e., 5 Å), the symmetry operation is considered valid. Since symmetry operations form a group, only a few are needed to complete the full point group.
This procedure allows the combination of axes that have been considered separately by CE-Symm. The point group is included in the final symmetry output and displayed to the user as a polyhedron box around the protein structure.
For evaluation purposes we used the manually curated dataset of 1,007 domains selected randomly from the set of SCOP superfamilies, introduced in our previous study [21]. The benchmark is intended to be representative of structural domains in the PDB, and repeats are present in 25.8% of domains. Domains were curated to require reasonably high coverage by repeats with conserved topology, but allowing for structural divergence and flexibility. A small number of classifications were updated to be more consistent with the new symmetry definitions, especially for the cases of open symmetry. The updated version of the internal symmetry dataset (v2.0), together with the reasons of the modified annotations, is summarized in S1 Table. The benchmark dataset and results are available in S2 Table or https://github.com/rcsb/symmetry-benchmark. An important note is that in the evaluation of the previous version the open symmetry cases in the benchmark were part of the asymmetric (negative) set, while they are part of the symmetric (positive) set in this evaluation.
RepeatsDB was used as an additional benchmark of positive cases [25]. At time of download, RepeatsDB contained 3,689 manually reviewed entries (accessed October 18, 2018). Entries consist of single chains, and may contain multiple structural domains or multiple repeat regions. We selected a total of 3,503 chains with repeats of classes III (solenoid repeats) and IV (closed repeats) that were part of the RepeatsDB-lite benchmarking dataset [24]. Chains which were either annotated in RepeatsDB as having multiple repeat regions or in ECOD [35] as having multiple structural domains were considered multidomain chains in the analysis. The RepeatsDB benchmarking dataset and CE-Symm results are listed in S3 Table.
In our previous article we compared the performance of CE-Symm against SymD. The performance in symmetry detection has only been affected by the additional order detection and alignment refinement steps. The ROC curves of both versions are very similar, with a slight reduction of false positives in the new one (S3 Fig). At the default TM-score threshold values for result significance, the false positive (FP) rate has decreased from 5.5% to 2.5%, while the true positive (TP) rate has been reduced from 81% to 76% on the benchmarking dataset. The bottleneck in symmetry detection continues to be finding a significant self-alignment.
The different methods for order detection perform similarly for closed symmetry cases in the benchmark (S4 Table). The simpler GraphComponent method performs worse than the others, but it is the only one that can be used for open symmetries, while the DeltaPosition detector performs better than the RotationAngle method, particularly for difficult cases.
On average for symmetric entries in the benchmark where CE-Symm could find symmetry, the optimization step extended the repeat length by 43%, reduced the RMSD by 1.8% and increased the average TM-Score of the repeat alignment by 19.6%. Furthermore, using optimization an additional 23 cases (9% of the symmetric structures in the benchmark) were correctly identified as symmetric (209 with optimization, 186 without), which is a 12% improvement in symmetry detection. Because the highest scoring alignment of the simulation trajectory is taken as the result, optimization can only improve the initial alignment.
A detailed evaluation of predicted number of repeats by CE-Symm is shown in S3 Fig. Among incorrect predictions, CE-Symm tends to underpredict the number of repeats, typically as a fraction of the true number of repeats (e.g. predicting 4 repeats for a C8 TIM-barrel). Examples with high-order rotational symmetry are also often missed due to the default maximum order of 8 used in the DeltaPosition method. Overall, CE-Symm 2.0 is able to predict the number of repeats correctly in 89% of cases. High order open symmetries remain challenging due to the prevalence of kinks and structural inhomogeneities in large open structures.
The RepeatsDB-lite method was also run on the benchmark for comparison (S5 Fig). RepeatsDB-lite only detects proteins with three or more repeats, so one- and two-repeat cases in the benchmark were binned together for analysis. With this definition, it predicts the correct number of structures for 72% of the benchmark. Since the method is based on a library of known repeat units, it tends to miss repeats in benchmark cases that are not similar to the training dataset. Additionally, the method does not enforce high coverage or symmetric orientations between repeats. This is desirable when identifying candidates that may have repeats, but it leads to a rather high false positive rate (24%) relative to the definitions of repeats used when curating the benchmark. Raw data for the benchmark results is available in S3 Table and at https://github.com/rcsb/symmetry-benchmark.
CE-Symm was run on the RepeatsDB dataset to assess its performance on a larger set of symmetric proteins. It was able to detect repeats in 69% of the dataset of RepeatsDB reviewed entries. Considering only single-domain proteins improves the recall of 77%, indicating that multi-domain proteins are challenging for the method. The database classifies repeat regions according to the Kajava tandem repeat classes [36]. CE-Symm achieves a recall of 64% for single-domain solenoid repeats (class III) and 89% for closed repeats (class IV). Thus, CE-Symm performs best for closed repeats with single-domain input.
The new CE-Symm tool is capable of presenting internal symmetry as a multiple structural alignment of repeats, which enables direct association of sequence and structure and can be used for comparative and evolutionary analyses of protein structures. The superposition used for the alignment is constrained by the axes of symmetry found in the structure, so that equivalent residue positions maintain the symmetric orientation. Symmetry-aware alignments are important to study, for example, binding and active sites at the internal symmetry interface, like calcium binding in βγ-crystallins (Fig 3).
Structural alignment of the repeats can reveal conserved motifs that have persisted since the duplication event. One example is the βγ-crystallin superfamily, which occurs in a variety of repeat arrangements. Many βγ-crystallins contain a calcium binding site motif [37]. As shown in Fig 3A, the calcium binding motif is structurally conserved after a 2-fold rotation around the symmetry axis, and the residue side-chains preserve their orientation. Furthermore, calcium coordinates residues from both repeats, making the two-fold symmetry an essential feature of the binding site.
On the other hand, duplication events allow the appearance of asymmetry by independent sequence and structural divergence of the repeats. An example is the MaoC-like thioesterase/thiol ester dehydrase-isomerase superfamily (SCOP: d.38.1.4). Members of this family fold into a characteristic ‘hot dog fold’ which binds coenzyme A and catalyzes the dehydration of various bound fatty acids. Typically the MaoC-like proteins contain one hot dog domain per chain and assemble into dimers, tetramers, or hexamers [38]. Some members of the family contain a duplication of the hot dog fold [39], accompanied by the loss of the catalytic motif R/N-####-H in one of the domains, in order to accommodate bulkier substrates which would otherwise not fit in a single domain [38]. The structural divergence of the catalytic site in one of the repeats of the double hot dog subunit can be easily observed with CE-Symm (Fig 4).
Some proteins contain more than one axis of symmetry. In those cases, the axes of symmetry can be collinear, orthogonal or independent to each other. If the axes are collinear, they can be combined into a single axis with higher symmetry order. If the axes are orthogonal, they can be combined into a point group of higher symmetry order.
If the axes are independent to each other, multiple levels of symmetry exist in the structure in a hierarchical organization. This can be an indication of multiple independent duplication events, like in the case of γ-crystallins (Fig 3B), where four repeats are related by two independent 2-fold axes corresponding to two successive duplication events.
Additionally, the internal symmetry axes can also combine with the quaternary symmetry axes. Therefore, internal symmetry can increase the order of symmetry of a protein complex. Returning to the previous MaoC-like protein example, the internal two-fold axis of the double hot dog domain in Fig 4B is orthogonal to the three-fold quaternary symmetry axis, combining for an overall dihedral symmetry. This arrangement is structurally similar to the D3 quaternary symmetry of hexameric single hot dog proteins (e.g. 1YLI). Accounting for internal symmetry when comparing the two protein assemblies is therefore important, because proteins can have a similar overall structure despite their different subunit compositions. It would be misleading to say that the structure of the trimeric and hexameric MaoC-like proteins are substantially different. Another well-know example of similar overall arrangement with different subunit composition are DNA clamps, which promote processivity in DNA replication. In archaea and eukaryotes, the clamp is a trimer, while in bacteria it is a dimer [40]. Furthermore, all DNA clamps have further internal symmetry axes leading to an overall D6 symmetry. As a historical note, the homology between bacterial and eurkaryotic DNA clamps was only acknowledged when the structures were solved and the similarity of their complexes was identified [41].
Furthermore, internal symmetry is important in understanding the stoichiometry of protein assemblies. Uneven stoichiometry assemblies are those with an unbalanced number of each entity type in the complex and occur rarely in the biological environment. It was previously reported that up to 40% of all protein assemblies with uneven stoichiometry in the PDB can be explained by the presence of internal symmetry in one or multiple of the subunits in the complex [42]. One such example is the artificial complex of Bowman-Birk inhibitor from snail medic seeds with bovine trypsin, which has an A2B stoichiometry (Fig 5). Although the complex is asymmetric, considering the internal symmetry of the inhibitor shows that the assembly is structurally comparable to an even A2B2 assembly with C2 overall symmetry. This property has also functional consequences, since the binding of two trypsin proteins symmetrically allows the inhibitor to efficiently induce dimerization and block the peptidase activity. Symmetry is characteristic of biological assemblies and can be considered by methods, like EPPIC, in order to predict the biological assembly in the context of crystal latices [43]. Including internal symmetry in these methods could further improve predictions for some known cases like, for example, uneven stoichiometries.
The majority of proteins have closed symmetry. In the case of quaternary structures, this is expected since homooligomers with open symmetry are disfavored due to their aggregation potential [44]. However, this is not the case for internal symmetry due to the ability for terminal repeats to diverge and avoid undesirable homotypic interactions.
The most general formulation of open repeats in the literature is that of superhelical symmetry, where the repeating unit is simultaneously translated along a helical path (curvature) and rotated around this path (twist) [28]. CE-Symm cannot identify superhelical symmetries, where both curvature and twist are relevant, because of the fundamental limit of the method to find a single symmetry axis (or multiple independent axes). However, we observe that the majority of structures containing tandem repeats that are classified as superhelical in the literature (solenoids) can be approximated with a single axis of symmetry by CE-Symm. They fall in one of the following four conditions: i) the twist is negligible (relative to CE-Symm tolerances); ii) the curvature is negligible; iii) both twist and curvature are negligible; or iv) the twist is much larger than the curvature. In all those cases, CE-Symm can identify the symmetry in the structures and annotate them as helical, translational or open rotational symmetries.
For instance, from the 18 solenoid protein representatives from table 1 in Kobe and Kajava [28], in 10 either the twist or the curvature are reported to be small (helical symmetry applies), in 5 both the twist and curvature are annotated as small (translational symmetry applies), and the remaining 3 structure representatives have irregular twist (asymmetric applies). Although many folds are classified as superhelical, only a small number have regular repeats but do not fit into one of the above categories. Therefore, in practice CE-Symm can also be a good method for identifying, classifying and characterizing solenoid and other repeat proteins with open symmetry. We hypothesize that the low prevalence of actual superhelical symmetry in proteins could be a consequence of the benefit in conserving interfaces between adjacent repeats.
We have extended our internal symmetry analysis tool in order to improve its usability, capabilities and the interpretability of results. In addition to detecting symmetry in protein structures, the tool can identify corresponding residues of the protein from each repeating element and the symmetry operations between them. CE-Symm 2.0 adds broad capabilities for the detection of all types of internal symmetry, providing information about the type and order of symmetry and the repeat boundaries. The alignments between the repeats are eminently useful in identifying conserved and differential features between repeats, and can be applied to understanding protein function and evolution.
The ability to run CE-Symm recursively to detect multiple axes of symmetry allows both higher-order point group symmetries to be identified and non-point group hierarchical symmetries. The simultaneous visualization of this rich information can lead to a better understanding of the structure and provide information about multiple duplication events. One limitation of CE-Symm 2.0 is that it does not yet integrate quaternary symmetry detection into it’s hierarchy. While it is possible to run the program on biological assemblies, it will have poor performance and may mistakenly fail to detect symmetry. Rather, methods specific to quaternary symmetry detection should be integrated with CE-Symm to provide this feature.
CE-Symm has been optimized for finding structures with global symmetry. While it does search for insertions, the length dependence of TM-Score means that structures with large insertions or multi-domain queries may not meet the default score thresholds. For multidomain proteins it may be needed to perform domain decomposition prior to running CE-Symm. Since it is based on CE rigid body alignment, the tool is also unlikely to detect all repeats in structures with conformational changes in some repeats, or with non-sequential rearrangements like circular permutations between repeats. Another limitation is the requirement for a consistent orientation between equivalent repeats. While for some applications preserving a conserved interface between repeats is desirable, there are many cases with large and functionally significant changes in repeat orientation (e.g. in many solenoid proteins).
Determining whether the high prevalence of internal symmetry in protein structures is predominantly a consequence of thermodynamic selection or an indication of the history of protein evolution remains an open question. Here, we have presented examples where internal symmetry is a result of evolution and tied to functional consequences, and how our tool can help researchers in the protein evolution, classification and annotation fields. The CE-Symm source code has been integrated into the BioJava library [45] and is freely available on GitHub. In the future, we would like to integrate CE-Symm into leading bioinformatics resources for protein analysis.
|
10.1371/journal.pntd.0006694 | Isolation of pathogenic Leptospira strains from naturally infected cattle in Uruguay reveals high serovar diversity, and uncovers a relevant risk for human leptospirosis | Leptospirosis is a neglected zoonosis with worldwide distribution. The causative agents are spirochete bacteria of the Leptospira genus, displaying huge diversity of serovars, the identity of which is critical for effective diagnosis and vaccination purposes. Among many other mammalian species, Leptospira infects cattle, eliciting acute signs in calves, and chronic disease in adult animals often leading to abortions. In South America, and including in Uruguay, beef and dairy export are leading sources of national income. Despite the importance of bovine health, food safety, and bovine-related dissemination of leptospirosis to humans, extremely limited information is available as to the identity of Leptospira species and serovars infecting cattle in Uruguay and the South American subcontinent. Here we report a multicentric 3-year study resulting in the isolation and detailed characterization of 40 strains of Leptospira spp. obtained from infected cattle. Combined serologic and molecular typing identified these isolates as L. interrogans serogroup Pomona serovar Kennewicki (20 strains), L. interrogans serogroup Canicola serovar Canicola (1 strain), L. borgpetersenii serogroup Sejroe serovar Hardjo (10 strains) and L. noguchii (9 strains). The latter showed remarkable phenotypic and genetic variability, belonging to 6 distinct serogroups, including 3 that did not react with a large panel of reference serogrouping antisera. Approximately 20% of cattle sampled in the field were found to be shedding pathogenic Leptospira in their urine, uncovering a threat for public health that is being largely neglected. The two L. interrogans serovars that we isolated from cattle displayed identical genetic signatures to those of human isolates that had previously been obtained from leptospirosis patients. This report of local Leptospira strains shall improve diagnostic tools and the understanding of leptospirosis epidemiology in South America. These strains could also be used as new components within bacterin vaccines to protect against the pathogenic Leptospira strains that are actually circulating, a direct measure to reduce the risk of human leptospirosis.
| Several species of the genus Leptospira cause leptospirosis, a disease that is transmitted from animals to humans (zoonosis). Leptospirosis is the most extended zoonosis worldwide, with over a million human cases each year. Leptospira spp. infect a broad range of wildlife and domestic animals, including cattle. In several South American countries beef and dairy exports rank among the most important national income sources, explaining why in Uruguay cattle outnumber human population by a factor of 4. Yet, we did not know which Leptospira species and serovariants (serovars) circulate among Uruguayan cattle. Current serologic diagnostic methods and whole killed-cell vaccination approaches, critically depend on using the proper serovars, which are hugely variable in Leptospira spp. from different regions of the world. Through a multidisciplinary consortium effort, we now report the isolation and typing of 40 strains of pathogenic Leptospira spp. An unexpectedly large variation in terms of species and serovars was found. These data are extremely important: 1- to improve diagnostics by updating the available reference antigen panels; 2- to evaluate the efficacy of novel vaccines; and, 3- to implement efficacious bovine vaccination as a means of reducing the incidence of bovine and human leptospirosis.
| Leptospirosis is a zoonotic disease of worldwide importance caused by pathogenic spirochetes belonging to the genus Leptospira [1]. It affects humans and a broad range of domestic animals and wildlife. In cattle, leptospirosis is an important cause of reproductive failure, including abortions and stillbirths [2]. Infected bovines also constitute an active reservoir for the spread of the zoonotic disease, especially for humans in direct contact with infected animals including veterinarians, abattoir and farm workers, hunters, as well as scientists handling laboratory animals or during fieldwork [3, 4]. Domestic and wild animals are important reservoirs in rural areas, unlike urban settings where rats play a major dissemination role [5, 6]. Human infection with Leptospira spp. results from direct exposure if the source of infection is animal tissue, body fluids or urine, and from indirect exposure if the source is environmental, such as soil or urine-contaminated water. While the disease is endemic in many countries, it often presents as epidemic outbreaks, causing severe, sometimes fatal disease in both humans and animals [7, 8].
Since the first systematic studies in 1960–1970, serologic studies in animals have repeatedly shown high prevalence of exposure to Leptospira in Uruguay, with individual seropositivity in the 25–50% range, and herd prevalence figures of 50–70% [9, 10]. Leptospirosis is considered as a re-emerging bovine disease in Uruguay since 1998 [10], after what stricter epidemiologic surveillance policies have been adopted by governmental agencies. Human leptospirosis has been included into the official list of diseases of mandatory notification. Leptospirosis in Uruguay is endemic, with limited epidemic outbreaks in rural areas. The annual incidence of human leptospirosis is estimated at 15 per 100,000 [11], with precise figures not determined due to under-reporting and extremely scarce systematic studies in southern Latin America of morbidity/mortality burden [7]. The human disease appears to be associated with bovine infection, as well as to rainfalls and floods [11], with recent isolation efforts revealing the presence of three L. interrogans serovars, two L. kirschneri and one L. borgpetersenii [12, 13].
Despite the relevance of bovine leptospirosis as a cause of bovine abortions and infertility in Uruguay, there have been no extensive studies on the actual identities of Leptospira species and serovars obtained from animals in the field. There are currently no repositories of autochthonous isolates available in the public domain, thus constraining vaccine companies to the use of foreign strains as vaccine antigens. Even though Hardjo serovars have been suspected for years to be involved in bovine infection cases [2, 14], to the best of our knowledge only four L. interrogans and two L. borgpetersenii isolates belonging to this serovar have been reported in South America [15–17] obtained in Brazil and Chile. An early study also reported six Hardjo isolates in Argentina, without distinguishing the species [18], and two isolates of L. interrogans Hardjo were also reported, one in sheep from Brazil [19] and one in cattle from Mexico [20]. We now report the first results of a multicentric effort, over the course of 3 years, aimed at isolating pathogenic Leptospira strains in Uruguay, from infected cattle in the field and at abattoirs. A detailed serologic and genetic characterization of such isolates uncovers a larger than expected variety of Leptospira species and serovars. These data will be instrumental for the design of better bacterin vaccines, as well as for improving diagnosis and epidemiologic studies in Uruguay and neighboring South American countries.
Urine and blood sampling from cattle in the field were performed by professional veterinarians, respecting international recommendations for animal welfare, with approval granted by the Ethics Committee for the Use of Animals for Experimentation (Comisión de Etica en el Uso de Animales de Experimentación CEUA), DILAVE, Ministry of Livestock, Agriculture and Fishery (Ministerio de Ganaderia, Agricultura y Pesca MGAP), Uruguay, according to national law #18,611. Permission to take samples for the study was received from the animal owners and the abattoirs.
Forty-eight herds from both dairy and beef farms were sampled in this study, during a 33-month period (Jan 2015-Sep 2017). Private veterinarians who suspected the disease sent the first samples to our laboratory at the Ministry of Livestock, Agriculture and Fishery. Following current protocols in Uruguay, serum samples from 12 animals from each suspected herd, were screened by the microscopic agglutination test (MAT) [21] for preexisting antibodies against Leptospira (S1 Table). Farm selection for subsequent sample collection prioritized those herds with presumptive diagnosis of leptospirosis (MAT titers ≥200 against ≥1 pathogenic Leptospira reference serogroups). Farms with recorded history of abortions, infertility or acute disease, were also prioritized. Selected farms were visited from January 2015 to September 2017, and individual blood and urine samples from 19 animals were collected (aiming for ≥1 seropositive animal with a 95% confidence interval, using a conservative seroprevalence figure of ≥15% on a reference population of 1000 individuals; seroprevalence estimates from background serologic data in Uruguay are actually higher; the number of individual animals to sample was calculated with the software WinEpi http://www.winepi.net). Due to logistic constraints, in a few cases the number of animals per herd was slightly higher, overall sampling a total of 963 individual animals. Individuals to be sampled in each farm were selected according to recorded history when available, prioritizing animals with clinical signs of acute disease (especially calves with rectal temperature ≥ 39.5°C, jaundice and/or hemoglobinuria), previous antibody titers ≥200 by MAT, and/or history of abortions or infertility. If less than 19 animals met the latter criteria, additional animals (heifers or adult cows) from the same herd were included to complete the required number. A questionnaire was distributed to farmers, gathering information about history of leptospirosis and recent vaccination (<12 months) in the farm.
Blood samples were collected by coccygeal venipuncture using 5 mL tubes with clot activator. Sera were then stored at -20°C. Intramuscular administration of diuretics (~150 mg furosemide, Furo R, Ripoll) and thorough genital organ cleansing (wiping with 70% ethanol) preceded urine collection from individual animals. Approximately 60 mL of midstream urine was collected in sterile 120 mL containers (Bioset, Medicplast).
Urine samples (100 μL) were inoculated in the field, immediately or within 2 h of sample collection (for the rationale, see first section of Results), in 5 mL Ellinghausen-McCullough-Johnson-Harris (EMJH) medium (prepared with Leptospira Medium Base EMJH [Gibco] and albumin BovoLep [Bovogen Biologicals PTY Ltd]), supplemented with 100 μg/mL 5-fluorouracil (5-FU; Sigma) [21], and transported at 4°C to the laboratory together with the corresponding blood/serum samples in Vacutainer tubes (Vacutainer, BD-NJ, USA). In the laboratory, two serial 1:50 dilutions were made from the first urine-inoculated tube, in 5 mL EMJH medium supplemented with 5-FU (EMJH/FU), and all three dilutions were incubated at 29°C. The remaining volume of urine samples was conserved at 4°C for subsequent lipL32 gene amplification (see below). Sera were used to determine anti-Leptospira titers by MAT following reported procedures [21]. Routine MAT tests used the national guide of positivity cutoff at titers ≥200. For comparison of reference vs local strains as MAT antigens (S5 Table), sera from animals from which pathogenic Leptospira spp. were isolated (only from those herds with no recent vaccination history) were tested by serial two-fold dilutions [21] starting from 1:100. The local strains used for the latter MATs, were chosen to represent each of the different serogroups identified in this work (IP1506001, IP1605021, IP1611024, IP1611025, IP1512017, IP1703027, IP1711049 and IP1512011, according to the numbering scheme defined in Table 1).
Random samples of urine (vesical puncture) and kidneys were obtained at 22 slaughterhouses that received animals from geographic regions throughout the country. No indications of reproductive failure nor of any other health problems were recorded for slaughtered animals. Due to pipeline logistics at slaughterhouses, kidneys and urine samples did not correspond to the same animal such that individual samples were treated as independent. Urine samples were immediately inoculated in EMJH/FU, according to the same protocol as with field samples. Kidneys were transported in 4°C-refrigerated boxes to the laboratory and processed on arrival, 2–6 hours after sampling. A fragment of approximately 10 g of tissue was placed in a funnel, surface-sterilized by dousing with alcohol and flamed with a Bunsen burner. The tissue was then placed in a sterile stomacher bag and 10 mL of phosphate-buffered saline (PBS) were aseptically added. After breaking the tissue down to a pulp in the stomacher machine, the obtained suspension was allowed to settle for 15 minutes, 250 μL of supernatant were drawn and inoculated in 5 mL EMJH/FU (called tube A). From tube A, 500 μL were transferred to a second 5 mL EMJH/FU tube (tube B), thus obtaining also a 10-fold diluted culture. Finally, a third culture was also prepared from each sample by directly inoculating 5 mL Fletcher medium with a small cylinder of kidney tissue obtained with a Pasteur pipette. All cultures were incubated at 29°C.
In order to define a precise protocol for culture inoculation in the field after urine collection, decreasing numbers of L. borgpetersenii serovar Hardjo strain Sponselee cells, ranging from 107 to 1 bacterium, were incubated in 1 mL filter-sterilized bovine urine. After variable times, 100 μL urine were inoculated in 5 mL EMJH for culture, and bacterial growth weekly monitored under a dark-field microscope.
For isolations, Leptospira cultures were incubated at 29°C and observed under dark-field microscopy weekly for up to 6 months [21]. In case of contamination by other microorganisms, the cultures were filtrated through a 0.22 μm sterile syringe filter (Millipore Corporation, MA, USA) and sub-cultured in fresh EMJH media. As soon as spirochete-like bacteria grew in specific cultures, the presence of pathogenic Leptospira species was assessed by PCR amplification of the lipL32 gene (see below). Once no contamination observed, PCR-confirmed cultures were sub-cultured in EMJH media without 5-FU until exponential growth phase. Leptospira spp. isolates were then conserved at ≥108 cells/mL in EMJH with 2.5% of dimethyl sulfoxide (Sigma) and flash-cooled in liquid nitrogen.
The lipL32 gene was chosen as a marker of pathogenic Leptospira species [22–24]. PCR amplification of lipL32 was performed using purified DNA from 10 mL of bovine urine samples. The urine was centrifuged at 10,000 g for 15 min, the pellet rinsed once with PBS pH 7.4, and total DNA was extracted with the PureLink Genomic DNA MiniKit (Invitrogen). lipL32 PCR-amplification was achieved using oligonucleotide primers lipL32F (5´-ATCTCCGTTGCACTCTTTGC-3´) and lipL32R (5´-ACCATCATCATCATCGTCCA-3´) [25]. The PCR was performed in 50 μL 10 mM Tris.HCl pH 8.4, 50 mM KCl, 1.5 mM MgCl2, 200 μM dNTPs, 0.25 mg/mL bovine serum albumin (Sigma), 2 μM oligonucleotide primers, 1 U Taq DNA polymerase (Invitrogen) and 5 μL template DNA. PCR cycling comprised 1 denaturation step (5 min at 95°C), 35 amplification cycles (each cycle 30 s at 94°C, 30 s at 58°C and 1 min at 72°C) and a final extension step (7 min at 72°C). PCR products were analyzed by agarose gel electrophoresis and ethidium bromide staining, seeking for the expected 474 bp amplicon. Bovine serum albumin (Sigma) was added in the PCR reaction mix, 0.25 mg/mL, greatly reducing sporadic inhibitory effects of certain urine samples on the amplification reaction. An internal control was always included to quantify this potential inhibition issue, by spiking analyzed samples with 40 ng of L. borgpetersenii DNA. Positive amplifications products were randomly chosen in a few field samples, and sequenced confirming specific amplification of Leptospira DNA.
This lipL32 PCR procedure was also performed to rank bacterial cultures (prioritizing more careful follow-ups), after DNA purification from 1 mL of EMJH cultures where suspect spirochetes had been observed by dark-field microscopy.
DNA from Leptospira spp. bovine and human isolates were purified from 1 mL of EMJH culture using the PureLink Genomic DNA MiniKit (Invitrogen). Primers LeptoA (5´- GGCGGCGCGTCTTAAACATG-3´) and LeptoB (5´- TTCCCCCCATTGAGCAAGATT-3´) were used to amplify the 5’-terminal 331 bp fragment of the 16S rRNA gene (rrs) as previously described [26]. The resulting amplicons were sequenced in both senses using internal primers LeptoC (Forward) (5´-CAAGTCAAGCGGAGTAGCA-3´) and Rs4 (Reverse)(5´-TCTTAACTGCTGCCTCCCGT-3´). Sequence quality was verified with the Chromas software, and consensus sequences were defined using BioEdit. All rrs sequences were deposited in GenBank (S2 Table). Consensus sequences were then compared with available sequences in GenBank using BLAST.
Multilocus variable-number tandem repeat (VNTR) analyses were performed according to published methods [27] using five discriminatory markers for VNTR loci 4, 7, 10, Lb4 and Lb5. Purified DNA from each isolate was used to amplify the VNTR4, VNTR7 and VNTR10 loci in L. interrogans, and the VNTR10, VNTRLb4 and VNTRLb5 loci in L. borgpetersenii. The GelAnalyzer 2010a software (http://www.gelanalyzer.com) was used to analyze the ethidium bromide-stained agarose electrophoresis gels, in which PCR products were resolved in parallel to 100-bp DNA ladder (Thermo Scientific) as molecular weight marker. The number of repeats for each VNTR locus was determined as: number of repeats = [PCR product size(bp)—flanking region (bp)] / repeat unit length (bp).
DNA from Leptospira spp. bovine and human isolates were purified from 1 mL of EMJH culture using the PureLink Genomic DNA MiniKit (Invitrogen). The secY gene was partially amplified by PCR with primers SecYF (5´-ATGCCGATCATTTTTGCTTC-3´) and SecYR (5´-CCGTCCCTTAATTTTAGACTTCTTC-3´) as described [28]. The resulting 549 bp amplicon was sequenced in both senses. Sequence quality was verified with the Chromas software, and consensus sequences were defined using BioEdit. All secY sequences were deposited in GenBank (S2 Table) and compared to those available in PubMed, MLST (https://pubmlst.org/leptospira) and PATRIC (https://www.patricbrc.org) [29] databases. The phylogenetic analyses based on secY sequences were performed with MEGA 6.0 software (www.megasoftware.net) using the neighbor-joining method. The evolutionary distances were computed using the Tamura-Nei method and are in the units of the number of base substitutions per site. The reliability of branches was validated by generating 1000 bootstrap replicates. Based on the analysis of sequence similarities, secY genotypes were assigned.
To determine the serogroup of isolated Leptospira strains, MAT was used with a panel of serogroup-specific rabbit antisera, spanning 24 Leptospira serogroups (KIT Royal Tropical Institute, S3 Table), performed in microtiter plates, mixing equal volumes of viable leptospires with serial 2-fold dilutions of each rabbit antiserum. After 2 h incubation at 37°C, agglutination of bacteria was observed under dark-field microscopy. The strain’s serogroup was assigned according to the antiserum that gave highest agglutination titer. Based on the combination of results from both serogroup determination and molecular typing (rrs gene partial sequencing and VNTR analysis), a presumptive serovar was assigned to all isolates belonging to L. interrogans, and L. borgpetersenii species, as previously described [27].
Initial attempts to isolate Leptospira strains from bovine urine samples were unsuccessful. The initial protocol was based on collecting the urine from all sampled animals, and then inoculating them into the tubes with culture media. We asked whether bacterial cell viability could be compromised due to exposure to urine over time. As a first approach to address this issue, the particularly fastidious L. borgpetersenii serovar Hardjo was chosen [30] to perform in vitro tests of viability kinetics in bovine urine. Indeed, a critical maximum time of exposure was defined at less than 2 h (S4 Table), above which subsequent isolation success rates decreased significantly. Although it cannot be ruled out that other serovars might behave differently, based on these observations, all urine samples were inoculated in the field within 2 h of collection, resulting in successful isolations.
A second logistic challenge for isolation efforts from urine samples, was the high number of cultures subject to follow-up under dark-field microscopy. PCR amplification of Leptospira lipL32 gene was optimized on bovine urine, eventually resulting in a robust method to prioritize cultures (Fig 1), identifying those samples that proved positive for pathogenic Leptospira spp. A strong inhibitory effect on lipL32 PCR amplification was frequently observed, dependent on the urine sample (Fig 1A). This sample-dependent inhibition issue was solved by washing the bacterial pellet obtained after urine centrifugation with PBS pH 7.4 (Fig 1B), and then adding bovine serum albumin in the PCR mix (Fig 1C). The sensitivity of this PCR method was ≥100 Leptospira cells, estimated by spiking known amounts of bacteria to sterile urine samples. Specificity was assessed confirming a positive reaction with relevant serovars of pathogenic Leptospira species (L. interrogans, L. noguchii, L. weilii, L. borgpetersenii and L. santarosai), while undetectable with non-pathogenic Leptospira (L. biflexa) nor with unrelated species (Escherichia coli, Pseudomonas aeruginosa, Salmonella sp., Staphylococcus aureus and Enterococcus sp.).
Using this screening strategy, the presence of pathogenic Leptospira spp. DNA was confirmed in 193 urine samples, indicating that at least ~20% (193/963) of all studied animals were excreting pathogenic Leptospira in their urine (Fig 1D and 1E). False positive results from collected samples are highly unlikely, considering that lipL32 is only present in the genomes of pathogenic Leptospira species [22], that no detectable amplification was observed with non-specific bacteria, and that randomly chosen amplicons from bovine urine samples confirmed 100% sequence identity with Leptospira lipL32. An environmental source of pathogenic bacteria during urine sample collection is highly unlikely as well, considering the sample collection procedure and the number of bacteria needed to attain the PCR sensitivity threshold. Following up with this approach at the herd level, 77% of the farms (37/48) that were studied, harbored ≥1 animal(s) excreting pathogenic Leptospira.
The sampling strategies, as detailed in Methods, were chosen to maximize the odds of isolating local strains of pathogenic Leptospira spp. from infected cattle. A two-pronged approach was followed: i- active and directed sampling in the field, at farms with suspicion of Leptospira infection; and, ii- random postmortem sampling of animals at slaughterhouses.
Field sampling. A total of 48 farms representing both beef and dairy cattle herds were visited from January 2015 to September 2017. They were distributed in 12 out of the 19 geographic departments in which the Uruguayan territory is divided. A total of 963 urine samples were collected and subjected to bacterial culture attempts and lipL32 PCR screening. On average, Leptospira growth was detected by dark-field microscopy on cultures after 28 days (range 7–56 days).
Cultures that showed suspect bacteria, were subjected to lipL32 PCR amplification, initially identifying 42 positive cultures from independent urine samples. Considering that 193 urine samples were positive by PCR screening, an estimated recovery rate of 21.7% (42/193) positive cultures from urine samples was achieved. From the original 42 positives, we ultimately obtained 32 pure cultures of Leptospira spp. (Table 1) from field animals, representing a 76.2% rate of success in isolating these bacteria from positive cultures, and a 3.3% global isolation success rate when considering the whole set of input urine samples (32/963). This latter figure should not be taken as a prevalence estimation of animals shedding leptospires (PCR-positive urine samples is a better indicator), since challenges in cultivating these fastidious bacteria are included in the global isolation rate.
Sampling at abattoirs. A total of 288 kidneys and 289 urine samples (representing 577 individual animals) were collected at slaughterhouses. According to the origin of slaughtered animals, all 19 departments of the country were included. 18 positive cultures of Leptospira were identified by dark-field microscopy and PCR amplification (rrs and lipL32 genes), from which 8 isolates were eventually obtained, 3 from urine and 5 from kidney samples (Table 1).
Overall, a total of 40 strains of pathogenic Leptospira were isolated from cattle along the course of this study, and characterized by combining serologic and molecular methods (Table 1). Recalling that initially 60 cultures had proved positive for Leptospira growth, the figures reveal that 20 could not be isolated (10 from field animals and 10 from slaughterhouses), due to overgrowth by contaminant species. Among the 40 characterized strains, 32 were isolated from live animals in the field (30 from cows or heifers, and 2 from calves with signs of acute leptospirosis), and 8 from adult carcasses at abattoirs (Table 1).
The Leptospira species were determined by PCR amplification and partial sequencing of the 16S rRNA gene (rrs). Three different pathogenic species were thus identified (Table 1): L. interrogans (n = 21), L. borgpetersenii (n = 10) and L. noguchii (n = 9).
Serogrouping of isolates was performed by MAT with a collection of 24 rabbit antisera against reference pathogenic serovars. All but one of the L. interrogans isolates corresponded to serogroup Pomona, the different one belonging to serogroup Canicola. The L. borgpetersenii strains all classed within serogroup Sejroe. In contrast, the L. noguchii isolates showed a broader variety of serogroups, including Pyrogenes (n = 1), Australis (n = 1), Autumnalis (n = 4), and 3 L. noguchii isolates that did not agglutinate with any of the reference antisera used.
Taking into account the identification of species and serogroup, together with the VNTR profiles (S1 Fig), it was possible to assign 20 L. interrogans strains to serovar Kennewicki, 1 L. interrogans to serovar Canicola, and the 10 L. borgpetersenii isolates to serovar Hardjo (Table 1). The serovars of the L. noguchii isolates could not be predicted, given that current VNTR profiling tables do not allow yet for serovar assignment of this species.
Twelve L. interrogans, five L. borgpetersenii and one L. noguchii strains, were isolated from farms with no history of vaccination (Table 2). Among such animals, MAT agglutination titers against reference strains were positive in ten cases (considering that national guidelines currently define less than 200 as non-reactive). However, when local isolates were added to the panel of MAT antigens for comparative purposes, 16 out of the 18 sera from non-vaccinated herds showed anti-Leptospira titers against the homologous autochthonous strain that was isolated (S5 Table). These results suggest that including local isolates of Leptospira spp. in the panel of antigens used for MAT may improve the sensitivity of the method. All the isolates recovered from herds with no history of vaccination, belonged to the homologous serogroup as shown by the seroreactivity data (S5 Table).
Genetic analysis of the 501bp secY allele was performed on the 40 typed isolates described in this work. Comparison to other L. interrogans (serovars Pomona and Canicola), L. borgpetersenii (serovar Hardjo) and L. noguchii sequences, obtained from other geographical regions and available in public databases, allowed to build a picture of related groups. Also included in this analysis were secY sequences obtained from 4 Leptospira strains recently isolated from human infections in Uruguay by one of the groups of our consortium [12, 13]. Such human isolates correspond to L. interrogans, L. kirschneri and L. borgpetersenii species. The dendrogram of partial secY sequence clustering, uncovered four phylogenetic clades that corresponded to genomospecies identified by partial rrs gene sequencing: L. interrogans, L. borgpetersenii, L. kirschneri and L. noguchii (Fig 2). The same 4-clades scenario emerged by calculating phylogeny with rrs gene sequences (S2 Fig). Only one homogeneous cluster was observed for the L. interrogans secY sequences, indicating that bovine isolates from Uruguay belonging to this species have close homology with isolates from South America (mainly from Brazil and Argentina) [31]. It is worth noting that two L. interrogans strains that had recently been isolated from human leptospirosis cases in Uruguay affecting rural workers [12, 13] clustered in the same secY clade together with the L. interrogans bovine isolates that we now describe. Concerning the L. borgpetersenii bovine strains, they also clustered with L. borgpetersenii serogroup Sejroe isolates from human and bovine sources in South America, Australia and USA; however, they showed no homology with the uruguayan L. borgpetersenii human isolate, which belongs to serogroup Ballum (F Schelotto, personal communication). Contrasting with such homogeneous clustering of L. interrogans and L. borgpetersenii strains, secY sequence analysis of the L. noguchii isolates revealed a substantially broader diversity, with isolates grouped in two distinct clusters. The first included two isolates, from Panama and Peru. The second cluster, with slight heterogeneity within, comprised all the L. noguchii isolates we are now reporting from Uruguay, as well as a number of other strains obtained from both human and animal origin in several countries of the American continent (Brazil, Nicaragua, Peru, Trinidad & Tobago, USA). Worth highlighting, the secY sequences of our bovine isolates IP1611024, IP1708035 and IP1709037, are identical to some of the L. noguchii strains recently reported in Brazil, isolated from cattle [32] and humans [33].
We are now reporting the isolation and typing of 40 native strains of pathogenic Leptospira spp. from infected cattle in Uruguay. This is the first systematic effort to isolate and type autochthonous Leptospira strains from cattle in this country, where bovine leptospirosis is a major concern as a cause of abortions and zoonotic dissemination. L. interrogans serovar Kennewicki (serogroup Pomona), our most frequent bovine isolate, has actually been also recovered from human patients with leptospirosis in Uruguay [12]. To further confirm this potential link between cattle and humans, we have now shown that the secY genotypes of both L. interrogans Kennewicki and Canicola serovars, are identical in Leptospira strains isolated from patients (rural workers) and from cattle (Fig 2), strongly suggesting that the latter disseminate the infection to exposed humans.
The successful culture of leptospires from bovine samples has likely been boosted by optimizing field sampling protocols, especially after quantifying time-dependent Leptospira viability in bovine urine. PCR screening has also been instrumental in prioritizing cultures, the number of which increased dramatically due to the systematic use of three culture dilutions per animal, themselves important to improve purity in some cases.
A total of 963 urine samples that were processed, eventually produced 42 positive cultures. Among these 42, 9 had produced negative PCR results at the time of urine sample screening. Two different scenarios explain such discrepancies: 8 of the 9 negative results, appeared early during our studies, and eventually proved to be the consequence of urine inhibition, triggering the optimization of our protocols (see Methods and Fig 1). Only in one sample we can strongly suggest that it is the PCR method’s sensitivity that explains the divergent result. In sum, lipL32 PCR screening is an instrumental strategy to prioritize culture follow-ups, albeit not leading to discarding ongoing cultures. We are now optimizing a more sensitive real-time PCR approach, anticipated to also being more robust for screening purposes.
Regarding important, and frequently neglected factors that can lead to success or failure in nation-wide efforts based on field sampling, it is worth highlighting the voluntary participation of farmers and private veterinarians. Early arrangements ensuring for such implications were critical logistic factors for a swift sample collection strategy and for gathering useful information about herds and individual animals. Serial dilutions of the biologic samples on separate culture tubes were successfully used as a means to tackle contamination issues. Most of the positive cultures were successfully purified using the first two dilutions A and B, roughly 50% success from each one. Further diluting the inocula (tube C) allowed the recovery/purification of only 4 additional isolates. Overall, EMJH media outperformed Fletcher in our hands, with only two isolates grown from the latter that were also obtained with EMJH.
Combined serologic and molecular approaches revealed the presence of three different Leptospira species. Besides the anticipated L. interrogans and L. borgpetersenii species, known to be major infectious agents in cattle [2, 34], an important number of isolates corresponded to L. noguchii, both from field samples as well as from abattoirs. L. noguchii has been isolated from cattle in South America [14, 32, 33], but had never been reported in Uruguay, and extremely limited information is currently available about its epidemiologic importance. Are L. noguchii strains a relevant cause of acute disease or reproductive problems in cattle? One of the two strains that we have isolated from calves with signs of acute leptospirosis, was actually identified as L. noguchii, but more information is urgently needed in order to establish the contribution of this unanticipated species in the burden of veterinarian and human leptospirosis in South America. The other strain infecting a suspected acute case was confirmed as L. interrogans serogroup Canicola serovar Canicola, a highly virulent variant often isolated from dogs. Serovar Canicola is however not considered to be adapted to cattle, although it has been reported to infect bovine hosts incidentally, including recent reports in Brazil [35]. It is interesting to note that the isolates belonging to L. interrogans and L. borgpetersenii, displayed limited variation. The latter revealed a single VNTR profile (consistent with a single serovar, Hardjo, within the Sejroe serogroup), also coherent with a unique secY genotype (B). As for the L. interrogans strains, once again quite homogeneous features were found for all isolates, with 20 out of 21 compatible with serovar Kennewicki (serogroup Pomona), and displaying a single secY genotype (A). Only one L. interrogans was different, VNTR clearly matching the one expected for serovar Canicola (in line with Canicola serogroup sero-agglutination), yet sharing the same secY genotype A as the Pomona Kennewicki strains. In stark contrast, the 9 L. noguchii isolates uncovered an unexpected variety of serogroups. We have not yet assigned serovar types to these L. noguchii strains, given that the VNTR multilocus analysis scheme has not been validated for this Leptospira species on the basis of cross-agglutinin absorption tests (CAAT) with serovar-specific antisera. We are currently sequencing the whole genomes for all isolates and actively pursuing direct serovar identification by CAAT for the L. noguchii strains. However, it can immediately be recognized that all nine L. noguchii strains likely correspond to 9 distinct serovars, combining the information of serogrouping and secY genotypes. Three of them did not agglutinate with any of the reference antisera tested, which span 24 serogroups that cover major pathogenic Leptospira [36]. The other six corresponded to serogroups Pyrogenes, Australis and Autumnalis, the latter including four different isolates, all of which differed in secY genotypes (D, F, G and H). The three L. noguchii isolates that did not react with serogroup-specific reference antisera, revealed as yet three additional secY genotypes (C, F and I), hence likely pertaining to three disparate serovars as well.
Serogroup Pomona is one of the most common variants isolated from animals worldwide [37]. This serogroup displays important genetic diversity, as revealed by restriction endonuclease analysis (REA) [38], even within serovars. However, the REA-based genetic profiles of Pomona serovar Kennewicki, show high stability among isolates from a single outbreak [39] and, interestingly, a strong correlation between specific hosts and corresponding REA profile. Those results are consistent with our study: analyzed by secY allele genotyping, a high homogeneity was observed in all Pomona Kennewicki isolates from cattle, despite the broad geographic distribution of the isolates, including those obtained in the field and from slaughterhouses. Serovar Kennewicki is recognized as an animal pathogen [40], apparently adapted to pigs as maintenance host. Even though in Uruguay domestic pigs are not usually raised together with cattle, a forbidden practice in dairy farms, we should not rule out wild boars or other wild animals as potential hosts for this serovar, nor an endemic cycle in domestic cattle [2].
More information is needed to evaluate the prevalence of the serovars we have isolated in the whole country, and neighboring ones in South America. Furthermore, the virulence of these strains in relevant leptospirosis models will be important evidence that must be investigated, regarding pathogenicity (e.g. mortality in the hamster model) and renal colonization (e.g. in the bovine host). It is worth highlighting that we have isolated similar Leptospira species and sero-variants from chronic and acute cases in the field, as well as from dead animals from abattoirs, suggesting they represent a genuine sampling of the true population distribution of infectious Leptospira spp. in cattle. To be conclusive, an epidemiologic study with national geographic coverage is a necessary next step, as well as an in-depth molecular analysis of the Leptospira DNA recovered from PCR-positive urine samples that did not result in positive cultures.
At the individual animal level, and only considering herds with no recent history of vaccination (18 cases), the MAT technique correctly predicted the serogroup (Pomona) of 9 out of the 12 animals where L. interrogans strains were isolated (Table 2). In contrast, none of the 5 cases with L. borgpetersenii infections, nor the one from which a L. noguchii strain was isolated, presented detectable antibody titers using the diagnostic panel of reference available at the national diagnostics laboratory (DILAVE, MGAP). This is likely due to low sensitivity of the MAT, a known issue when it comes to host-acclimated serovars such as Hardjo in cattle [41]. The MAT did not identify any of the L. noguchii isolates, as these were not included within the reference antigen panel in the national diagnostics laboratories (DILAVE, Ministry of Livestock, Agriculture and Fishery). This finding is important, as L. noguchii is a recognized pathogenic species for animals and humans [33, 42]. However, when autochthonous L. interrogans serogroup Pomona, L. borgpetersenii serogroup Sejroe and representative serogroups of the L. noguchii strains were included for anti-Leptospira antibodies titration by MAT, we did observe an increase of sensitivity: analyzing those herds with no history of recent vaccination, all the animals from which L. borgpetersenii strains were isolated showed reactivity against the local isolate, as it was also the case for an animal from which L. noguchii serogroup Pyrogenes was isolated (S5 Table).
As a consequence of this study, the inclusion of these native strains among the antigens for MAT diagnostics and seroprevalence epidemiologic studies, must be an immediate action. Such policies will be important to increase MAT-based diagnostics sensitivity and accuracy [43], and to improve the estimations of prevalence and incidence of bovine leptospirosis infection in the country. Furthermore, isolation and characterization of circulating Leptospira strains, are ongoing activities as a result of our multicentric consortium efforts. We anticipate that new variants and/or species may be discovered, achieving a more complete understanding of current diversity of Leptospira in South America.
A recent study of bovine Leptospira spp. isolates obtained from animals in slaughterhouses in Brazil, shows an important diversity in terms of species and serovars [14]. Libonati et al. report two L. interrogans strains belonging to serogroup Sejroe, and four different serogroups assigned to each of the other two L. santarosai and L. noguchii species identified. Our results now demonstrate a similar diversity of bovine isolates in terms of species and serovars. We have isolated L. borgpetersenii serogroup Sejroe strains, although so far, no L. santarosai isolates nor L. interrogans serogroup Sejroe have been recovered. Instead, we did isolate several strains of L. interrogans serogroup Pomona (presumptive serovar Kennewicki) and one Canicola (presumptive serovar Canicola). With regards to L. noguchii, the broad range of serogroups that we have detected seems to be a shared scenario with the situation in Brazil, with Autumnalis, Australis and Pyrogenes identified in both countries (additionally, serogroup Panama has also been identified in Brazil [32]). However, three L. noguchii isolates could not be classified in any serogroup, failing to agglutinate with the broad panel of reference antisera that was used. These results were confirmed in three different laboratories within our consortium, including the Paris center (WHO Collaborating Center and French reference laboratory for leptospirosis). In any case, these novel serogroups are distinct from the L. noguchii strains so far isolated in Brazil.
It does not escape our attention that most of the serovars that we are now reporting, are not included in the vaccines currently available to the farmers. Except for L. borgpetersenii serovar Hardjo and L. interrogans serovar Canicola, to the best of our knowledge neither serovar Kennewicki (L. interrogans) nor any of the L. noguchii serogroups/serovars that we identified, are being included in bacterin formulations that different companies produce and commercialize as bovine vaccines in South America (Table 2). Bacterins confer little or no cross-protection between serovars, hence the serovars that actually circulate in each region should be included to aim for efficacious vaccines [34]. Indeed, in our study we have obtained several isolates from one herd before and after vaccination. We will now perform closer analyses of naturally exposed herds, following up the effects of vaccination at the individual level. That current vaccines might have shifted the serovar profile of currently circulating Leptospira strains in Uruguay, is a plausible scenario. Proper bacterin vaccination should result in herd protection. We should have thus observed lower isolation rates from vaccinated herds, but we have not. Urine shedding of leptospires can be effectively controlled or significantly reduced in livestock, by using the correct bacterin formulations, according to recent studies with naturally exposed sheep herds [44] or with experimental vaccination/challenge approaches in cattle [45]. Significant reduction in bovine renal colonization and bacterial urinary shedding are achieved by vaccination with bacterins that include the infectious serovars [46], ultimately controlling endemic cycles of infection. Moreover, a systematic vaccination and surveillance program for pig and cattle leptospirosis in New Zealand, demonstrated a correlative dramatic decrease in the incidence, not only of the animal disease, but also of human leptospirosis [47]. Nevertheless, further research is needed to obtain long-lasting vaccination effects and complete protection against bacterial infection. Likely a protective cellular immune response is needed in the cattle model [46, 48, 49] to generate a highly efficacious vaccine against leptospirosis, and not only the humoral response triggered by killed-cell bacterins. The latter are also known to trigger a biased response towards the serovar-specific bacterial lipopolysaccharide antigen, T-independent with lack of memory response [50].
A more thorough understanding of leptospirosis epidemiology, including maintenance hosts and impact in livestock production, is essential to understand and design effective control strategies for this zoonosis. Efficacy studies with currently available vaccines for bovine leptospirosis in our region are also urgently needed. The assembly of this multicentric consortium (S1 Text) gathering the complementary expertise of several key research and governmental institutions in Uruguay, has made possible to obtain the first repository of Leptospira isolates in the public domain, most of them already typed in terms of species, serogroup and serovar. This is a major milestone in the way of controlling leptospirosis in Uruguay, with the associated far-reaching aim of reducing the risk for the human population.
|
10.1371/journal.pgen.1003142 | Dynamic and Differential Regulation of Stem Cell Factor FoxD3 in the Neural Crest Is Encrypted in the Genome | The critical stem cell transcription factor FoxD3 is expressed by the premigratory and migrating neural crest, an embryonic stem cell population that forms diverse derivatives. Despite its important role in development and stem cell biology, little is known about what mediates FoxD3 activity in these cells. We have uncovered two FoxD3 enhancers, NC1 and NC2, that drive reporter expression in spatially and temporally distinct manners. Whereas NC1 activity recapitulates initial FoxD3 expression in the cranial neural crest, NC2 activity recapitulates initial FoxD3 expression at vagal/trunk levels while appearing only later in migrating cranial crest. Detailed mutational analysis, in vivo chromatin immunoprecipitation, and morpholino knock-downs reveal that transcription factors Pax7 and Msx1/2 cooperate with the neural crest specifier gene, Ets1, to bind to the cranial NC1 regulatory element. However, at vagal/trunk levels, they function together with the neural plate border gene, Zic1, which directly binds to the NC2 enhancer. These results reveal dynamic and differential regulation of FoxD3 in distinct neural crest subpopulations, suggesting that heterogeneity is encrypted at the regulatory level. Isolation of neural crest enhancers not only allows establishment of direct regulatory connections underlying neural crest formation, but also provides valuable tools for tissue specific manipulation and investigation of neural crest cell identity in amniotes.
| FoxD3 is an important stem cell factor expressed in many types of embryonic cells including neural crest cells. In the embryo, neural crest cells are a type of stem cell that forms diverse derivatives, including nerve cells, pigment cells, and facial structures. To better understand neural crest development and differentiation, we have explored how FoxD3 expression is regulated in these cells. By examining non-coding DNA, we have identified distinct genomic regions that mediate expression of green fluorescent protein (GFP) in a pattern that recapitulates FoxD3 expression. Interestingly, we find two genomic “on–off” switches or enhancers, called NC1 and NC2, that drive GFP expression in a pattern that recapitulates FoxD3 expression at different times and places during neural crest development. We find that Pax and Msx proteins turn on both NC1 and NC2 enhancers by directly binding to them. In addition, cranial expression driven by NC1 requires a protein called Ets1, whereas trunk expression of NC2 requires a different protein called Zic1. The results show that FoxD3 in differentially regulated in distinct neural crest cell populations in a manner that is specifically encoded in the genome. These enhancers provide valuable tools for understanding neural crest development in birds and mammals.
| The neural crest (NC) is a transient population of cells that migrates throughout the embryo and forms many different cell types, including neurons and glia of the peripheral and enteric nervous systems, bone and cartilage of the craniofacial skeleton and melanocytes [1], [2]. Induction of the neural crest is thought to involve a number of growth factors, including Wnts and BMPs, that establish the neural plate border region that contains the prospective neural crest. This region is characterized by the collective expression of a number of transcription factors, including Msx1/2, Pax3/7 and Zic1, termed neural plate border genes [3]. Subsequently, as neurulation progresses, additional transcription factors are expressed by neural crest precursors residing within the neural folds and dorsal neural tube. These transcription factors, termed neural crest specifier genes, include Sox9, FoxD3, Ets1, Snail1/2 and Sox10, amongst others [2]. Regulatory interactions between neural plate border genes, neural crest specifier genes and signaling inputs generate a complex gene regulatory network (GRN) that orchestrates essential steps in neural crest ontogeny, including emigration from the neural tube, migration to appropriate locations and differentiation into many different cell types.
An important challenge is to establish direct connections within the neural crest GRN. For example, the neural plate border marker, Pax7, is essential for expression of a number of different neural crest specifier genes [4] such that its loss-of-function results in the subsequent loss of Sox10 and Snail2 in the cranial neural crest. Thus, these genes act downstream of Pax7, either by direct or indirect interactions. For the case of Sox10, regulatory analysis revealed direct inputs from Sox9, Ets1 and Myb, but not Pax7 [5], suggesting that effects of loss of Pax7 on Sox10 expression are likely to be indirect. This raised the question of what genes might be direct targets of neural plate border genes like Pax7.
Of the neural crest specifier genes, FoxD3 is one of the first markers of premigratory neural crest in many vertebrate species including mouse, chick, Xenopus and zebrafish [6]–[12]. Its initial expression in the neural tube precedes that of Sox10 and several pieces of evidence suggest that FoxD3 is critical for initiating a cascade of neural crest gene expression that controls their emigration from the neural tube. For example, ectopic expression of FoxD3 in the chick neural tube induces expression of neural crest markers and increases emigration from the neural tube [13]. Similarly in Xenopus ectopic expression at the 8-cell stage increases the expression of neural crest markers, while expression of dominant-negative FoxD3 reduces expression of genes like Snail2, Twist and Ets1 [11] and depletes some neural crest derivatives [9], [11], [14], [15].
Despite its important role both in stem and neural crest cells, no regulatory element(s) controlling the onset of FoxD3 expression are known. To define linkages and assess direct regulatory interactions in the neural crest gene regulatory network with particular interest in possible targets of Pax7, we set out to dissect the cis-regulatory regions of the critical neural crest gene, FoxD3. Taking advantage of the chick's compact genome and ability to assay putative regulatory regions by electroporation, we have identified two enhancers, NC1 and NC2, that mediate reporter expression in spatially and temporally distinct manners in the chick embryo, and in combination closely recapitulate the endogenous expression of FoxD3. Detailed regulatory analysis shows that initial expression of FoxD3 in both cranial and trunk neural crest requires direct input from neural plate border genes, Pax7 and Msx1/2. These factors function in combination with the neural crest specifier gene, Ets1, to bind to the cranial NC1 regulatory element. However, at vagal/trunk levels, they function together with the neural plate border gene Zic1 to activate the NC2 enhancer. These results not only reveal region-specific enhancer activity in the neural crest, but also expand the neural crest GRN and inform upon direct interactions therein. Conserved between mouse and chick, these enhancers further provide excellent tools for assaying gene regulation and manipulation of neural crest gene expression in amniotes.
In the cranial neural tube, expression of FoxD3 initiates in premigratory neural crest cells at HH8- (Figure 1A), with strong and rapid onset of expression that precedes that of Sox10 or Snail2. At this stage, the FoxD3 expression domain includes the neural folds of the forebrain and midbrain. Subsequently, at HH8+, FoxD3 expression expands posteriorly to the hindbrain (Figure 1B). As neural crest cells delaminate at HH9 and migrate at stage 10, FoxD3 is maintained or expressed de novo at high levels by many migrating cranial crest cells (Figure 1C, 1D). The domain of expression of FoxD3 at this stage includes premigratory and migratory crest extending from the midbrain to the trunk, with the exception of the neural folds at the level of rhombomere 3 (dotted arrow in Figure 1C). Expression persists through subsequent stages as the neural crest advances to surround the optic vesicle and populate the first branchial arch.
At vagal and trunk levels, FoxD3 is also expressed by premigratory and migrating neural crest cells. FoxD3 transcript expression initiates as the neural folds appose in the midline at stage HH9. At HH12, FoxD3 is detected in migrating vagal crest as the first wave of cells leaves the neural tube (white arrow in Figure 1I) and with time, FoxD3 expression initiates at progressively more caudal levels in the trunk neural tube and migrating neural crest. At later stages, FoxD3 expression is maintained in a subset of neural crest derivatives, including peripheral ganglia [16].
Double fluorescent in situ hybridization for FoxD3 and Sox10 reveals differences in the expression domains of these neural crest specifier genes. FoxD3 transcripts are detected in the premigratory population prior to Sox10, which is expressed in cranial, vagal and trunk neural crest cells only as cells leave the neural tube (arrows on Figure 1E and 1F). Therefore, even though FoxD3 and Sox10 both have been placed in the same hierarchical level in the neural crest GRN, they are recruited at different time points during neural crest specification.
The genomic region of FoxD3 was examined for conservation across multiple vertebrate taxa including chick, mouse, human, opossum, Xenopus and zebrafish using the UCSC Genome Browser and ECR Browser. The region analyzed spanned 160 kb between the genes immediately up and downstream of FoxD3, Atg4C and Alg6 respectively (Figure 1O). To test putative enhancers for neural crest regulatory activity, eighteen conserved regions varying in size from 1 kb to 4 kb were cloned into an eGFP reporter vector [17] and electroporated into the entire epiblast of stage HH4 or dorsal neural tube of HH8–14 chick embryos, together with pCI-H2B-RFP as a ubiquitous tracer to verify efficacy of transfection.
By testing conserved regions within the FoxD3 locus, we found two enhancers that drive specific expression of eGFP in the neural crest, in a manner that collectively closely recapitulates the endogenous pattern of FoxD3 expression. Enhancer NC1 directed expression of eGFP in the premigratory cranial neural crest analogous to the early endogenous expression of FoxD3 (Figure 1G–1H). eGFP in the cranial neural folds was detected from stage HH8+ (Figure 1G), in the dorsal neural tube and on a few neural crest cells during emigration (Figure 1H, 1J, 1L), lasting until approximately stage HH14, at which time only very weak eGFP expression could be detected. While migrating neural crest from the midbrain to rhombomere (R) 2 exhibited NC1 mediated eGFP expression (Figure 1J), no eGFP expression was observed caudal to R3. This contrasts with the endogenous FoxD3 pattern, which is observed in R4, R6 and more posterior crest (Figure 1I).
Enhancer NC2, in contrast to NC1, mediated strong eGFP expression in the premigratory, delaminating and migrating neural crest at and caudal to R6 (Figure 1K), beginning at HH9. This expression pattern of eGFP in the vagal and trunk neural crest recapitulates endogenous expression of FoxD3 (Figure 1I) at this axial level. The expression of both eGFP and endogenous FoxD3 mRNA extends to the premigratory/delaminating crest at the level of the 4th most caudal somite. Interestingly NC2 activity also controlled eGFP expression in a large subpopulation of migrating cranial neural crest at the level of the midbrain, R1 and R2, which was detectable only after stage HH9+, and expression in premigratory and migratory NC from R4.
To examine the activity of NC2 enhancer at later stages and in neural crest derivatives, we electroporated stage HH8–14 embryos in ovo, and fixed the embryos after 24–48 h (HH15–20). FoxD3 is expressed in most premigratory and migratory vagal and trunk neural crest, but is down-regulated in melanoblasts (Figure S1A), which migrate underneath the ectoderm and initiate emigration approximately 24 h after the emigration of ganglionic neural crest in the chick. Interestingly, we observed expression of eGFP in melanoblasts prior to and during migration (Figure S1B), in addition to expression in the dorsal root and trigeminal ganglia. To confirm that this expression was due to activity of the enhancer and not stability of eGFP, we performed in situ hybridization for eGFP and detected mRNA for eGFP in melanoblasts and dorsal root ganglia (Figure S1D–S1E), suggesting that eGFP is indeed ectopically expressed by melanoblasts, under control of the NC2 enhancer. Expression of eGFP was also seen in neural crest cells migrating along the enteric neural crest pathway (Figure S1F). In contrast to NC2, at HH14 very weak expression of NC1 activity was confined to the branchial arches whereas no expression was observed in cranial ganglia.
We next examined overlap of endogenous FoxD3 expression with reporter expression driven by NC1 and NC2 by performing double labeling with eGFP and FoxD3 antibodies. The results show that NC1-driven eGFP expression completely overlapped with that of endogenous FoxD3 protein in stage HH9 embryos (Figure 2A–2C). Similarly, NC2-driven eGFP expression in migrating cranial neural crest overlapped with endogenous FoxD3 protein expression at stage 11 (Figure 2D–2F) and with FoxD3 in delaminating and migrating crest at the trunk/vagal levels at stage 12 (Figure 2G–2H). The complete overlap between enhancer activity and FoxD3 expression strongly suggests that NC1 and NC2 are the responsible regulatory modules for the control of endogenous FoxD3 in the neural crest.
To determine if orientation of the enhancers affects their activity, NC1 and NC2 enhancers were cloned in reverse orientation into ptk-eGFP and electroporated in HH4 embryos. The results show that both have equivalent ability to drive reporter expression in reversed as in their endogenous orientation, without significant changes in pattern or levels of activity (Figure 2J, 2K).
Finally, we examined whether these enhancers were conserved across amniotes. To this end, we cloned the homologous conserved regions from the mouse genome and mouse (m) NC1 and mNC2 constructs were electroporated into chick embryos at gastrula stages. The results show that the patterns of eGFP expression driven by mNC1 and mNC2 were identical to those observed with chick NC1 and NC2 (Figure 2L, 2M), suggesting that these enhancers are conserved between chick and mouse and likely throughout amniotes.
To examine the dynamic nature and combined activity of the two enhancers in the migrating cranial neural crest, we co-electroporated NC1 (green) and NC2 (blue) enhancers in combination with a previously identified cranial neural crest Sox10E element (red) [5] that expresses in all emigrating and migrating neural crest cells. Reporter expression of multiple fluorophores was then visualized in transverse sections of slices through the midbrain region, using a novel slice culture protocol [18].
Time-lapse movies revealed differential temporal and spatial activity of NC1 and NC2 enhancers. While NC1 activity was present in the premigratory neural crest, the expression it drove in the dorsal neural tube appeared transient in most cells and preceded that driven by the Sox10E enhancer (white arrow in Figure 3A). NC1 activity then recurred in a small subpopulation of actively migrating cranial crest cells that concomitantly displayed Sox10E activity (black arrows in Figure 3B and 3C, Video S1). In contrast, NC2 activity was observed in very few cells within the neural tube (red arrow in Figure 3C), and only a few delaminating neural crest cells coincident with Sox10E activation. Thereafter, NC2 drove expression in a large subset of migrating cranial neural crest cells, which were also positive for Sox10E activity (black arrows in Figure 3C and 3D).
To further investigate neural crest heterogeneity with respect to enhancer-driven expression, we co-electroporated embryos with NC1 (green) and NC2 (red) enhancers and observed neural crest formation and migration by time lapse microscopy. Analysis of the movies suggested that there was little overlap between cells showing activity of NC1 and those with NC2 (Figure 3E–3H, Video S2). Only a few cells co-expressed eGFP and RFP and this may reflect perdurance of the reporter that may be more stable than the endogenous transcription factor. These results suggest that there is highly dynamic regulation of FoxD3 in migrating neural crest cells and suggest that there may be distinct subpopulations that reflect activity of NC2 but not NC1, or vice versa. They further suggest that NC1 activity may be largely responsible for the transient early expression of FoxD3 in the neural tube (arrow in Figure 3F), whereas NC2 activity at cranial levels may be primarily responsible for FoxD3 expression in migrating neural crest cells (arrows in Figure 3G and 3H).
The dynamic expression driven by the enhancer NC1 and its early activation, correlating with the onset of endogenous FoxD3, led us to explore upstream regulators and their binding motifs within NC1, responsible for early activation in the premigratory neural crest. To this end, conservation across vertebrates was used as a guide to identify putative core regions within the enhancer. The central region of NC1 was highly conserved with human, mouse and Xenopus, but showed no sequence conservation with zebrafish. Primers were designed to amplify fragments of NC1, which were tested for activity at stages HH9–10, corresponding to the time it drove strongest expression. Using this approach, NC1 was reduced to 553 bp (NC1.1) without loss of activity (Figure 4A, 4B). A further deletion to 303 bp (NC1.2) resulted in weak eGFP expression specifically in the cranial neural crest (Figure 4A, 4D), suggesting that the regions at the ends of NC1.1 amplify activity of the enhancer, although the critical regions are present within NC1.2. The sequence of NC1.2 was further analyzed by substituting 100 bp regions of sequence with eGFP coding sequence within the NC1.1 fragment. eGFP coding sequence was chosen as a random sequence to substitute for enhancer regions, so that the size and spacing was maintained, but did not alter expression in control experiments. This analysis revealed that 200 bp was required for expression mediated by the enhancer (Figure 4A). We then substituted 20 bp blocks of sequence with eGFP coding sequence across this region within NC1.1. This analysis revealed a region of 80 bp that was critical for detectable expression of eGFP (Figure 4A). An adjacent 92 bp region was required as a unit for eGFP expression; however substitutions of 20 bp blocks within this secondary region weakened but did not eliminate eGFP expression. None of the substitutions resulted in expansion of enhancer-driven expression. The 172 bp fragment (NC1.3) containing the most critical and supportive regions was amplified and electroporated into embryos, and the 80 bp putative core region (NC1.4) was tested by placing two copies in tandem into the ptkeGFP construct. NC1.3 alone drove very weak expression of eGFP in the neural crest. Interestingly, the NC1.4 cancatamer was sufficient to drive eGFP expression in the same pattern as the full-length NC1 enhancer, albeit slightly weaker, suggesting that the 80 bp NC1.4 fragment contains the core elements essential for activity of this enhancer.
Potential transcription factor binding sites within the core region were identified using Rvista and Jaspar databases (Figure S2A). Mutations were made to these sites by substituting 6–8 bp of the core binding site (marked in red or blue in Figure S2A). Mutations to the Ikaros binding site or to the Ets/Zeb binding site did not affect expression of eGFP (Figure S2B and S2C). In contrast mutation of the homeodomain site (Figure S2D and S2E) or Elk/Ets site resulted in loss of eGFP expression. Additionally, mutation of an Msx site (Figure S2A) reduced activity of the enhancer. Pax7, Msx1 and Msx2 are neural plate border genes expressed in the neural folds prior to expression of FoxD3, and candidates for direct activators of FoxD3. All of these can potentially bind to the homeodomain sites. Furthermore, Ets1 is expressed specifically in the cranial neural crest concomitant with the onset of FoxD3 expression.
Similar to the dissection of NC1, we performed a series of deletions and substitutions to identify the core structure critical for activity of the NC2 enhancer (Figure 4H). NC2 is highly conserved in mouse, human, Xenopus and zebrafish. Stepwise deletions revealed a fragment, termed NC2.9, with similar albeit weakened activity to that seen with NC2 in the vagal and trunk neural crest, as well as weak activity in cranial migratory neural crest (Figure 4H, 4I). Subsequently, 100 bp and 30 bp substitutions were made within NC2.9, narrowing the essential regions of the enhancer to approximately 120 bp, encompassing a 90 bp core region surrounded by auxiliary regions required for strong expression (Figure 4H). Several deletions of the NC2 enhancer resulted in eGFP expression in the developing retina (NC2.6, NC2.9 M20) and otic vesicle (NC2.7), and using the full-length NC2 enhancer to drive eGFP, occasional weak expression could also be seen in these structures.
Importantly, deletion of the Zic site within the 90 bp critical core region resulted in complete loss of activity of the enhancer (Table S1). The auxillary (amplifying) region contains Pax, Ets and SoxE sites. Deletion of Pax or SoxE binding sites in the auxiliary region caused loss of NC2 activity in cranial neural crest, but did not affect vagal/trunk NC2 activity (Table S1). Similarly, deletion of an Ets1 site in the auxiliary region (M20), abolished activity in R1–R3 of the cranial neural crest, but did not affect vagal/trunk activity (Table S1). The results suggest that the NC2 enhancer itself is differentially regulated in the cranial neural crest versus trunk neural crest.
We next tested whether the putative transcription factors implicated by enhancer dissections could regulate enhancer driven reporter expression. To this end, individual embryos were electroporated on one side with FITC-conjugated control morpholino plus enhancer directing Cherry expression and with FITC-conjugated blocking morpholino plus enhancer-Cherry on the contralateral side.
For NC1, morpholino-mediated loss of Pax7 (Figure 5B, 5K) protein resulted in significant loss of reporter expression on the target morpholino side (right). Whereas Msx1 knock-down alone resulted in a mild loss of Cherry expression and Msx2 knock-down had almost no phenotype (data not shown), the double MO knock-down exhibited a strong loss of reporter expression (Figure 5C, 5K). Additionally, knock-down of the transcription factor Ets1 resulted in strong loss of NC1 enhancer activity (Figure 5D, 5K). In contrast, morpholinos against other neural crest or neural plate specifiers like Zic1 (Figure 1E), Sox9 or AP-2 failed to alter NC1 reporter expression. These findings support the possibility that Pax7, Msx1/2, and Ets1 are direct inputs into the NC1 enhancer.
To confirm that the loss of Cherry positive cells was not due to loss of neural crest cells on the morpholino-treated side of the embryo, we examined other neural crest markers in embryos in which enhancer-driven Cherry expression was depleted (Figure S3). At the concentration of morpholinos used, we observed little change in Sox9 expression (Figure S3G–S3I), demonstrating that the neural crest population was present in morpholino-treated embryos. Similarly, immunostaining with the HNK-1 antibody at stage HH10 confirmed the persistence of neural crest cells after morpholino treatment (Figure S3J–S3L).
We next examined the effects of knocking down putative regulators on expression driven by the NC2 in the vagal/trunk neural crest. Electroporation of both Pax7 and Msx1/2 morpholinos resulted in loss of NC2 activity in the trunk neural crest (Figure 5G, 5H, 5L) similar to the effects observed for NC1 activity in the cranial crest. In addition, electroporation of the Zic1 morpholino caused strong loss of NC2 activity specifically in the trunk (Figure 5J, 5L), suggesting this transcription factor is a key player in the regulation of trunk expression of FoxD3. On the other hand, Ets1 knock-down had no affect on trunk activity of NC2, which is not surprising given that this transcription factor is not expressed in the posterior neural crest (Figure 5I, 5L). Taken together, these results place Pax7 and Msx1/2 as general regulators of FoxD3 expression, while Ets1 and Zic1 seem to specifically regulate NC1 and NC2, respectively.
To examine the effects of these regulators on endogenous gene expression, we performed morpholino-mediated loss-of-function of Pax7, Ets1, Msx1/2 and Zic1 and examined endogenous FoxD3 expression in newly forming cranial and trunk neural crest cells. Detection of FoxD3 was assessed by hybridization chain reaction (HCR), which reflects transcript levels more accurately than in situ hybridization and at subcellular resolution [19]. We found that morpholino mediated knock-down of Msx1/2, Pax7 and Ets1 caused a significant loss of cranial FoxD3 expression (Figure 6A–6C) at stage HH9, but not Sox9 or HNK-1 expression.
At trunk levels, knock-down of Msx1/2, Pax7 and Zic1 resulted in a significant reduction of endogenous FoxD3 expression (Figure 6D–6F), whereas loss of Ets1 had no effect. The results show that Pax, Msx and Zic transcription factors are not only important for mediating enhancer activity but also for endogenous expression of FoxD3 in the vagal/trunk neural crest.
To demonstrate in vivo association of Pax7, Msx1 and Ets1 transcription factors with the NC1 enhancer, we performed quantitative chromatin immunoprecipitation (ChIP) experiments. Cross-linked chromatin isolated from the midbrain dorsal neural tube of HH8–9 embryos was immunoprecipitated using Pax7, Msx1 and Ets1 antibodies and ChIP-enriched DNA was used in site-specific qPCR, with primers designed to amplify fragments within the NC1 region. For all three factors, we found significant enrichment of the NC1 region amplicon, expressed as a percent of the total input chromatin, compared to IgG controls (Figure 6G–6I). No enrichment was detected in the negative control regions in chromosome 8 (Figure 6G–6I), confirming they are direct inputs into NC1. These data demonstrate that Pax7, Msx1 and Ets1 bind in vivo to the NC1 enhancer element in the cranial neural crest.
Given the striking effects of Zic1 knockdown on NC2 activity in the trunk region, we hypothesized that this transcription factor directly binds this enhancer in vivo. To examine this, we dissected dorsal trunk neural tubes of stage HH12 embryos, crosslinked and immunoprecipitated chromatin with a Zic1 antibody. The results show significant enrichment of the NC2 region amplicon, expressed as a percent of the total input chromatin, compared to IgG (Figure 6J). The results confirm that Zic1 directly associates with the NC2 enhancer in trunk neural crest.
Taken together, these results reveal direct transcriptional regulators of FoxD3 in the neural crest GRN, and highlight the differential regulation of FoxD3 in the cranial and trunk neural crest cells.
As proposed in a putative gene regulatory network [3], FoxD3 is predicted to be downstream of neural plate border specifier genes such as Msx1/2, Pax3/7 and Zic1. Indirect support for this regulatory connection comes from several previous studies. Mice null for Pax3 lack expression of FoxD3 in the neural crest [13]. Knockdown of several genes in Xenopus, including Msx1, Pax3 and Zic1 results in loss of FoxD3 expression in the neural crest [20]–[22]. Similarly, knockdown of these genes and others expressed at the neural plate border in lamprey result in loss of FoxD3 expression [23]. Conversely, misexpression of these genes can induce expression of FoxD3 and other neural crest markers in Xenopus [20]–[22]. However, little was known about direct binding of any of these potential upstream transcription factors to a regulatory region of FoxD3, or the exact placement of these genes in relation to FoxD3 within the neural crest gene regulatory network. Importantly, these studies did not consider that FoxD3 may be differentially regulated at different axial levels.
Our results suggest that expression of FoxD3 is regulated in the avian neural crest by at least two enhancers, which direct expression in largely distinct spatiotemporal domains (head versus vagal/trunk regions), as well as in different subpopulations of the cranial neural crest. The enhancer NC1 is active in premigratory and some migratory cranial neural crest rostral to R3, while enhancer NC2 activity initiates in a single continuous wave caudal to rhombomere 4, including vagal and trunk regions, but also later in a subpopulation of migrating cranial neural crest. In our analysis of the conserved regions within the FoxD3 locus, only these two regions were able to mediate reporter expression in patterns reflecting the distribution of neural crest. The proximity of the NC1 and NC2 enhancers to the FoxD3 coding region, the recapitulation of endogenous FoxD3 expression by the combined activity of the enhancers, and the effect of manipulating upstream regulators on both enhancers and endogenous FoxD3 expression, strongly suggest that NC1 and NC2 act as enhancers regulating endogenous expression of FoxD3 in the neural crest.
Comparison of the activity of these two enhancers with the cranial Sox10 enhancer Sox10E2 [5] using time-lapse imaging demonstrated for the first time that there is dynamic regulation of multiple enhancers within a population of cranial neural crest cells. We observed that the activity of the cranial NC1 enhancer is initially restricted to cells in the dorsal neural tube; only later is it activated de novo in actively migrating cranial neural crest cells, where its activity is preceded by that of Sox10E2. NC2 is active in only a few delaminating/emigrating cranial neural crest cells but in a majority of the migrating neural crest population. Interestingly, there is little overlap of NC1 and NC2 activity in the cranial neural crest, whereas both overlap with Sox10E2, which appears to be active in all of the migrating cranial crest population.
The minimal overlap in activity of NC1 versus NC2 in cranial neural crest populations raises the interesting possibility that there may be a regulatory switch of enhancers from NC1 to NC2 at the endogenous promoter of FoxD3 when the cells reside within the dorsal neural tube and/or are emigrating. Such competition at the promoter could occur if only a single enhancer can be functional at any given time on the FoxD3 promoter. If this is the case, the very few double labeled cells expressing NC1- and NC2-driven reporter expression may represent perdurance of eGFP protein rather than the actual levels of enhancer activity. The finding that NC1 and NC2 enhancers are active in generally separate cranial neural crest populations further suggests that the cranial neural crest represents a heterogeneous population, even as the cells are delaminating from the neural tube, and that this heterogeneity may be encrypted at the regulatory level.
It is intriguing to speculate that the differential activity of NC1 and NC2 in distinct subpopulations may reflect differential cell fate and commitment status of future neural crest derivatives. Consistent with the possibility that NC1 and NC2 activity may reflect commitment to different lineages, NC2 is later active in neural crest-derived dorsal root and trigeminal ganglia, whereas NC1 is active transiently in the branchial arches, but not in peripheral ganglia.
The activity of NC2 in the vagal and trunk neural crest recapitulated expression of endogenous FoxD3 in premigratory and migratory neural crest cells. In addition, FoxD3 is retained by a subset of neural crest derivatives [16]. Consistent with this, conditional knockout of FoxD3 in neural crest cells using the Wnt1-Cre line suggests that FoxD3 is required to maintain neural crest progenitors and that its loss biases their derivatives toward a mesenchymal fate at the expense of neural derivatives [24]. Thus, it appears to regulate the switch between neural/glia and melanocyte lineages [16].
NC2 not only was active in neuronal derivatives, but also directed activity in neural crest cells migrating along the dorsolateral pathway, which are melanocyte precursors that migrate 24 h after the ventrolateral population migrate to the ganglia. Cells on the dorsolateral population do not normally express FoxD3 [7]. Thus, NC2 likely is missing a repressor region for the pigment lineage that is present in the endogenous regulatory region. In fact, ectopic expression of FoxD3 in melanoblasts inhibits their migration onto the dorsolateral pathway, while down-regulation of FoxD3 results in premature dorsolateral migration and increases melanocyte differentiation in cultured neural crest [7]. FoxD3 represses transcription of Mitf, a key transcription factor required for melanocyte development [16], [25]. Our finding of an active enhancer in melanoblasts suggests that FoxD3 is normally repressed in melanoblasts, and this repression does not occur within the NC2 region. In the zebrafish histone deacetylase 1 (hdac1) mutant, a severe loss of mitfa positive melanophores can be rescued by partial reduction of FoxD3; suggesting hdac1 is required to repress FoxD3 in melanophores [25]. It is not yet clear whether this repression is direct or indirect and if it is conserved across species.
The current results establish for the first time a direct regulatory connection between the neural plate border genes, Pax7 and Msx1/2, and FoxD3, suggesting it is an immediate downstream target. This confirms and validates previous indirect evidence in Xenopus, lamprey and mouse, and provides further support for a conserved gene regulatory network in the neural crest. Pax7 and Pax3 are closely related paralogs which have overlapping expression and function [26]. Pax3 and Pax7 bind identical DNA binding domains, and while they show equal affinity for binding to the paired domain, Pax7 shows a higher affinity for the homeobox domain [27]. Both Pax3 and Pax7 are expressed in the developing neural crest, but in overlapping and distinct regions of the neural crest, and these patterns differ between species. In mouse and Xenopus, Pax3 is expressed in premigratory neural crest along the neural axis, and Pax7 is restricted to cranial levels (and very weak in Xenopus) [13], [28], [29]. In chick and zebrafish, Pax7 is expressed throughout the developing crest, and whereas Pax3 expression in neural crest is restricted to trunk levels in chick, in zebrafish it is also seen at cranial levels [4], [30], [31]. Evidence from Xenopus, mouse and lamprey suggests that Pax3 and/or Pax7 is required for FoxD3 expression and neural crest specification [13], [22]. In chick, Pax7 but not Pax3 knock-down at gastrula stages depletes neural crest specifier expression [4]. Mouse Pax3 mutants have a neural crest phenotype, and lack expression of FoxD3 in the trunk neural crest. However at cranial levels, where Pax7 is expressed, FoxD3 also is expressed [13]. Pax7 mutant mice have some craniofacial abnormalities, but survive well [28], and the impact of Pax3/Pax7 combined knockout on the neural crest has not been described. Substitution of Pax3 by Pax7 rescues the development of the neural crest [26], suggesting that there is partial redundancy between Pax3 and Pax7 in the mouse neural crest. In Xenopus, Pax3 is necessary for expression of FoxD3 [22], and in lamprey the Pax3/7 gene is similarly necessary for expression of the FoxD3 homolog FoxD-A [23].
Msx1 has been proposed to lie upstream of Pax3, FoxD3 and Snail2 during neural crest induction in Xenopus [21]. Loss of Msx1 or Msx2 in mice causes craniofacial abnormalities [32], [33], while the combined loss resulted in major defects in cranial neural crest derivatives, including mispatterning or reduction in size of cranial ganglia, loss, hypoplasticity or malformation of cranial bones, and conotruncal abnormalities [34]. Ablation of FoxD3 in mice in neural crest using Wnt-cre causes a similar phenotype at cranial levels; loss or reduction of many craniofacial structures, reduction in the size of cranial ganglia, subtle cardiac neural crest defects and also reduction in dorsal root ganglia size, and loss of enteric neural crest [24]. Cranial neural crest cells are still capable of undergoing migration in the absence of FoxD3 or Msx1/2, but many undergo apoptosis; in FoxD3 mutants apoptosis was seen in the neural tube or during migration [24], and in Msx1/2 mutants in the trigeminal ganglia and branchial arches [34]. As yet, the expression pattern of FoxD3 in the Msx1/2 mouse mutants is not known; however the strong similarities between the FoxD3 and Msx1/2 mutants at cranial levels provide support to the idea that Msx1/2 are immediately upstream of FoxD3 in the cranial neural crest. Differences at cranial levels between the mutant mice may reflect other roles of Msx genes, such as in neural tube and bone development. Other differences between the phenotypes suggest that in mice, Msx1/2 is not critical for neural crest development at trunk levels, unlike FoxD3. Although Msx transcription factors have been primarily described as transcriptional repressors [35], there is growing evidence for their role as transcriptional activators as well [36], [37]. Our results demonstrate that during avian cranial neural crest specification, Msx1/2 act as transcriptional co-activators of FoxD3.
Our data also show that Ets1 is necessary for initial FoxD3 expression since electroporation of Ets1 morpholinos during gastrulation (at HH5) depletes FoxD3 expression at HH9. In contrast, a dominant-negative Ets1 inhibited cranial neural crest migration but did not result in decreased FoxD3 expression [38]. Examination of the expression of FoxD3 and Ets1 by in situ hybridization suggests that Ets1 and FoxD3 are expressed concomitantly in the cranial neural crest. The difference in results between these two studies likely rests in the stages at which the knock-down reagents were effective, with the present results uncovering an earlier role for Ets1.
Recent work on the Sox10E2 enhancer showed that Sox10 expression in the cranial neural crest is regulated by Ets1, Sox9 and cMyb [5], [39]. The finding that Ets1 participates in activation of both FoxD3 and Sox10 at cranial level solidifies its potential crucial activation role in regulation of the cranial neural crest as a factor that initiates the specification module of the neural crest gene regulatory network. Interestingly, mis-expression of Ets1 in trunk levels confers cranial neural crest-like characteristics on trunk neural tube cells; namely increased delamination of neural crest independent of cell cycle phase [38]. This suggests that it plays a critical role in conferring head/trunk differences in the neural crest. However, conservation of this regulation across vertebrates remains to be determined. Ets1 is expressed by premigratory and migratory cranial neural crest in mice [40] and Xenopus [41]. Mice null for Ets1 have defects in cardiac neural crest, but none reported in cranial neural crest [40], [42]. Whether there is compensation for Ets1 in the cranial neural crest by other family members remains to be determined. Ets1 expression in chick neural crest is restricted to cranial levels; R4 and more rostral regions [38], [43]. Interestingly, the NC1 enhancer for FoxD3 is not active in R4, whereas the Sox10E2 enhancer is active to R6 [5], and Ets1 is active in R4 but not further caudally [38].
Although there is little published information regarding the molecular players are involved in the establishment of the more caudal neural crest populations, the present results implicate the neural plate border specifier, Zic1, as a critical factor in the control of FoxD3 expression at vagal and trunk levels. Zic1 has been shown to be required for FoxD3 expression in Xenopus neural crest [21], [22] where it is likely to partner with Pax3 in neural crest specification. Conversely, over-expression of Zic1 causes expansion of the FoxD3 and Snail2 expression domains [22], albeit it is unclear whether this occurs via direct or secondary interactions. The role of Zic1 as a trunk specific activator of FoxD3 is corroborated by expression data (Simões-Costa M., unpublished observations) suggesting much higher Zic1 transcript levels in the vagal/trunk avian neural folds than at cranial levels at the onset of FoxD3 expression. Our results are consistent with complementary functions of Zic1 in trunk and Ets1 in cranial neural crest specification in the avian embryo.
The present study expands the number of known direct regulatory interactions in the cranial neural crest gene regulatory network, confirming a direct regulation of FoxD3 by Pax3/7 and Msx1/2, and revealing a previously unknown regulation of FoxD3 by Ets1. We have also identified Zic1 as a key player in setting up the FoxD3 expression domain in the trunk neural crest. Several other genes, like Hairy2, Sox10 and Sox5, have been suggested to regulate FoxD3 expression [44]–[46]; however it remains to be determined whether this regulation is direct or indirect.
It is well known that the developmental potential of neural crest cells varies along different levels of the neural axis. Quail/chick chimeras have elegantly demonstrated that both the pathways of migration and derivatives differ depending upon the axial level from which neural crest cells emigrate [47]. For example, cranial but not trunk neural crest cells normally contribute to bone and cartilage. Similarly, vagal neural crest cells contribute to the enteric nervous system whereas other neural crest populations normally do not [48].
Our data show that the inputs to FoxD3 in the vagal/trunk region are distinct from those functioning at cranial levels, suggesting a model for region-specific expression of FoxD3 (Figure 7). Whereas the neural plate border specifier Zic1 appears to be a critical input for NC2 activity in the trunk, Ets1 is critical for activating NC1 at cranial levels. Both Zic1 and Ets1 transcription factors appear to act in concert with Pax7 and Msx1/2 which are expressed along the entire neural axis. To date, no transcription factors have been found to be selectively expressed in particular regions of the premigratory neural crest. However, the discovery of cranial-specific enhancers for FoxD3 (this study) and Sox10 [5] clearly suggest that these differences are inherent at the regulatory level. The existence of these enhancers supports the idea that both spatial and temporal information is encoded in the genome.
The genomic region of chicken FoxD3 was compared to other vertebrates using the ECR Browser (http://ecrbrowser.dcode.org) and comparative analysis tracks of UCSC Genome Browser (http://genome.ucsc.edu/). We analyzed a 160 kb genomic region encompassing the FoxD3 locus up to the first upstream (Atg4C) and downstream (Alg6) of neighboring genes. Regions containing elements found to be highly conserved across most vertebrates including human, mouse and Xenopus were amplified using Expand High Fidelity Plus system (Roche, Indianapolis, IN) with CH261-166E22 and CH261-100C15 (CHORI BAC Resources, http://bacpac.chori.org) BAC clones as templates and directionally cloned into the ptkeGFP or ptkCherry vectors [5], [17]. Mouse neural crest enhancers mNC1 and mNC2 were amplified using Expand High Fidelity Plus system from genomic cDNA. For use in multiple enhancer time-lapse experiments, ptkCitrine and ptkCerulean plasmids were constructed by swapping the eGFP coding region by Citrine and Cerulean sequences, respectively. Appropriate enhancer elements (Sox10E from [5] and NC1, NC2 – this study) were cloned into ptkCherry, ptkCitrine and ptkCerulean, respectively.
Chicken embryos were electroporated at HH4 using previously described techniques [1], [49]. FITC-conjugated morpholinos (against target factors or control morpholino) at concentrations from 1–3 mM combined with 1 mg/ml of enhancer_reporter Cherry constructs were electroporated only on one half of the embryo. For morpholino knockdown of regulators, electroporations were performed at HH5 or HH5+ to avoid disruption of the neural plate border. Fifteen to twenty embryos were analyzed for each of the morpholinos used. To electroporate HH8–14 chicken embryos in ovo, previously described techniques [50] were used with minor modifications; both constructs were injected at a concentration of 2 mg/ml each, and embryos were electroporated with 5 30ms-square pulses of 22 V with 100 ms rest in between each pulse. After incubation, embryos were collected and fixed in 4% paraformaldehyde for 1 hour, then viewed using fluorescence microscopy. Images were captured using a Zeiss Axioskop 2 plus microscope with AxioVision 4.6 software, and compiled using Adobe Photoshop 7.0 and Adobe Illustrator 10. For dynamic multiple enhancer analysis in slice culture, embryos were electroporated at stage HH4 with three constructs simultaneously (Sox10E-Cherry/NC1-Citrine/NC2-Cerulean) to allow for proper spectral separation of reporter signals from different enhancers. After roughly 16 hours of incubation, cranial midbrain regions were prepared and imaged as described previously [18].
Whole mount and section in situ hybridization for FoxD3 were performed using previously described procedures [51]. Whole mount in situ hybridization for eGFP was modified using the guidelines in [52]. Double fluorescence in situ hybridization was performed according to [53], and hybridization chain reaction (HCR) to detect endogenous FoxD3 was conducted according to [19].
Some embryos expressing RFP and eGFP were processed and cryosectioned at 14 mm. Select sections were labeled using the HNK-1 antibody (diluted 1/50), secondarily detected using goat anti-mouse IgM Alexa 350 (1/200; Molecular Probes). For whole mount immunostaining we used the protocol described by [54] (FoxD3 antibody generously provided by Patricia Labosky and Michelle Southard-Smith).
Regions of NC1 and NC2 were replaced with eGFP coding sequence using fusion PCR protocol. For 100 bp substitutions, the region of eGFP used was tggagtacaactacaacagccacaacgtctatatcatggccgacaagcagaagaacgg catcaaggtgaacttcaagatccgccacaacatcgaggacgg, for 30 bp substitutions acaagcagaagaacggcatcaaggtgaact and for 20 bp substitutions tggagtacaactacaacagc. Fragments were amplified using primers detailed in Table S2 and fused using the method adapted from [55]. Amplified fusion fragments were cloned into ptkeGFP and sequenced to ensure no additional mutations were present. ECR browser (http://rvista.dcode.org/) and Jaspar database (http://jaspar.genereg.net/cgi-bin/jaspar_db.pl) were used to predict and analyze binding motifs within highly conserved regions. Individual sites were mutated by substituting 6–8 adjacent critical base pairs with GFP coding sequence, using fusion PCR and sub-cloning into ptkeGFP. Primers used are listed in Table S3. Mutated enhancer constructs were electroporated into stage HH4 embryos as described above and analyzed for expression of eGFP and RFP at stages HH8–12. A minimum of five embryos was examined for each condition.
ChIP was performed using chromatin prepared from dorsal neural tube regions of HH8–10 (4–10 somite) chicken embryos using Ets1 (sc-350;Santa Cruz), Pax7 (ab34360. Abcam) and Msx1 antibodies (Sigma M0944) with normal rabbit IgGs (sc-2027,Santa Cruz;ab27478, Abcam) as previously described [56]. For the Zic1 ChIP chromatin was isolated from the dorsal neural tube regions from the trunk of HH11 embryos. Immunoprecipitation was performed with a Zic1 antibody from Sigma (HPA004098).
|
10.1371/journal.pgen.1005672 | The QTL within the H2 Complex Involved in the Control of Tuberculosis Infection in Mice Is the Classical Class II H2-Ab1 Gene | The level of susceptibility to tuberculosis (TB) infection depends upon allelic variations in numerous interacting genes. In our mouse model system, the whole-genome quantitative trait loci (QTLs) scan revealed three QTLs involved in TB control on chromosomes 3, 9, and in the vicinity of the H2 complex on chromosome 17. For the present study, we have established a panel of new congenic, MHC-recombinant mouse strains bearing differential small segments of chromosome 17 transferred from the TB-susceptible I/St (H2j) strain onto the genetic background of TB-resistant C57BL/6 (B6) mice (H2b). This allowed narrowing the QTL interval to 17Ch: 33, 77–34, 34 Mb, containing 36 protein-encoding genes. Cloning and sequencing of the H2j allelic variants of these genes demonstrated profound polymorphic variations compare to the H2b haplotype. In two recombinant strains, B6.I-249.1.15.100 and B6.I-249.1.15.139, recombination breakpoints occurred in different sites of the H2-Aβ 1 gene (beta-chain of the Class II heterodimer H2-A), providing polymorphic variations in the domain β1 of the Aβ-chain. These variations were sufficient to produce different TB-relevant phenotypes: the more susceptible B6.I-249.1.15.100 strain demonstrated shorter survival time, more rapid body weight loss, higher mycobacterial loads in the lungs and more severe lung histopathology compared to the more resistant B6.I-249.1.15.139 strain. CD4+ T cells recognized mycobacterial antigens exclusively in the context of the H2-A Class II molecule, and the level of IFN-γ-producing CD4+ T cells in the lungs was significantly higher in the resistant strain. Thus, we directly demonstrated for the first time that the classical H2- Ab1 Class II gene is involved in TB control. Molecular modeling of the H2-Aj product predicts that amino acid (AA) substitutions in the Aβ-chain modify the motif of the peptide–MHC binding groove. Moreover, unique AA substitutions in both α- and β-chains of the H2-Aj molecule might affect its interactions with the T-cell receptor (TCR).
| Many genes of the host regulate interactions with Mycobacterium tuberculosis and determine the level of susceptibility to, and severity of, tuberculosis (TB). Identification of these genes and their alleles is continuing and contributes new knowledge about the host-pathogen interactions. So far, forward genetic approaches (from phenotype to gene) have identified several chromosomal segments involved in genetic control of TB in mice (quantitative trait loci—QTL), but only one particular gene, Ipr1, has been identified. Here, we report the identification of a second TB-controlling gene. On the basis of a pair of mouse inbred strains with polar susceptibility to TB infection (susceptible I/St and more resistant C57BL/6) we established a panel of recombinant strains carrying small segments of Chromosome 17 from I/St on the genetic background of C57BL/6. A combination of genetic mapping, gene sequencing, TB phenotypes assessment and immunological approaches demonstrates that the H2-Ab1 gene encoding the beta-chain of the Class II heterodimer H2-A determines susceptibility to TB infection. The importance of allelic polymorphisms in Class II genes encoding antigen-presenting molecules in susceptibility to infection has been suspected. This is the first prove of this role obtained by the methods of classical forward genetics.
| Tuberculosis (TB) remains a significant public health problem: one-third of the human population is infected with Mycobacterium tuberculosis (MTB) and 10% of those are at a risk of developing overt TB during their lifetime [1, 2]. Although there is growing body of evidence that the outcome of infection is modulated both by bacterial and host genetics [3, 4], genetic factors regulating susceptibility to infection, transition from latency to reactivation and severity of the disease remain largely unknown. The important role of host genetic factors in TB disease control in humans has been clearly demonstrated in numerous studies, including adoption [5], twin [6–8], genome-wide association (GWAS) [9–12], and case-control population [13–16] studies. Apart from rare cases of Mendelian susceptibility to mycobacterial diseases (MSMD) due to nonsense and missense mutations in key genes involved in protective immunity against intracellular pathogens (reviewed in [17]), the complex patterns of TB susceptibility and disease manifestations clearly correspond to a polygenic type of genetic control with numerous epistatic interactions (reviewed in [18]). Naturally, identification of TB control-relevant genes and alleles in humans remains a very difficult task which is complicated by the environmental and strain diversity, as well as by the lack of consensus in the definition of, and distinction between, clinical phenotypes.
TB infection can be readily induced in mice, and some refined mouse TB models reproduce human-like pulmonary infection with appreciable accuracy (see [19, 20] for the review). In a few independent studies employing different inbred mouse strains, the whole genome scan approach has been applied for genetic mapping of quantitative trait loci (QTL) involved in TB susceptibility and disease control [21–25]. Since different inbred strains were selected as susceptible and resistant parental prototypes, and different phenotypes (survival time post-infection, multiplication of mycobacteria in organs, dynamics of cachexia) were analyzed, it is not surprising that the genomic locations of most of the QTL reported in different studies did not coincide. In our TB models, we use I/St TB-susceptible and A/Sn or C57BL/6 TB-resistant mice as prototypes. TB-infected I/St mice differ profoundly from their more resistant counterparts by early onset of mortality, rapid body weight loss, increased mycobacterial multiplication in lungs and spleens, and exacerbated lung histopathology [26]. Whole genome scans performed in F2 and N2 generations identified three QTL on chromosomes 3, 9 and 17 whose allelic variation affected TB susceptibility [22, 23]. The QTL on chromosome 17, peaking at the D17Mit175 marker, overlaps with the location of the mouse major histocompatibility complex Н2. Remarkably, this locus remains the only known case of co-localization of TB-controlling QTL reported in previous studies: Kramnik and colleagues mapped a QTL within the H2 region using a different combination of parental strains [21].
Associations of TB susceptibility/severity with the MHC polymorphic haplotypes have been previously reported both in humans [27–30] and mice [31–35]. In mice, allelic variations within the H2 complex were shown to affect survival time following challenge, the level of T-lymphocyte-mediated delayed type hypersensitivity (DTH) response, T-cell proliferation after stimulation with mycobacterial antigens and the efficacy of BCG vaccination against tuberculosis, production of IFN-γ by mycobacteria-specific T-cells, and production of mycobacteria-specific antibodies [31, 36–39]. However, these early studies provided no information about any particular gene within the H2 complex affecting TB immunity. Progression from a defined QTL region to a particular gene remains a major challenge: about 3,000 QTLs have been mapped in mice and rats but less than 1% of the genes have been identified at the molecular level [40]. This is especially true for the ~3.5 Mb MHC region, which provides the highest density of coding genes in the genome. Furthermore, many of these genes, which display a very high level of allelic variation and extensive linkage disequilibrium, have fundamental roles in immunity. Not surprisingly, numerous associations with several diseases for this part of the genome have been reported [41–44].
To begin identification of the gene, we have started to narrow the interval for the chromosome 17 QTL using a classical homologous recombination approach and have developed a panel of recombinant congenic mouse strains bearing different intra-H2 segments from TB-susceptible I/St mice on the resistant B6 genetic background. Given the previous demonstration that an allelic variant of chromosome 17 QTL inherited from B6 mice determined resistance to infection [21], we decided to use these common and genetically well-characterized mice as the TB-resistant prototype strain. At the initial stage of this study, we succeeded in narrowing the region on chromosome 17 which determines the level of TB-susceptibility from 8–65Mb to 33,77–34,34 Mb [45]. Since gene sequencing data for the I/St inbred strain are unavailable from the databases, in the present study we cloned and sequenced coding parts of all genes annotated for this region using I/St cDNA. As expected, the region displayed a very high level of genetic polymorphism and only a few out of 36 genes demonstrated identical sequences for B6 and I/St. In addition, the region under study contains many genes of importance for immunity and cell biology, thus being realistic candidates for the infection control. Thus, we searched for recombination events inside the TB-controlling region and established new mouse strains narrowing the region to the 34,24–34,33Mb interval. This interval contains only five coding genes, all belonging to classical and non-classical Class II: H2-Ob, H2-Aa, H2-Ab1, H2-Eb1 and H2-Eb2. Two recombinant strains with substantial differences in response to TB infection displayed recombination events in different parts of a single H2-Ab gene, which was critical for gene identification.
We transferred genomic regions covering the vicinity of the H2 complex from TB-susceptible parental I/St (H2j) mice onto the B6 (H2b) genetic background in successive backcross generations. Starting with the BC1 (N2) generation, we applied simultaneous selection for the presence of two traits: TB-susceptible phenotype and Chromosome 17 markers of the I/St origin. At the generation N10-11, more than forty B6.I-H2 recombinant congenic strains on the B6 background carrying different, partly-overlapping genomic regions of the extended H2j- haplotype (17 Chr: 8,44–65,34 Mb) were generated. Fig 1 displays the most informative B6.I strains whose pheno- and genotyping allowed us to narrow the region of interest to the interval 33,77–34,34Mb (a total of 0,57 Mb). Mice of all strains which inherited this region from I/St ancestors were significantly more susceptible to infection than those bearing B6 alleles as indicated by survival curves (Fig 2) and the dynamics of cachexia (S1 Fig). Fine mapping within this region was achieved by superposition: the resistant strain B6.I-249.1.16 carries H2j alleles more proximal than the SNPs rs13482956 (17:33, 773331) whereas the strain B6.I-9.3.19.8 is susceptible although all distal genes starting with and including H2-Ea are of B6 origin (Fig 1). Being more susceptible than B6, all recombinant strains carrying the region 33, 77–34, 34Mb inherited from I/St were more resistant than their I/St ancestors, indicating the influence of B6 background genes on survival. This is further supported by the fact that the level of resistance was identical in parental B6 mice and in recombinant mice which inherited the identified H2 segment from B6 (Fig 2). According to http://www.ensembl.org, the identified region contains 36 protein-coding genes, most of which may have important immunological and regulatory functions.
No information was available about the genome sequence of I/St mice, so it was impossible to start searching for candidate genes by direct sequence comparison. Therefore, we cloned and sequenced the protein-coding regions of all 36 I/St-originated genes in the region (GenBank, accession numbers KJ650201-KJ650234). Table 1 displays all amino acid (AA) substitutions between H2b and H2j haplotypes, as judged from the cDNA sequencing data. As expected, the region appeared to be highly polymorphic: only seven genes (Zbtb22, H2-Ke2, B3galt4, Slc39a7, Brd2, H2-DMb2, H2-DMb1) displayed no allelic polymorphism for the two haplotypes.
The H2 segment under study contains numerous genes generally involved in immune response control, and for many of these genes evidence is available indicating their possible involvement in the control of TB infection. S1 Table briefly summarizes the data illustrating this point. In addition, alternative splicing isoforms for many I/St alleles in this region not annotated previously were revealed (see: GenBank, accession numbers KJ663713- KJ663725), making the general picture of genetic diversity even more complex. The rich deposit of polymorphic genes potentially influencing susceptibility to, and severity of, TB infection, as well as the potential contributions of both polymorphism and expression regulation of other genes in the region, justified further narrowing the interval by genetic recombination.
To search for new recombination events inside the region 33, 77–34, 34 Mb, we performed several crosses between novel recombinant and B6 mice. In particular, the F2 progeny of (B6.I-249.1.15 x B6) F1 mice was used to develop a new set of congenic strains. In two new recombinant strains, B6.I-249.1.15.100 (hereafter–B6.I-100) and B6.I-249.1.15.139 (hereafter–B6.I-139), standard genotyping identified the point of recombination between markers D17Mit21 and D17Mit22 (Fig 3A). Surprisingly, these strains demonstrated sharply contrasting TB phenotypes (Fig 3B). After aerosol challenge, B6.I-139 mice did not differ by survival time from parental B6 mice (mean survival time (MST) = 238.9 ± 13.41 and 249 ± 10.21 days, respectively, P > 0.5). B6.I-100 mice did not differ from the B6.I -249.1.15.46 strain (MST 152 ± 13.3 and 153 ± 10.97 days, respectively, P > 0.5), but did differ significantly from the B6.I-139 strain (P < 0.001). Phenotypic differences were confirmed by evaluation of cachexia dynamics (S2 Fig), and by assessment of mycobacterial loads in the lungs at weeks 4 and 10 post-challenge (Fig 3C). Annotation in the http://www.ensembl.org database provides the length of 98, 588bp for the genomic region between D17Mit21 and D17Mit22, which contains only 5 protein-coding genes, 2 lincRNA genes and no genes for micro-RNAs (Fig 3D).
The chromosomal segment sufficient to determine the contrasting TB phenotypes appeared to be very small, and we identified genetic material inside the segment by gene sequencing. Both strains carried the b allele of H2-Ob and the j allele of H2-Aa, but differed at the H2-Ab1 gene (Fig 4). The H2-Ab1 gene in both strains originated from recombination events between b and j haplotypes, but the crossing-over occurred at different sites. In B6.I-139 mice, the whole polymorphic part of the H2-Ab1 gene encoding the extracellular functional domain of the molecule was of the H2b origin: only substitutions W222R in the connecting peptide and H249Y in the cytoplasmic domain were inherited from the H2j haplotype. In contrast, in B6.I-100 mice this polymorphic part of the H2-Ab1 gene was identical with that of the H2j haplotype, except for a single substitution N29D (Fig 4). As far as both recombination events occurred in the translated part of the gene, we assume that the promoter region of B6 origin was identical for both strains and played no role in infection response. The fact that B6.I-139 mice displayed the resistant phenotype similar to parental B6 mice suggests that two AA substitutions of I/St origin in the connecting peptide and the cytoplasmic domain are not major players in TB susceptibility. Analogously, the presence of the H2b-encoded aspartic acid in the H2-Ab1 of the B6.I-100 strain is unlikely to influence the level of TB susceptibility, since B6-I.100 mice display a phenotype identical to that of B6.I-249.1.15.46 mice, whose entire H2-Ab1 gene was inherited from I/St mice. Taken together, these results demonstrate that the differences in TB susceptibility/severity between these two recombinant mouse strains were determined by allelic polymorphisms in a single β1 domain of the H2-Aβ molecule.
Thus, independent recombination events within a single gene created genetic variation sufficient to markedly alter the response to TB infection. The newly identified H2-Ab1 alleles were designated as j* in the B6.I-100 strain and b* in the B6.I-139 strain.
We performed fine genetic mapping using the most integrative TB characteristics–survival curves, mycobacterial multiplication in the lungs, and body weight loss. Differences in the regulation of lung tissue inflammation after infection in B6 and I/St mice are of critical importance for TB pathogenesis [46, 47]. To characterize the influence of the H2-Ab1 polymorphism on TB-induced inflammation, we compared lung pathology 35 days post-infection in mice of both parental and new recombinant strains. As shown in Fig 5A, in B6 and B6.I-139 mice, lung pathology was represented by granulomatous areas well-delimited from the breathing tissue, whereas I/St and B6.I-100 mice developed diffuse TB pneumonia that was more severe in I/St mice.
In good agreement with the histological results, the levels of key Type 1 inflammatory cytokines, IL-6 and TNF-α, after TB challenge were significantly lower in the lungs of resistant B6 and B6.I-139 mice compared to susceptible I/St and B6.I-100 mice (Fig 5B). No difference in the levels of the TB-irrelevant Type 2 cytokine IL-5 between all four strains was found. Importantly, production of the key TB-protective Type 1 cytokine, IFN-γ, by CD4+ T-lymphocytes isolated from the lungs of infected mice and stimulated in vitro with a mixture of mycobacterial antigens followed the H2-Ab1-determined pattern. As shown in Fig 5C, the 5-fold difference in the numbers of IFN-γ-producing CD4+ T cells between parental B6 and I/St strains was reduced to a 2-fold difference between B6.I-139 and B6.I-100 mice but remained highly significant (P < 0.01). Calculation of the total numbers of IFN-γ-producing CD4 cells per lung provided the results consistent with the percentile evaluation (S2 Table). These results establish connections between anti-TB protective immune responses, CD4+ T-cell function and allelic heterogeneity of the classical Class II antigen-presenting molecule, providing a mechanistic explanation for the differences in the severity of disease determined by a single MHC gene.
IFN-γ production by the CD4+ T cell in response to MTB is generally considered the major mechanism of host defense [48]. TB-resistant B6 mice express only one MHC Class II molecule - H2-Ab on their antigen-presenting cells (APC), whereas I/St mice express both MHC Class II molecules - H2-Aj and H2-Ej. A priori it was impossible to judge whether the defect in TB defense in mice bearing the H2j haplotype was determined by sub-optimal antigen presentation by the H2-Aj compared to the H2-Ab molecule, or by the parallel presentation of mycobacterial antigens by two Class II molecules which somehow interfered with the development of protective immunity. To resolve this issue, we assessed the presentation of mycobacterial antigens by the APC derived from mice with different Class II allelic composition.
A mycobacteria-specific CD4+ T cell line derived from I/St mice [49] readily proliferated in the presence of mycobacterial antigens if the APC were derived from mice expressing the H2-Aj molecule, even if the H2-E molecule was not expressed (Fig 6A). Moreover, the presence or absence of H2-E did not change the level of response, suggesting that the H2-A molecule presents mycobacterial antigens to the vast majority of T-cell clones. To prove that the H2-E-recognizing T-cell clones have not been lost due to repeated stimulation during T-cell line development, we repeated the experiment with highly purified CD4+ T cells from TB-immune lymph nodes of I/St mice and obtained similar results (Fig 6B).
Recombinant B6.I-100 and B6.I-139 mice express the H2-E molecule due to the presence of the H2-Ej α-chain. These mice possess identical H2-Aj α-chains but differ in their H2-Aj β-chains (Fig 4). To determine in which context mycobacterial antigens were presented to T cells by newly originated H2 haplotypes and to evaluate the efficacy of antigen presentation by recombinant H2-Aβ chains, we assessed the response of T-cells from recombinant mice in the presence of different APC. As shown in Fig 6C and 6D, B6.I-100 and B6.I-139 T cells recognized mycobacterial antigens in the context of the H2-A, but not the H2-E, molecule. Interestingly, B6.I.100 T-cells did not distinguish fully syngenic Aj* and progenitor Aj, whereas B6.I-139 T cells recognized only Ab* β-chain, most likely because the hybrid H2-Aβb/Aαj molecule was formed. Taken together, these results indicate that the H2-A molecule plays a pivotal role in the presentation of mycobacterial antigens and the generation of TB-specific CD4+ T cell responses. The Aj and Aj* allelic variants are not intrinsically defective in the antigen-presenting function and elicit a level of T cell proliferation in response to soluble mycobacterial antigens similar to the Ab and Ab* alleles.
These observations provided an opportunity to functionally test whether or not the result of T cell interaction with infected macrophages depends upon H2-A alleles. To this end, we performed co-culture experiments with macrophages from B6 and congenic B6.I-9.3.19.8 (see Fig 1) mice and immune effector CD4 T-cells obtained from (B6 x B6.I-9.3.19.8) F1 mice, which are able to interact with both allelic forms of H2-A. The H2-E-negative strain B6.I-9.3.19.8 was used instead of B6.I-100 to further exclude possible influence of the H2-E expression. As shown in Fig 6E, T-cells profoundly increased the ability of B6 and F1 peritoneal macrophages to inhibit mycobacterial growth, whereas only marginal effect was seen in B6.I-9.3.19.8 macrophages, suggesting that recognition of “protective” H2-Ab vs. “non-protective” H2-Aj molecules by CD4+ T-cells leads to profound differences in macrophage activation. Remarkably, moderate capacity to inhibit mycobacterial growth in the absence of T-cells was similar in all macrophages, regardless their genetic origin. This is in full agreement with theoretical expectations: Class II alleles do not regulate the level of innate protective response. These results provide additional independent conformation in support of the conclusion that the H2-A allelic variation is sufficient to determine prominent variations in acquired anti-mycobacterial immunity. In contrast with experiments on genetic restriction of antigen-specific response described above, reciprocal functional experiment (activation of F1 macrophages by H2-Ab and H2-Aj T-cells) would not be informative, since allogenic Class II recognition provides unpredictable effects.
A BLAST search in the Protein Data Bank (PDB) for α- and β-chains of the H2-Aj molecule revealed the highest score of sequence similarity (88% and 85% identity, respectively) with the protein 2P24 [50], with deletions or insertions lacking. Comparison between H2-Aj and H2-Ab (PDB id 1MUJ) provided sequence identity of 93% and 89% for α- and β-chains, respectively, with the 2-AA deletion (P65 and E66) in the former allele. Two molecular models of the H2-Aj protein based upon atomic coordinates of 1MUJ and 2P24 provided a high level of similarity around the deletion point (S1 Fig), which justified the use of the 1MUJ model for further comparisons.
Comparison between the H2-Aj and H2-Ab molecules suggests that the most prominent structural dissimilarities occur in two protein backbone regions: the 310 helical fragment of the α-chain and the P65E66 deletion in the H2-Aβj chain (Fig 7A). These deviations are not unique and are present in other H2 haplotypes (reviewed in Ref [51] and displayed in S3 Fig). Analysis of the hydrogen bond network between H-2A Class II molecules and, as a model, invariant CLIP peptide stabilizing the complex before an antigenic peptide is loaded, demonstrated that the conserved H-bond interactions and the total number of H-bonds are identical for the H2-Ab and H2-Aj products despite two b → j substitutions, T71K and E74A in the Aβ-chain (Fig 7B and 7C). The Aβ-position E74 is highly conserved among all known mouse H2-A haplotypes (S4 Fig) and the majority of human HLA-DQ molecules [52]. In the H2-Aj molecule, the A74-provided H-bond is lacking; however, it might be functionally substituted by the H-bond from the Aβ K71.
However, prominent differences in the structure and size of the peptide-anchoring pockets between the two allelic forms of the H2-A molecule were observed. Structural data for the peptide-anchoring pockets were either available from the 3D structure of the H2-Ab [53] or deduced for the H2-Aj from our model. AA substitutions distinguishing the H2-Aj protein from the prototypic H2-Ab should have profoundly changed the structure of the peptide-binding groove in binding pockets P1, P4, P6, P7 and P9. Due to the substitution L31W and A52T in the α-chain, the volume of the H2-Aj P1 pocket should be appreciably smaller compared to that of H2-Ab (Fig 7D) and, therefore, may accept smaller side chains capable of forming a hydrogen bond with the Aβ T52 hydroxyl group. Changes in other binding pockets are due to substitutions in the β-chain (Fig 7E). The differences between H2-Aj and H2-Ab in the P4 binding pocket include substitutions Y26L, T28S, G13S and the most prominent one—E74A, allowing occupation of the pocket by a neutral residue in H2-Aj instead of a positively charged residue in H2-Ab. Changes in the P6, (substitutions Y30N and T71K), make this pocket in the H2-Aj molecule more permissive for negatively charged residues. Due to the H47F substitution, the pocket P7 is neutral in H2-Aj but positively charged in H2-Ab, while a minor substitution Y37F makes pocket P9 more permissive for lipophilic residues. We consider the most important differences between H2-Aj and H2-Ab to result from a unique substitution Q61K in the α-chain and substitution R70Q in the β-chain, since these substitutions alter the polarity of interactions between two chains (S5 Fig). Two positively charged residues occupying opposite positions in the hybrid H2-A (αj+βb*) (S5 Fig) could well affect the Class II-TCR interactions.
An extremely high level of genetic polymorphism in the H2 chromosomal region and its unique saturation with immunity-relevant genes complicate the identification of allelic variants in a particular gene influencing the disease severity and outcome. This is particularly true for the classical Class I and II genes, for which genetic silencing approaches are hardly applicable since they lead to a severe abrogation of overall functions of acquired immunity. It took us about 7 years to develop the panel of more than 40 H2-recombinant inbred strains on the B6 genetic background sufficient for identification of the H2-Ab1 gene as the TB severity determinant using the forward genetic approach. The key point was establishing the fact that in the B6.I-100 and B6.I-139 strains distinct recombination breakpoints between the H2j and H2b haplotypes were located within the same H2-Ab1 gene. This resulted in a H2-Ab1b-like allele in TB-resistant B6.I-139 and in a H2-Ab1j-like allele in TB-susceptible B6.I-100 mice, which, being compared with previously characterized phenotypes and genotypes in other strains from the panel, identified the H2-Ab1 as the gene underlining the Chromosome 17 TB-controlling QTL. This is the first direct demonstration of the differences in TB infection susceptibility/severity depending on allelic polymorphisms in a single Class II MHC gene. Associations of TB susceptibility/severity with the MHC polymorphic haplotypes have been previously reported both in humans [27–30] and mice [31–35], but direct evidence that the alleles of H2-A1 or its human orthologous gene HLA-DQ differentially regulate TB control by the host was lacking. Importantly, in our system both allelic variants apparently retain normal functional activity and are not defective in mycobacterial antigen presentation/recognition (Fig 6). This may reflect the situation which exists in natural populations: fine genetic differences may lead to pronounced shifts in adaptively important responses providing the subject for slowly operating natural selection on quantitative basis, whereas loss-of-function mutations in key immunologically active genes are eliminated rapidly. Allelic differences in a single H2-A1 gene influenced all major phenotypes characterizing severity of TB infection (bacterial loads in affected organ, histopathology, cachexia, survival time post-challenge), suggesting that H2-Ab1 is one of major players in the TB control in mice. Naturally, this does not exclude the presence of other genes within an extended H2 region involved in TB control, but their possible contribution is likely to be weaker.
The established panel of recombinant mouse strains may serve a useful tool for dissecting genetic control of susceptibility/severity in other models of TB and in other experimental infections. Indeed, one TB-susceptibility QTL, Sst5, has been mapped to the H2 region in the B6 –C3HeB/FeJ mouse strain combination [21]. An extended H2 region contains QTL involved in genetic control of susceptibility/severity of several protozoal and metazoal pathogens: Lmr1 for Leishmania major [54], Char3 for Plasmodium chabaudi [55], Belr1 for Plasmodium berghei [56], Tir1 for Trypanosoma congolense [57] and Sm2 for Schistosoma mansoni [58]. It is very unlikely that co-localization of all these QTL is coincidental, and our new panel of mouse strains may shed light on the architecture of the H2-driven genetics of host-parasite interactions in these other disease models.
Regarding the molecular mechanisms underlining differences in TB susceptibility in the absence of overt gene dysfunctions, several possibilities are being considered. The most evident is differences in mycobacterial antigenic peptide repertoire presented to T-cells by structurally different Class II molecules. Alignment of the AA sequence of the polymorphic H2-Ab1j domain with annotated H2 haplotypes (S4 Fig) demonstrates that it shares the common deletion P65-E66 with haplotypes k, g7, u, s, f, s2, retains conserved AA residues at positions determining the basic 3D structure of the protein, and is not unique with this regard. However, there are non-conservative and potentially important substitutions located in four β-strands and in the α-helix that form the peptide-binding groove. Molecular modeling predicts that the motif for peptide binding should differ between H2-Aj and H2-Ab due to substitutions in the pockets P4, P6, P7 and P9 (Fig 7). However, at present no information is available concerning sets of mycobacterial peptides providing more vs. less protective anti-TB responses.
We also considered the level and stability of the cell surface expression of the H2-A allelic forms as a factor potentially influencing the level and quality of T-cell activation. Since antibodies reacting with the H2-Ab and H2-Aj molecules with equal affinity are lacking, a direct quantitative comparison of expression levels was impossible. Thus we applied an antibody dilution approach described earlier [59, 60] and found no differences in the expression levels for the H2-A molecule between B6, B6.I-100 and B6.I-139 mice (S6 Fig), most likely excluding this explanation.
Yet another possible reason for the differences in anti-mycobacterial immunity between the carriers of H2-Ab1j and H2-Ab1b alleles could be selection of CD4 T cells in thymus and/or their maintenance in the periphery. Our preliminary studies demonstrated a significant difference in the CD4: CD8 ratio between B6 and I/St mice, as well as between some of the novel recombinant mice. More data and, possibly, new recombinant mouse strains expressing no H2-E molecule will be needed to precisely evaluate the importance of this MHC-dependent pathway of immune response regulation.
Mice of inbred strains I/StSnEgYCit (abbreviation I/St, H2j) and C57BL/6JCit (abbreviation B6, H2b) were bred and maintained under conventional, non-SPF conditions at the Animal Facilities of the Central Institute for Tuberculosis (CIT, Moscow, Russia) in accordance with the guidelines from the Russian Ministry of Health # 755, and under the NIH Office of Laboratory Animal Welfare (OLAW) Assurance #A5502-11.
The B6.I panel of MHC-congenic strains (fragments of the H2j haplotype transferred onto B6 genetic background; totally, more than 40 strains) was established using the classical cross–backcross–intercross protocol [61]. Selection of backcross progeny carrying the H2j haplotype of the I/St origin and its inter-MHC recombinant derivatives with the H2b haplotype was performed in each backcross generation. Genotypes of the simple sequence length polymorphisms in the region under study (SSLPs D17Mit, www.jax.org) were determined using PCR of isolated tail DNA samples (Wizard Genomic DNA Purification Kit, Promega, USA) followed by product separation on 4–6% agarose gels. The following markers were used: D17Mit13, 16, 21, 22, 28, 47, 49, 57, 81, 82, 103, 133, 143, 147, 152, 175, 177, 233. Two SNP markers, rs13482956 and rs50366565, were genotyped by PCR followed by enzymatic digestion of PCR products with FspBI and Hin1II, respectively (Thermo Scientific, Lithuania).
All carriers of novel MHC-region allelic variants were serially backcrossed on B6 parental mice. After generation N = 10–14, homozygous animals were obtained by sib mating and further maintained by brother-sister mating. Water and food were provided ad libitum. Mice of 8-12wk of age at the beginning of experiments were used. All experimental procedures were approved by the CIT Animal Care Committee (IACUC protocols #2, 7, 8, 11 approved on March 6, 2013).
To evaluate severity of the disease, mice were infected with ~5 x 102 colony-forming units (CFU) of standard virulent M. tuberculosis strain H37RV (sub-strain Pasteur) using an Inhalation Exposure System (Glas-Col, Terre Haute, IN) exactly as described earlier [49]. Mortality was monitored daily starting at week 5 post-infection. To assess CFU counts, lungs from individual mice were homogenized in 2.0 ml of sterile saline, and 10-fold serial dilutions were plated on Dubos agar (Difco) and incubated at 37°C for 20–22 days. Pathology of the lung tissue was assessed as described [50]. Briefly, mice were euthanized by a thiopental (Biochemie GmbH, Vienna, Austria) overdose. Lung tissue (the middle right lobe) was frozen in a –60°C to –20°C temperature gradient in an electronic Cryotome (ThermoShandon, UK), 6–8μm-thick sections were cut across the widest area of the lobe, fixed with acetone, stained with hematoxylin-eosin and mounted.
To prepare T-cell lines, cells from the popliteal lymph nodes of I/St and B6 mice, immunized into rear footpads with 10μg/mouse of mycobacterial sonicate mixed 1:1 with incomplete Freund’s adjuvant, were cultured as described previously [62]. Briefly, 2 x 106/ml immune cells isolated on day 21 post-immunization were cultured in 24-well plates (Costar, Netherlands) in RPMI-1640 containing 10% FCS, 10 mM HEPES, 4 mM L-glutamine, 5 x 10−5 M 2-ME, vitamins, pyruvate, non-essential amino acids and antibiotics (all components–HiClone, Logan, UT, USA) for 14–16 days in the presence of 10μg/ml mycobacterial sonicate. Live immune cells (>93% viability by trypan blue exclusion) were isolated by centrifugation at 2500 g for 20 min at 20°C, on the Lympholyte M gradient (Cedarlane Labs, Ontario, Canada), washed twice and counted. The next stimulation cycle was accomplished by co-culturing 2 x 105 isolated cells with mitomycin C-treated 1.5 x 106 splenic APC in the presence of sonicate for another 14–16 days. These cycles were repeated 4 times and resulted in stable antigen-specific CD4+ (>99% purity by flow cytometry) T-cell lines. To obtain fresh immune CD4+ T cells, at day 21 following immunization lymph node cells were purified by negative selection using magnetic beads (CD4+ T-cell Isolation kit II, Miltenyi Biotec) according to the manufacturer’s recommendations.
To assess antigen-specific proliferation, either 105 purified CD4+ T cells, or 104 T-line cells were co-cultured with 2 x 105 mitomycin C-treated splenic APC in a 96-well flat-bottom plate (Costar), at 37°C, 5% CO2, in supplemented RPMI-1640 containing 10 μg/ml of H37Rv sonicate. Non-stimulated wells served as controls. Triplicate cultures were pulsed with 0.5 μCi [3H]-thymidine for the last 18 h of a 40 h incubation. The label uptake was measured in a liquid scintillation counter (Wallac, Finland) after harvesting the well’s contents onto fiberglass filters using a semi-automatic cell harvester (Scatron, Norway).
Peritoneal macrophages were obtained after stimulation with peptone as described previously [63]. 50 x 103 macrophages per well of 96-well plates in RPMI-1640 supplemented with 2% FCS and containing no antibiotics were infected with M. tuberculosis H37Rv at MOI 5:1 for 1.5 h. CD4+ T cells (~97% purity) were obtained from spleens of (B6 x B6.I-9.3.19.8) F1 mice at day 21 after i. v. infection with 5 x 105 CFU of M. tuberculosis H37Rv using magnetic separation (see above). T cells were added to infected macrophages at 1:1 ratio, and co-cultures kept for 72 h at 37°C in CO2 incubator. To assess mycobacterial viability, [3H]-uracil label was added for last 18 h of incubation, and the uptake assessed exactly as described in [63]. This method provides >99% correlation with CFU counting [63].
Infected B6 and I/St mice were euthanized by thiopental overdose, and lung cell suspensions were prepared using the methods described earlier [64]. Briefly, blood was washed out by repeated broncho-alveolar lavage with 0.02% EDTA-PBS with antibiotics, lungs removed, sliced into 1–2 mm3 pieces and incubated at 37°C for 90 min in supplemented RPMI-1640 containing 200 U/ml collagenase and 50 U/ml DNase-I (Sigma, MO). Single cell suspensions obtained by vigorous pipetting were washed twice in HBSS containing 2% FCS and antibiotics. Suspensions of spleen and lymph node cells were obtained using routine procedures. Cells were incubated 5 min at 4°C with an anti-CD16/CD32 mAb (BD Biosciences) for blocking Fc-receptors and stained with FITC-anti-CD3, APC-anti-CD8 and PerCP-anti-CD4 antibodies (BD Biosciences).
For intracellular IFN-γ staining, 1.5 × 106 cells were cultured in 24-well plates in the presence of 10 μg/ml mycobacterial sonicate for 48 h; GolgiPlug block (1 μl/ml; BD Biosciences) was added for the last 10 hours. Cells were then stained with anti-IFN-γ mAb XMG1.2 (BD Biosciences) using the Cytofix/Cytoperm kit (BD Biosciences). The expression of the Н2-Еα molecules on cell surface, discriminating Н2b (Н2-Еα-negative) and Н2j (Н2-Еα-positive) haplotypes was assessed using the PE-14-4-4S mAb (BD Biosciences) Cells were analyzed on BD Biosciences FACSCalibur flow cytometer using CellQuest and FlowJo software.
The levels of cytokines in the lung tissue was measured individually in infected animals using whole-lung homogenates in 2 ml of sterile saline stored at –70°С before assessment. After thawing, debris was removed from the samples by centrifugation at 800 g, and cytokine levels in supernatants were assessed in an ELISA format using mouse OptEIA TNF-α Set, OptEIA IL-6 Set and OptEIA IL-5 Set (BD Biosciences) and mouse INF-γ Set (Biolegend) according to the manufacturer’s instructions.
RNA was extracted from spleens using the SV Total RNA Isolation System (Promega, USA) and treated with DNase I (AMPD1, Sigma). Complementary DNA (cDNA) was synthesized with oligo-dT18 primers (Thermo Scientific, Lithuania) and M-MLV reverse transcriptase (Promega, USA). Primer sequences for cloning were obtained from the Ensembl database (version GRCm38.p2) for the C57BL/6 strain. 5'(forward) primers ended at the start codon (ATG); reverse primers started at the (TGA) stop codon. Coding DNA was amplified with Advantage GC Genomic LA Polymerase (Clontech, USA), PCR products were purified by gel extraction with Cleanup Mini Set (Evrogen,Russia) and cloned into the PCR-Script Amp Cloning vector using the PCR-Script Amp Cloning Kit (Stratagene, USA) or in pAL-TA (Evrogen, Russia) with preliminary 3 cycles of amplification of PCR products with Taq polymerase (Helicon, Russia). The 4–6 positive clones were sequenced for each gene. Nucleotide sequences have been submitted to the GenBank (http://www.ncbi.nlm.hih.gov/genbank) under accession numbers KJ650201-KJ650234, KJ663-713-KJ663725.
Molecular modeling was performed using an Octane2 workstation (Silicon Graphics, USA) equipped with the programs Insight II/Discover (Accelrys, USA). Atomic coordinates of the mouse Class II MHC H2-Au MBP125-135 (PDB id 2P24) [50] and Class II MHC H2-Ab in complex with the human CLIP peptide (PDB id 1MUJ) [54] were used for homology modeling. AA substitutions were deduced using the Biopolymer program. In order to minimize inter-atomic clashes, all individual AA conformations were chosen automatically, using the criteria of the lowest energy. To deduce the structure of the H2-Aj molecule using the H2-Ab template, the deletion P65-E66 was introduced manually using the Biopolymer program. Atomic coordinates available from two models, 1MUJ and 2P24, for the Class II-CLIP structures were subjected to further refinement for the I-Aj using the Discover program and AMBER force field. Refinement stages included short energy minimization (the steepest descent algorithm), followed by 1 ps molecular dynamics simulations at 298K and by the final energy minimization (the conjugate gradient algorithm). Results were visualized using the Insight II and Discovery Studio software (Accelrys, USA).
All analyses were done using Graphpad Prism version 4. Mortality was assessed using Kaplan-Meier survival analysis and the log-rank tests, CFU counts using Student’s t-test. P < 0.05 was considered statistically significant.
|
10.1371/journal.pntd.0002281 | Phylogenetic Analysis Reveals a High Prevalence of Sporothrix brasiliensis in Feline Sporotrichosis Outbreaks | Sporothrix schenckii, previously assumed to be the sole agent of human and animal sporotrichosis, is in fact a species complex. Recently recognized taxa include S. brasiliensis, S. globosa, S. mexicana, and S. luriei, in addition to S. schenckii sensu stricto. Over the last decades, large epidemics of sporotrichosis occurred in Brazil due to zoonotic transmission, and cats were pointed out as key susceptible hosts. In order to understand the eco-epidemiology of feline sporotrichosis and its role in human sporotrichosis a survey was conducted among symptomatic cats. Prevalence and phylogenetic relationships among feline Sporothrix species were investigated by reconstructing their phylogenetic origin using the calmodulin (CAL) and the translation elongation factor-1 alpha (EF1α) loci in strains originated from Rio de Janeiro (RJ, n = 15), Rio Grande do Sul (RS, n = 10), Paraná (PR, n = 4), São Paulo (SP, n = 3) and Minas Gerais (MG, n = 1). Our results showed that S. brasiliensis is highly prevalent among cats (96.9%) with sporotrichosis, while S. schenckii was identified only once. The genotype of Sporothrix from cats was found identical to S. brasiliensis from human sources confirming that the disease is transmitted by cats. Sporothrix brasiliensis presented low genetic diversity compared to its sister taxon S. schenckii. No evidence of recombination in S. brasiliensis was found by split decomposition or PHI-test analysis, suggesting that S. brasiliensis is a clonal species. Strains recovered in states SP, MG and PR share the genotype of the RJ outbreak, different from the RS clone. The occurrence of separate genotypes among strains indicated that the Brazilian S. brasiliensis epidemic has at least two distinct sources. We suggest that cats represent a major host and the main source of cat and human S. brasiliensis infections in Brazil.
| Sporotrichosis is a subcutaneous mycosis acquired by traumatic inoculation of soil and plant material contaminated with infectious propagules of the pathogen. The transmission of the disease by cats to other animals and humans occurs by biting or scratching, promoting direct inoculation of yeast cells into host tissue. This may represent an alternative and a successful transmission of the fungus. In order to understand the impact of felines on the epidemiology of sporotrichosis, we evaluated the phenotypic and genotypic features of isolates obtained from animals and humans living in outbreak areas. Although sporotrichosis is caused by a complex of species, in this study we observed that S. brasiliensis is the prevalent etiological agent of feline sporotrichosis, having been recovered from 96.9% of the samples. Moreover, this approach allowed us to recognize that isolates from RJ, SP, PR and MG states are genetically similar among them but different from feline isolates recovered from the RS epidemic. Our study brings new insights into the eco-epidemiology of sporotrichosis in Brazil, clarifying the distribution and prevalence of S. brasiliensis in feline outbreaks. Knowledge about the source and distribution of the etiological agent between outbreak areas may help to establish public strategies for the containment of the epidemic of sporotrichosis in Brazil.
| Mycotic diseases, particularly those caused by dimorphic fungi such as Sporothrix, can be considered as an emerging threat to various species of animals. Upon introduction of propagules into the mammalian host, the fungus undergoes a thermodimorphic transition to a yeast-like phase, leading to infections varying between fixed localized cutaneous lesions to severe, disseminated sporotrichosis.
The first connection between Sporothrix and animals was made by Lutz and Splendore [1]. Since then sporotrichosis has been reported in dogs, cats, horses, cows, camels, dolphins, goats, mules, birds, pigs, rats, and armadillos, as well as in humans. However, the cat is the animal species most affected by this mycosis [2]. Over the last two decades, Brazil has experienced its largest epidemic of sporotrichosis due to zoonotic transmission, whereby cats were pointed out as key susceptible host. The zoonotic potential of infected cats has been demonstrated by the isolation of S. schenckii s.l. from feline skin lesions, nasal, oral cavities, and claw fragments [3], [4].
In contrast to the classical route of infection by Sporothrix, where soil and plant material loaded with saprophytic hyphae of the pathogen were the source of contamination [5], transmission of Sporothrix spp. by cats to other cats and to humans via direct inoculation of yeast cells represents an alternative and a successful type of dispersal of the disease. The yeast form is more virulent than the mycelial form [6], [7]. Transmission of yeast cells may enhance the appearance of more severe forms of the disease.
Until recently, S. schenckii was considered to be the only species causing sporotrichosis. The infection has a worldwide distribution, mainly in tropical and subtropical countries [8]–[10]. The most common clinical manifestations in humans are the lymphocutaneous and fixed forms, but other clinical types, such as a disseminated form, may also occur [11], [12], partly depending on the immune status of the host.
Multilocus sequencing combined with morphological and physiological data support the separation of at least four distinct Sporothrix species within the S. schenckii complex, uniting the species with high pathogenic potential to mammals. The original taxon S. schenckii (Clades IIa and IIb) and the novel species S. brasiliensis (Clade I), S. globosa (Clade III), and S. luriei (Clade VI) todays are referred to as the S. schenckii complex [13], while the mildly pathogenic species S. mexicana (Clade IV) takes a remote position near the environmental species S. pallida [11], [14]–[19]. The Sporothrix species differ in their pathogenic potential for mammals [20], [21], their geographical distribution [11], [13], [15], [17], and in their sensitivity to antifungal therapy [22]. All species have been reported from Brazil [11].
Endemic areas of sporotrichosis in Brazil are characterized by poor sanitation, substandard housing and little or no access to health services – a challenge to control and eradication of the disease. The oldest outbreaks of sporotrichosis among humans and cats have been reported in the states of Rio de Janeiro [3], [23], [24] and Rio Grande do Sul [25], [26]. Delayed diagnosis and treatment in cats may lead to a rapid spread of the disease through the community members. The increase in the number of cases in cats is followed by higher numbers of human cases, which constitutes a serious public health problem.
Despite the increasing frequency and severity of cases, the eco-epidemiology of feline sporotrichosis in Brazil is still unknown. The aim of the present study was to determine the distribution and prevalence of Sporothrix species among naturally infected felines using phenotypic and molecular phylogenetic approaches.
Thirty three (33) Sporothrix isolates from Rio de Janeiro, RJ (n = 15); Rio Grande do Sul, RS (n = 10); Paraná, PR (n = 4); São Paulo, SP (n = 3) and Minas Gerais, MG (n = 1) were obtained from lesions of cats and dogs with sporotrichosis (skin or mucosa lesions) (Fig. 1). Fungal cells were recovered directly from lesions and cultured on Mycosel agar (Difco Laboratories, Detroit, Mich.). Suspected colonies were subcultured on potato dextrose agar (Difco Laboratories, Detroit, Mich.) at room temperature. Isolates were identified phenotypically as S. schenckii s.l. As a control, human clinical isolates (n = 66) inside and outside the Brazilian feline outbreaks areas were included in the study (Table 1).
Morphological identification of cultures was performed according to Marimon et al. [17], [18] including vegetative growth on PDA media at 30, 35, 37 and 40°C, colony colors on corn meal agar (Difco Laboratories, Detroit, Mich.), assimilation profiles of raffinose, ribitol and sucrose, and microscopic morphology in vitro. Growth at different temperatures was measured according to Mesa-Arango et al. [27]: the percent growth inhibition (GI) was calculated at 37°C by the following formula [(colony diameter at 30°C – colony diameter at 37°C)/colony diameter at 30°C]×100. The GI was evaluated by analysis of variance/Tukey test using the GraphPad (GraphPad Prism v. 5.00 for Windows, San Diego California USA, www.graphpad.com), considering statistically significant when p<0.05. Observed data were used for taxonomic characterization applying the dichotomous key to species of the S. schenckii complex proposed by Marimon et al. [18].
For molecular analysis, genomic DNA was extracted and purified directly from mycelial colonies following the Fast DNA kit protocol (MP Biomedicals, Vista, CA, USA) with the homogenization step repeated three times with a Precellys 24 instrument (Bertin, Montigny le Bretonneux, France). DNA was quantified with NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). The calmodulin (CAL) locus region was amplified directly from the genomic DNA by PCR, as described by O'Donnell et al. [28], using the degenerate primers CL1 (5′-GAR TWC AAG GAG GCC TTC TC-3′) and CL2A (5′-TTT TTG CAT CAT GAG TTG GAC-3′), which generated an 800-bp amplicon corresponding to exons 3 through 5. The translation elongation factor-1 alpha (EF1α) locus region was amplified using the newly designed primers EF1-F (5′-CTG AGG CTC GTT ACC AGG AG-3′) and EF1-R (5′-CGA CTT GAT GAC ACC GAC AG-3′) which amplified an 850-bp fragment, covering the last exon of this gene, matching the same region evaluated by the consortium Assembling the Fungal Tree of Life (AFTOL).
Thermal cycling conditions were as follows: one cycle of 5 min at 95°C, followed by 35 cycles of 1 min at 95°C, 1 min at 60°C (CAL) or 57°C (EF1α) and 1 min at 72°C, followed by one cycle of 10 min at 72°C.
Amplified products were gel-purified with the Wizard® SV Gel and PCR Clean-Up System (Promega, USA) following the manufacturer instructions. DNA samples were completely sequenced with an ABI 3730 DNA Analyser (Applied Biosystems, Foster City, CA, USA) using BigDye® Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems). The fragments were sequenced on both strands to increase the quality of sequence data and assembled into single sequences via CAP3 using bases with quality of phred ≥30. Sequences were aligned with MAFFT v. 5.667 [29] and retrieved alignments were manually edited in order to avoid mis-paired bases.
Calmodulin sequences deposited at GenBank belonging to the clades of clinical importance in the S. schenckii complex (Table 1) were collected and included in the present alignment as reference strains for the phylogenetic distribution. We choose the saprophytic fungus Grosmannia serpens (Ophiostomataceae), CBS 141.36 [30] as outgroup for CAL analysis [11]. All Sporothrix EF1α sequences used in the phylogenetic analysis were generated in this study (Table 1). The outgroup for the EF1α analysis included the saprophytic fungus Ophiostoma piliferum, CBS 158.74 (AFTOL-ID 910) [31]. This species was chosen because the genus Ophiostoma (Ophiostomataceae) is considered a close related group to Sporothrix species [32].
Phylogenetic analyses were carried out using Neighbor-joining, Maximum Likelihood and Bayesian methods. Neighbor-Joining and Maximum Likelihood trees were constructed using MEGA 5 software [33] and 1000 bootstrap replicates were used to estimate confidence values for individual clades and are shown next to the branches [34]. The evolutionary distances were computed using the Tamura 3-parameters method [35] and the rate variation among sites was modeled with a gamma distribution (shape parameter = 1). For Bayesian analysis by Markov Chain Monte Carlo (MCMC), two independent analyses of four chains each as default were initiated from a random tree and processed for 1.000.000 generations; sample trees were retrieved every 1000 generations. Log-likelihood scores were plotted against its generation number in order to evaluate convergence; samples collected prior to “burn-in” (25%) were ignored. The remaining samples were used to determine the distribution of posterior probability values [36]. The posterior probabilities values of generated clades and overall topology of each replicate were compared in order to verify that each consensus tree converged on a similar phylogeny. Phylograms generated by Bayesian analysis were used to represent the phylogenetic distribution and were produced with the help of the Figtree 1.0 software (available at http://tree.bio.ed.ac.uk/software/figtree/).
Evolutionary relationships at the intraspecific level were evaluated using haplotype networks in order to visualize differences and diversity among S. brasiliensis sequence data. The number and diversity of CAL and EF1α haplotypes were estimated using the software DNAsp v5.0 [37]. Gaps and missing data were excluded in the calculation. Median-joining networks [38] for the CAL and EF1α dataset were obtained and visualized using the software Network 4.610 (available at www.fluxus-engineering.com).
Evidence of recombination in S. brasiliensis population isolated from animals and humans samples was inferred by the split decomposition method [39], implemented by the Splitstrees4 software, version 4.8 [40] which is used to identify branching ambiguities attributable to recombination events. The presence of recombination networks can be detected by bridges between members of the genetically isolated groups. Each isolated group will have an independent branch, showing that it does not share genetic material with the others. This analysis allowed the assessment of recombination possibilities within and between the seven phylogenetic groups considered.
The PHI-test incorporated in the SplitsTree software [40] was used to test signals of recombination (p<0.05, significant evidence of recombination). The test is proven to be a robust calculation and no previous knowledge about population history, recombination rate, mutation rate and rate heterogeneity across sites [41] is necessary. Although large splits in networks do not necessarily imply recombination, split decomposition networks in conjunction with the PHI-test can easily detect which sequences in a given data set contribute the most to the recombination signal [42]. The PHI-test is repeated after possible recombinants are deleted from the alignment until p>0.05 (no evidence of recombination). Also, DNAsp v5.1 [43] was used to evaluate minimum number of recombination events in the history and haplotypic diversity of S. brasiliensis population. The software computes the recombination parameter R = 4Nr, where N is the population size and r is the recombination rate per sequence -or between adjacent sites [44].
The animals included in this study were examined by a veterinarian with experience in small animal internal medicine. The procedures performed in these animals were approved by the Ethics in Research Committee (CEUA) of the FIOCRUZ, Rio de Janeiro, Brazil, under license number L-041/06.
Our study included indoor and feral cats from five different geographic regions in Brazil (RJ, RS, MG, SP, and PR). Diagnosis of sporotrichosis was performed by the clinical evaluation of skin lesions and confirmed by isolation of the pathogen. The suspected colonies of Sporothrix species were grown on Mycosel agar until purification of the pathogen. The fungus was easily isolated from material from the nasal, oral mucosa and skin lesions. Lesions in the cephalic region and/or respiratory tract were observed in most of the animals (Fig. 2).
Phenotypic characterization, i.e. growth at various temperatures, macroscopic and microscopic features, and carbohydrate assimilation, yielded data similar to those found for the reference strains of S. brasiliensis (CBS 120339) and S. schenckii (CBS 359.36) reported by Marimon et al. [17]. Among the 33 strains of Sporothrix isolated from cats (n = 31) and dogs (n = 2) from different geographic regions of Brazil, 32 belonged to S. brasiliensis (96.9%) and 1 to S. schenckii (3%). These phenotypic results showed that S. brasiliensis is highly prevalent among cats with sporotrichosis. The two isolates recovered of canine sporotrichosis (CBS 132994 and CBS 133004 from RS and SP, respectively) were identified as S. brasiliensis.
Using CL1 and CL2A primers we amplified 800 bp of the CAL locus. The complete alignment included 100 strains. Aligned sequences of CAL were 727 bp long, including 366 invariable characters, 214 variable parsimony-informative (29.43%), and 125 singletons. Comparison with sequences available at GenBank revealed a match of 99–100% with the type strain of S. brasiliensis (CBS 120339, AM116899) corroborating our phenotypic data. The single isolate of S. schenckii (CBS 132961) matched 99% with the S. schenckii s. str. strain (FMR 8678, AM117446) from Argentina.
Phylogenetic analysis of isolates from cats and dogs revealed that S. brasiliensis is the prevalent species (32/33); only a single isolate clustered with S. schenckii s. str. The clade of pathogenic Sporothrix species was well supported with high bootstrap and posterior probability values. The S. brasiliensis isolates recovered from animal sources clustered in a single branch together with clinical isolates, indicating that they belonged to the same genotypes and confirming that the disease is transmitted by cats. A cryptic branch was observed in the S. brasiliensis clade composed of the isolates Ss27, Ss125, Ss128, CBS 132997, CBS 132999, CBS 133000, CBS 133001 CBS 133002 and CBS 133003, supported by bootstrap and posterior probabilities values (64/66/1) (Fig. 3A).
Sporothrix brasiliensis presented low genetic diversity compared to its sister taxon S. schenckii when CAL was used as a marker. Elongation factor (EF1α) was used as marker to assess the genetic diversity in the species. All isolates presented similar fragments of 850 bp of the EF1α locus which were amplified and sequenced with primers EF1-F and EF1-R. Aligned sequences of EF1α were 707 bp long, including 639 invariable characters, 34 variable parsimony-informative (5.08%), and 33 singletons. The 100 OTUs were distributed into 7 main groups (Fig. 3B), which were congruent with the CAL phylogeny.
Judging from the EF1α dataset, the S. brasiliensis isolates recovered from animal sources in RJ and RS clustered in two branches with human clinical isolates from the same states, indicating two epidemics with distinct genotypes are concerned (Fig. 3B). Sporothrix brasiliensis presented low genetic diversity in EF1α, in accordance with results obtained for the CAL locus.
The haplotype diversity of S. brasiliensis species was assessed using the DNASp software. Only 7 haplotypes for CAL (Fig. 4A) and 3 haplotypes for EF1α (Fig. 4B) were found. The low values of haplotype (HdCAL = 0.36 and HdEF1α = 0.37) and nucleotide diversities (πCAL = 0.00152 and πEF1α = 0.00062) lead us to hypothesize that this species is clonal (Table S1). Geographical separation between the RJ and RS epidemics for the EF1α locus was clear. The median-joining network based on the EF1α haplotype showed an intraspecific separation (Fig. 4B, haplotypes H11 and H12) resulting from a nucleotide transition from A to G, between isolates from RJ and RS epidemics (Table S2). The average divergence between S. brasiliensis and its sister species S. schenckii is much higher, suggesting that the species experienced different evolutionary processes.
Recombination analysis of S. brasiliensis was first assessed by split decomposition method and no networks linking different isolates were observed in both datasets (Fig. 5), in agreement with the concept of clonal species. Also PHI-test analysis showed no evidence of recombination (pCAL = 0.757 and pEF1α = 0.903), and no recombination events were detected by DNAsp5 software. Taken together, these analyses indicated that S. brasiliensis is a clonal species.
Aiming to evaluate possible phenotypic characteristics that explain the success of this pathogen adaptation to the feline host we evaluated the thermal resistance of strains of clinical interest (human and animal) and environmental strains. Strains of S. brasiliensis from feline origin (n = 30) showed highest temperature tolerance, being inhibited 77.1±6.32% on average at 37°C (Fig. 6). The group differed statistically from other species evaluated herein (S. schenckii s. str., S. globosa, and S. mexicana), suggesting that this factor may confer advantage during the process of infection in the feline host.
Supplementary information reported in this section is complementary to the results and describe the genetic diversity of the Sporothrix isolates.
Epidemiology of fungal infections can be influenced by several factors, including: (i) biological factors such as fungal virulence and host resistance, (ii) ecological factors such as temperature, atmospheric humidity, ultraviolet radiation, geological conditions, and inter-relationships with other living beings, and (iii) socio-economic factors such as poverty, sanitation, clothing, profession, prophylactic habits and population migrations. In the Brazilian epidemic of feline sporotrichosis a combination of a highly virulent fungus and a susceptible host coupled to low sanitary conditions in the suburbs has made the state of RJ a highly endemic area of this mycosis among animals and humans. The epidemic proportions are noted only since the last two decades.
Little is known about the eco-epidemiology of feline sporotrichosis and its impact on the epidemiology of human sporotrichosis. Cats play a significant role in outbreak areas of sporotrichosis such as RJ and RS. Classically, humans can acquire sporotrichosis by cat scratches or bites, the reason why cats are considered important source of infection in the spread of the disease. In our study we found that S. brasiliensis is the prevalent etiological agent of feline sporotrichosis in Brazil. Among cats, S. brasiliensis was identified in a total of 96.9% of the samples, by isolation of the pathogen from lesions and posterior phenotypic and molecular characterization.
Interestingly, a correlation between cat outbreaks and prevalence of S. brasiliensis among humans was found in the same geographic area, such as in RJ (Table 1). This fact matches with our hypothesis that outbreaks among cats directly influence the prevalence of S. brasiliensis in human cases of sporotrichosis in the same geographic area. A similar situation was observed in the state of RS where S. brasiliensis was isolated with high frequency from cats as well as from humans.
Marimon et al. [17] analyzed 127 Sporothrix isolates using the calmodulin locus and five major clades (I–V) were recognized. The maximum likelihood, neighbor-joining and Bayesian analyses based on the calmodulin (Fig. 3A) or EF1α (Fig. 3B) loci placed our animal Sporothrix isolates in Clade I (S. brasiliensis) composed of clinical samples from the RJ State epidemic, with strong bootstrap and posterior probability support. All pathogenic Sporothrix species are known to occur in Brazil [11], but S. brasiliensis is relatively frequent among feline sporotrichosis outbreaks.
The geographic origin of S. brasiliensis of the Brazilian epidemic is difficult to establish. At least two distinct genotypes occur: one in RS and another in RJ. The latter is the oldest and longest recorded in the literature [3], [4], [23], [24]. Our data show that humans and animals infected in the RS epidemic harbor the same S. brasiliensis genotype, which is distinct from the one of the RJ epidemic. The RJ genotype is also present in the recent outbreaks in PR, MG and SP, which suggests spread of S. brasiliensis from RJ. Additionally, our results showed absence of recombination events in the CAL and EF1α loci, demonstrating that S. brasiliensis is a clonal species. Despite a recent indication of intraspecific variability within the species S. brasiliensis using RAPD [45] we believe that this phenomenon is not frequent or strong enough to break the prevalent pattern of clonal population structure, i.e., recombination or scarce exchange of genetic material may occur in some point of the evolutionary course of the pathogen life without compromise or affect its population structure. This hypothesis has been discussed by Tibayrenc and Ayala [46] through different group of pathogens including fungi.
The existence of clonal populations has repeatedly been proven in fungal pathogens [47]–[50], although most of these species are surmised to have occasional sexuality in any phase of their life cycle. Under permissive conditions, most fungi reproduce very effectively by asexual propagation. Sexual reproduction provides advantages to the pathogen under adverse conditions, generating suitable genotypes that enhance survival. Many fungal epidemics are driven by populations showing low levels of genetic diversity, as demonstrated in Penicillium marneffei [51], [52], Cryptococcus gattii [53], [54] and Batrachochytrium dendrobatidis [55]. Also feline and human sporotrichosis in Brazil caused by S. brasiliensis is driven by the spread of a clonal species. In contrast, outbreaks of other human pathogens such as Coccidioides immitis [56]–[58] and Paracoccidioides brasiliensis [59]–[61], spread by a diversity of genotypes.
The ecological aspects of the pathogenic species within the genus Sporothrix needs to be reevaluated, and this information can be crucial to find the source of S. brasiliensis in nature. Classically, soil [5], thorny plants [62], Sphagnum moss [63]–[65] and hay [66] have been pointed as source of S. schenckii s.l. To date, just a single environmental isolate (FMR 8337) of S. brasiliensis was isolated and reported from domiciliary dust in Brazil [17], [19]. Distant relatives of Sporothrix in the fungal order Ophiostomatales are mainly associates of bark beetles on woody plants [67], [68]. Zhou et al. [13] demonstrated that different ecologies are corroborated by phylogenetic separation.
It is challenging to obtain environmental isolates of S. brasiliensis, and the low number of subjects contaminated with propagules from soil or woody plants is indeed low compared to the high occurrence in warm-blooded hosts [3], [69], [70]. This suggests successful transmission among animals (cat-cat and cat-humans). This scenario is different from epidemics occurring in South Africa [71], [72], India [10], [73], the USA [63], [64], Australia [66], [74], China [75], and Japan [76], where patients are mainly infected through soil and decaying wood. Possibly the Brazilian epidemics of S. brasiliensis are related to the emergence of a pathogenic clone front of a highly susceptible feline host, rather than to an increase in population size of S. brasiliensis in nature. This is corroborated by the high degree of virulence observed in naturally infected cats in the outbreak area [24], as well as demonstrated in a murine model [21]. Besides that, we do not discharge the hypothesis that the emergence of pathogenicity could also be attributed to a recent host-shift from an unknown host to cats as discussed in other groups of pathogens [77]–[79]. Feral cats present a great potential to spread the disease in a short period of time due to their mobility and digging behavior, whereas dispersal from soil or vegetable remains is ineffective.
Classically, accumulation of mutations in fungal populations can lead to speciation processes. However, rapid emergence of a new, highly virulent pathogen which is able to explore different ecological niches may result from other processes than those observed in natural selection. In many plant-pathogenic fungi, such as Fusarium and Alternaria, pathogenicity is determined by mobile, dispensable small chromosomes [80], [81]. Genetic processes such as hybridization of two distinct, sympatric species [82], parasexual recombination [83], [84] or mechanisms of inactivation/activation of virulence genes by insertion of transposons [85] can also drive the emergence of pathogenicity. Hybridization is one of the possible mechanisms of emergence of phytopathogenic fungi [86], [87] as well as fungi pathogenic to animals [88]. It has also been discussed in the genus Ophiostoma, which is phylogenetically related genus to Sporothrix [89]. All these genetic processes, alone or in combination, may be the reason of the emergence of virulence in the species S. brasiliensis. The lack of variation in the populations of S. brasiliensis also may be the result of a strong selective pressure imposed by the feline host. Presence of opposite mating types and sexual reproduction leads to genetic recombination and may increase fitness and widen host ranges. So far, no evidence of sexual recombination was demonstrated experimentally for the species from the S. schenckii complex and this fact, combined with the hostile selective pressure of the cats may provide possible explanations for the lack of diversity in S. brasiliensis.
The association of S. brasiliensis with cats may play an important role in the evolution and spread of this pathogen. The interaction between cats and S. brasiliensis is not an exclusive relationship, since S. schenckii s. str. was also found in the feline host. However, S. brasiliensis has become predominant in this host within less than a decade, indirectly indicating a recent adaptation to the conditions of the feline body. Therefore, cats represent a natural habitat for S. brasiliensis. In contrast to the situation in opportunistic fungi, Sporothrix species are able to escape from the host and be transmitted to the next host, which is one of the hallmarks of a pathogen. Transmission is either direct during fights, or indirect, the fungus returning to soil after the cat has died.
Given the role of the mammal host in Sporothrix evolution, variance in fitness between clonal lineages of S. brasiliensis is expected to lead to populations that are better adapted to host conditions. For example, the body temperature of the feral cat Felis catus is around 38–39°C, depending on its activity [90]. Interestingly, S. brasiliensis has the best rate of vegetative growth when incubated at 37°C, followed by S. schenckii s. str. (Fig. 6). Remaining species of Sporothrix such as S. globosa and S. mexicana appear to be more sensitive to temperature, having a maximum around 35°C. The cat's body temperature could be considered an important selective pressure event, selecting thermo-resistant strains during sporotrichosis outbreak episodes. Transmission of S. brasiliensis by cats promotes inoculation into human hosts of yeast cells of rather than of hyphae and conidia, yeast cells having been reported to be more virulent [6].
The endotherm developed by mammals is a natural defense mechanism against pathogens [91]–[93], and in our study this factor appears to restrict the occurrence of species of the S. schenckii complex that are sensitive to temperatures above 35–37°C [11], [17].
Another important factor in understanding the success of the epidemic of sporotrichosis among cats in RJ, has a socio-economic character. Most cat owners are living in neglected areas and many abandon dead animals in the street [94], favoring contact with other feral cats, or simply bury their pets after death in their backyard or in nearby wastelands. This directly allows the return of the agent into the environment, increasing outbreak risks of the pathogen, and enhancing the spread of the clonal species. In an epidemic scenario, domestic pets such as cats and dogs are the first animals to become infected with the fungus. Subsequently human cases of sporotrichosis are likely to emerge. Thus, we believe that cats can act as sentinel animals for epidemiological services, and its notification should be compulsory by regulatory agencies as the Centers for Zoonosis Control. The predominance of a species that is highly virulent to humans and animals requires fast implementation of public health policies to contain the epidemic, lowering harmful effects to the population.
|
10.1371/journal.pntd.0006594 | Evidence of vertical transmission of Zika virus in field-collected eggs of Aedes aegypti in the Brazilian Amazon | Arboviruses are viruses transmitted to humans and other animals by the bite of hematophagous arthropods. Infections caused by chikungunya virus (CHIKV), dengue virus (DENV), Zika virus (ZIKV), and the deadlier yellow fever virus (YFV) are current public health problems in several countries, mainly those located in tropical and subtropical regions. One of the main prevention strategies continues to be vector control, with the elimination of breeding sites and surveillance of infested areas. The use of ovitraps for Aedes mosquitos monitoring has already demonstrated promising results, and maybe be also useful for arboviral surveillance.
This work aimed to detect natural vertical transmission of arboviruses in Aedes aegypti and Aedes albopictus. Mosquito egg collection was carried out using ovitraps in Itacoatiara, a mid-size city in Amazonas state, Brazil. Collected eggs were allowed to hatch and larvae were tested for CHIKV, DENV, and ZIKV RNA by RT-qPCR.
A total of 2,057 specimens (1,793 Ae. aegypti and 264 Ae. albopictus), in 154 larvae pools were processed. Results showed one positive pool for CHIKV and one positive pool for ZIKV. The active ZIKV infection was further confirmed by the detection of the negative-strand viral RNA and nucleotide sequencing which confirmed the Asian genotype. The Infection Rate per 1,000 mosquitoes tested was assessed by Maximum Likelihood Estimation (MLE) with 0.45 and 0.44 for CHIKV and ZIKV, respectively, and by Minimum Infection Rate (MIR) with 0.45 for both viruses.
To our knowledge, this is the first detection of ZIKV in natural vertical transmission in the Ae. aegypti, a fact that may contribute to ZIKV maintenance in nature during epidemics periods. Furthermore, our results highlight that the use of ovitraps and the molecular detection of arbovirus may contribute to health surveillance, directing the efforts to more efficient transmission blockade.
| The control of the vast majority of arbovirus infections relies on entomological measures to reduce mosquito infestation. Therefore, this study analyzed the use of ovitraps for arboviral surveillance in a mid-size city of the Amazonas state, Brazil. We found one larva pool infected with chikungunya virus, before the first human case confirmed in this municipality. Another pool was infected with Zika virus, demonstrating the first evidence that vertical transmission occurs in naturally infected Aedes aegypti mosquito populations.
| The arboviruses transmitted by mosquitoes of the genus Aedes, like chikungunya virus (CHIKV), dengue virus (DENV), Zika virus (ZIKV), and yellow fever virus (YFV) have reached threatening numbers in the last years, with a huge impact on public health systems in several countries throughout the world [1–7]. Nevertheless, the detection and identification of circulating arboviruses are most often taken from human cases, mainly when an outbreak is already in place. With the fast-worldwide expansion of new emerging or reemerging arboviruses such as CHIKV, DENV, and ZIKV, the need to establish the role of each mosquito species in the spread of these pathogens is clear. This knowledge is fundamental to the implementation of effective surveillance and control measures against these vectors in order to avoid the early establishment of an arboviral epidemic [8].
The first report of natural infection of mosquitoes with ZIKV in Brazil was from Aedes aegypti adults collected in Rio de Janeiro [9] and this species was also considered as the primary vector during the epidemic. Indeed, the detection of ZIKV in Ae. aegypti mosquitos occurred soon after the emergence of ZIKV in this city.
A recent study showed the first description of Ae. aegypti infected with CHIKV ECSA genotype in Brazil [10]. These authors reinforce the role of this species as an important vector of CHIKV in urban areas of northeastern Brazil and emphasize the benefits of entomological surveillance programs for public health, regarding the immediate implementation of diseases prevention.
A study conducted in the Amazonas State, Brazil, demonstrated that the entomological surveillance using ovitraps could be successfully used to monitor the different DENV serotypes circulating in the municipalities of the interior of the state. Moreover, this study suggests that the use of arboviral monitoring strategies in routine surveillance helps for early detection of virus circulation before outbreaks, contributing to more efficient and effective control measures [11].
The continuous monitoring of Ae. aegypti infestation, associated with the early detection of arbovirus circulation, may contribute to the development of epidemic prediction models for diseases transmitted by this vector. Besides, vertical transmission, even at a low rate, contributes to the preservation of an arbovirus in nature, without a well-known cycle involving invertebrates and vertebrates, and may be of paramount importance in endemic areas as an alternative arboviral maintenance mechanism [12,13].
In the present study, we monitored and detected the natural vertical transmission of arboviruses in a mid-size city of the Amazonas State, Brazil, during the emergence of ZIKV in 2015–2016.
The collection of Aedes eggs doesn't require special permission in Brazil. All the house owners agreed with and allowed the installation of ovitraps in their properties.
The study area was the city of Itacoatiara in the Amazonas State, located in Northern Brazil. Amazonas is the largest state of the Brazilian federation with 1,559,161 km2 and an estimated population of 4,063,614 inhabitants (2.23 inhabitants/km2). Amazonas has international borders with Venezuela, Colombia, and Peru, and domestic borders with the states of Roraima, Pará, Mato Grosso, Rondônia, and Acre. Itacoatiara (latitude 03° 08' 35" S; longitude 58° 26' 39" W) is located 189km from the capital Manaus and is the third most populous municipality in the Amazonas with an area of 8,891.9 km2 and a population estimated at 100,000 inhabitants (9.77 inhabitants/km2). The predominant climate is equatorial (tropical monsoon), characterized by high temperatures and a significant rainfall over ≥ 2500 mm per year, but with a dry season also known as Amazonian winter. An important fluvial harbor is located in Itacoatiara for the transport of inhabitants and agricultural products [14].
Our group has been using ovitraps to monitor Aedes spp. infestation. The ovitraps consist of a dark plastic container with a capacity of 700 mL containing 300 mL of a 0.04% brewer's yeast solution as an attractant for mosquito females, and an oviposition substrate on which the eggs are laid (Eucatex, Brazil) with the rough part facing the inner area of the trap for oviposition. On each monitoring cycle, a total of 100 ovitraps were installed around selected houses in the city, with a weekly exchange of the pallets. Ovitraps distribution was equidistant on every 200 meters, covering the entire inhabited area of the city.
All of the properties where the ovitraps were installed were georeferenced with the use of a GPS device (Garmin Ltd, USA#MP62SC) with UTM—SIRGAS 2000 projection. The coordinates were inserted into a geographic information system (GIS) in the software QGIS 2.16.2, where each property had its spatial location identified as a point (attribute) in a layer (shape) of the city.
Initially, collected pallets were sent to the Itacoatiara entomology laboratory, placed to dry and analyzed under a stereoscopic magnifying glass for Aedes spp. eggs count. Once every 2 months, the positive pallets were sent to the entomology laboratory of the Fundação de Vigilância em Saúde do Amazonas (FVS-AM), where the pallets were individually immersed in glass flasks containing 200 mL of dechlorinated water for egg hatching.
The larvae were raised until the third stage when species were identified and separated into pools of a maximum of 30 specimens, placed in cryotubes, and sent frozen to Instituto Leônidas e Maria Deane (ILMD)–Fiocruz Amazônia, where they were kept in a -80°C freezer, until molecular analysis for arboviral RNA detection.
Firstly, each larvae pool was spiked with the Escherichia coli bacteriophage MS2 (ATCC 15597-B1) to be used as an internal positive control (IPC) using the same conditions previously described [15]. All pools were individually disrupted in 2 mL microtubes containing a 5 mm stainless-steel bead and 250 μL of TRIzol Reagent (Invitrogen, USA, #15596026) with the aid of the TissueLyser LT bead mill (Qiagen, Germany, #85600), 50Hz for 5 minutes. The reservoir containing the microtubes was frozen and kept in an ice bath during the process. Posteriorly, the homogenized pools were clarified by centrifugation, and the supernatant was added with 750 μL of Trizol. Therefore, RNA extraction followed the manufacturer’s recommendations. The RNA pellet was resuspended in 40 μL of nuclease-free water and evaluated for quantity and quality with the BioDrop DUO UV spectrophotometer (BioDrop Ltd, United Kingdom, #80-3006-61). A total of 2.5 μL microliters of the extracted RNA was used for the RT-qPCR assays. The remaining volume was stored at -80°C for further analyses.
All samples were evaluated for the detection of three arboviral RNAs by RT-qPCR in a StepOnePlus Real-Time PCR System (Applied Biosystems, USA, #4376598) located at the Real-Time PCR Platform of ILMD. The protocols used for the detection of each virus were previously published: DENV [16]; CHIKV [17]; and ZIKV [18], but we conducted the assays with some modifications. All probes were used at a final concentration of 0.1μM, whereas all primers were used at a final concentration of 0.3μM. All reactions were performed with TaqMan Fast Virus 1-Step Master Mix (Applied Biosystems, USA, #4444432), following manufacturer’s recommendations, except for the number of cycles that was increased to 45. For each lot of analyzed samples, three blank reactions (nuclease-free water as the template) and external positive controls (RNA extracted from viral culture) were included. The number of viral copies in each positive sample was estimated by RT-qPCR using absolute quantification by the standard curve method and reported as viral RNA copies/μL (of the eluted RNA).
With the objective to evaluate the ZIKV replication in naturally infected larvae, we conducted the same RT-qPCR assay used for detection of ZIKV RNA, but in a two-step protocol targeting the negative-strand RNA. Firstly, two different cDNA assays were made: I) with only the reverse primer (which hybridizes to positive-strand RNA) and II) with only the forward primer (which hybridizes to negative-strand RNA). All cDNAs were made with SuperScript IV Reverse Transcriptase (Invitrogen, Carlsbad, CA, #18090050), according to the manufacturer’s recommendations.
Subsequently, a qPCR assay was conducted with TaqMan Fast Advanced Master Mix (Applied Biosystems, USA, #4444558), following manufacturer’s recommendations, except for the number of cycles that was increased to 45. Three different master mixes we used: I) with only the reverse primer; II) with only the forward primer and III) with both reverse and forward primers.
The positive pools were submitted to a conventional RT-PCR amplification protocol using the primers ZIKA_ASIAN_FNF1 (5’–CCGCGCCATCTGGTATATGT– 3’) and ZIKA_ASIAN_FNR (5’–CTCCACTGACTGCCATTCGT– 3’) designed to target the NS5 coding region of Asian ZIKV lineages. For the CHIKV sample, we used a protocol already described for alphaviruses amplification [19].
Thereafter, amplicons were precipitated with PEG-8000 and submitted to the nucleotide sequencing reaction with BigDye Terminator v3.1 Cycle Sequencing Kit. Capillary electrophoresis was conducted in an ABI3130 sequencer located at the genomics platform of ILMD, Fiocruz Amazônia. The final FASTA sequences were initially submitted to BLAST analysis [20] and further evaluated by a web-based Dengue, Zika and Chikungunya Subtyping Tool (Version 1.0), freely available at http://bioafrica.mrc.ac.za/rega-genotype/html/index.html.
The virus infection rate (IR) was calculated with PooledInfRate, version 4.0 by Biggerstaff, a Microsoft Excel add-in that computes the IR using data from pooled samples (even with different pool sizes) by both Minimum Infection Rate (MIR) and Maximum Likelihood Estimation (MLE) methods. Freely available at https://www.cdc.gov/westnile/resourcepages/mosqsurvsoft.html.
Between January and April 2016, 154 larvae pools containing 2,057 specimens (1,793 Ae. aegypti and 264 Ae. albopictus) were analyzed for CHIKV, DENV, and ZIKV RNA. By using the RNA extraction method described under Material and Methods we were able to obtain high quality RNA for most samples (260/280 median value: 1.96; IQR: 1.87–2.01) and all IPC reactions were positive (Ct median value: 30.9; IQR: 30.0–31.4).
One pool containing only one larvae, obtained from the trap P056/ITA, collected at 15-Feb-2016, was positive to ZIKV (47 viral RNA copies/μL, approximately 1.8 x 103 ZIKV RNA copies in the infected larvae). Another pool containing three larvae, obtained from the trap P026/ITA, collected at 25-Feb-2016, was positive to CHIKV (2 viral RNA copies/μL) Fig 1.
The positive ZIKV pool was further evaluated for productive infection by the specific detection of the negative-strand RNA. We detected amplification for both cDNAs, derived from the positive or negative-strand RNA, with a positive-to-negative strand RNA ratio of approximately 2:1, represented by the difference of one Ct (Table 1).
Subsequently, conventional RT-PCR amplification was performed on both positive pools as described in Material and Methods, but unfortunately, only the Zika sample could be amplified. The final ZIKV sequence corresponds to a fragment of 450bp that was deposited in GenBank under the accession number MG279550. The analysis of the viral genotyping confirmed the Asian ZIKV lineage.
The virus IR per 1,000 mosquitoes tested was calculated for CHIKV and ZIKV based in the total of pools and individuals tested. The CHIKV IR was 0.45 for both MIR and MLE methods (MIR: lower Limit = 0.00; upper limit = 1.32 and MLE: lower Limit = 0.03; upper limit = 2.15). The ZIKV IR was 0.45 for MIR (MIR: lower Limit = 0.00; upper limit = 1.32) and 0.44 for MLE (MLE: lower Limit = 0.03; upper limit = 2.15).
The first report about the detection of Ae. aegypti in Manaus, the capital of Amazonas State, was in November 1996 and for Ae. albopictus in September 1997. Since then, Aedes spp. mosquitos began to be found in other municipalities of the interior of the Amazonas State, and successive Aedes-related arboviral infections have been reported [21–25].
Between 2015 and 2016, a total of 878 suspected cases of CHIKV and 4,485 cases of ZIKV were reported in the Amazonas State [26,27]. Preventive and reactive measures regarding the vector control were carried out by the health authorities under the coordination of FVS-AM, to decrease Aedes infestation in different municipalities. In addition to the routine actions, ovitraps were installed with the purpose of directing the field efforts more efficiently.
The primary aim of our study was to evaluate natural arboviruses vertical transmission in the field. Since this study was conducted under circumstances where the cold-chain could not be guaranteed, we decided to use a study design that favored the molecular detection, protecting viral RNA as soon as possible. Therefore, we decided to disrupt the Aedes larvae directly in Trizol reagent, which consists of a solution of phenol and guanidinium isothiocyanate that concurrently solubilizes biological material and immediately inactivates RNases [28], but also inactivates viral particles, preventing the possibility of viral isolation.
The present study confirms other studies demonstrating natural vertical transmission of CHIKV [29–31], in a municipality with no human case previously confirmed. It is important to emphasize that most acute febrile cases in Brazil are diagnosed by clinical examination, without specific laboratory confirmation, especially during an ongoing outbreak. At the time we collected our samples, a Zika outbreak was already established, which could compromise the diagnosis of chikungunya cases. Although the CHIKV positive sample was amplified in duplicate in the probe-based RT-qPCR assays, we were unable to amplify this sample using conventional RT-PCR, preventing its sequencing. RT-qPCR would have been more sensitive compared to conventional RT-PCR, particularly for samples with low viral load. Another important point is that the sequence variation at the primers sites that may also decrease the efficiency of nucleic acid amplification methods.
In an experimental study of vertical ZIKV infection, a total of 69 pools of Ae. aegypti adult mosquitos (F1) were tested, and six were positive in an indirect immunofluorescent antibody assay [32]. While other studies have demonstrated the natural vertical transmission of DENV [11,33–37], the detection of ZIKV in naturally infected larvae had not yet been described.
We report the first detection of Zika virus vertical transmission in an Ae. aegypti larvae under the natural conditions found in the field. Therefore, we evaluated if there was a productive infection in the Ae. aegypti larvae by strand-specific amplification of viral RNA. ZIKV is a positive-sense, single-stranded RNA virus that belongs to the Flaviviridae family. During flaviviruses replication, a complementary negative-strand RNA is produced, which is used as a template for synthesizing new positive-strand RNA copies. Viral replication progresses asymmetrically, producing more positive-strand than negative-strand RNA. The positive-strand RNA molecules are packaged into the virions, acts as templates for viral protein production and promotes evasion of innate cell response [38,39]. Therefore, the specific detection of negative-strand flavivirus RNAs is an indicator of active viral replication, and different studies have been using this approach [40–43].
In this study, ZIKV cDNA was produced with the reverse primer, which hybridizes with the positive-strand RNA, e.g., the genomic RNA found into the virion particle and also detected during viral replication, or with the forward primer, which hybridizes with the negative-strand RNA, only found during replication. The qPCR results clearly showed that both cDNAs were amplified in a positive-to-negative strand RNA ratio of approximately 2:1.
To the best of our knowledge, there is only one plausible explanation that could explain the detection of ZIKV RNA in the larvae, besides a canonical viral infection. Recently, different studies showed that naked viral RNA from hepatitis C virus (HCV), as well as from human pegivirus, two members of the Flaviviridae family (genus Hepacivirus and Pegivirus, respectively), may spread infection in exosomes vesicles [44,45]. Importantly, these studies showed the phenomenon in vitro, and further investigations are required to prove if similar events occur during ZIKV infection in vivo. Furthermore, the same studies also showed that, regardless of the way of viral RNA release in the cytoplasm of newly “infected” cells, productive viral RNA replication was observed, leading to the release of infectious particles. Given this, although our study design does not allow us to assert if ZIKV RNA reaches larvae cells by a classic route of viral infection, the specific detection of the negative strand RNA provides substantial evidence that active viral replication has occurred in Ae. aegypti naturally infected larvae.
We detected arboviral RNA in larvae, which does not necessary means that larvae would become infected adults due to possible transstadial loss of infection during development to adulthood. On the other hand, if some of the infected larvae achieve maturity still infected, they will be readily able to transmit ZIKV to other mosquitos by venereal transmission [46] or, in the case of females, to human hosts. Thus, this phenomenon may contribute to the epidemic potential of this arbovirus because mosquitoes that emerge as virus-infected adults will have more opportunity to transmit virus than mosquitos that become infected after blood meal in an infected vertebrate. Further studies are necessary to evaluate all variables contributing to maintaining a virus circulating in a specific area until the number of new susceptible human subjects raise, by immigration or births, sufficiently to support a new epidemic cycle.
According to previous work, the rapid detection of arbovirus in specimens collected in the field may contribute to the effectiveness of vector control measures, decreasing the viral transmission among the human population [47]. Altogether, the results showed in the present manuscript strengthen the importance of continuous monitoring of arboviral infections in both mosquitoes, as well as in human hosts, before the establishment of a new outbreak.
|
10.1371/journal.pcbi.1006285 | Learning the payoffs and costs of actions | A set of sub-cortical nuclei called basal ganglia is critical for learning the values of actions. The basal ganglia include two pathways, which have been associated with approach and avoid behavior respectively and are differentially modulated by dopamine projections from the midbrain. Inspired by the influential opponent actor learning model, we demonstrate that, under certain circumstances, these pathways may represent learned estimates of the positive and negative consequences (payoffs and costs) of individual actions. In the model, the level of dopamine activity encodes the motivational state and controls to what extent payoffs and costs enter the overall evaluation of actions. We show that a set of previously proposed plasticity rules is suitable to extract payoffs and costs from a prediction error signal if they occur at different moments in time. For those plasticity rules, successful learning requires differential effects of positive and negative outcome prediction errors on the two pathways and a weak decay of synaptic weights over trials. We also confirm through simulations that the model reproduces drug-induced changes of willingness to work, as observed in classical experiments with the D2-antagonist haloperidol.
| The basal ganglia are structures underneath the surface of the vertebrate brain, associated with error-driven learning. Much is known about the anatomical and biological features of the basal ganglia; scientists now try to understand the algorithms implemented by these structures. Numerous models aspire to capture the learning functionality, but many of them only cover some specific aspect of the algorithm. Instead of further adding to that pool of partial models, we unify two existing ones—one which captures what the basal ganglia learn, and one that describes the learning mechanism itself. The first model suggests that the basal ganglia weigh positive against negative consequences of actions according to the motivational state. It hints how payoff and cost might be represented, but does not explain how those representations arise. The other model consists of biologically plausible plasticity rules, which describe how learning takes place, but not how the brain makes use of what is learned. We show that the two theories are compatible. Together, they form a model of learning and decision making that integrates the motivational state as well as the learned payoffs and costs of opportunities.
| What guides rational behavior in a complex environment? Certainly, knowledge of the typical payoffs and costs of acting a certain way is critical for successful action selection. Those payoffs and costs do not only depend on the action that is carried out, but also on the environmental state, henceforth referred to a ‘situation’. If payoffs and costs are represented separately in the animal’s brain, they can be weighted depending on animal’s motivational (i.e. internal) state, which can vary independently of the environmental situation. For example, consider the action ‘harvesting fruit from a tree’ in the situation ‘close to a fruit-bearing tree’. It has a payoff connected with the nutrients in the fruit, but also costs related to the effort, the risk of pain and the exposure to predators associated with climbing a tree. The nutrients in the fruit are only valuable for the animal if it is hungry, i.e. if it is in the appropriate internal state. So, when it is hungry, the payoffs of climbing a tree which was identified as fruit-bearing should be weighted more than the costs, to ensure that the animal searches for food. By contrast, when the animal is not hungry at all, the payoffs should be weighed less than the costs, to make sure that it does not climb the tree without necessity. In summary, the payoffs and costs of a specific action (‘climbing a nearby tree’) carried out in a certain environmental situation (‘near fruit-bearing tree’) should be weighed against each other according to the motivational state (‘hunger’) to correctly asses the action’s utility.
In all vertebrates, an important role in this process of action evaluation and selection is played by a set of subcortical structures called the basal ganglia [1]. The basal ganglia are organized into two main pathways shown schematically in green and red in Fig 1. The Go or direct pathway is related to the initiation of movements, while activation of the No-Go or indirect pathway results in targeted movement inhibition [2]. These two pathways include two separate populations of striatal neurons expressing different dopaminergic receptors [3]. The striatal Go neurons express D1 receptors and are excited by dopamine, while the striatal No-Go neurons express D2 receptors and are inhibited by dopamine [4]. Thus dopamine changes the balance between the two pathways and promotes action initiation over inhibition.
The competition between Go and No-Go pathways during action selection and the role of dopaminergic modulation are subject of many interpretations and models, e.g. [5–7]. In particular, the Opponent Actor Learning (OpAL) hypothesis suggests that the Go and No-Go neurons specialise in encoding the values of actions with positive or negative consequences respectively [8]. We extend the OpAL hypothesis further by proposing that for each individual action, the direct and indirect pathway separately encode the learned positive and negative consequences. As the dopaminergic neurons modulate the Go and No-Go neurons in opposite ways, dopamine controls the extent to which positive and negative consequences affect the activity in the thalamus, through the output of the basal ganglia [8]. For example, when motivation is high, the dopaminergic neurons will excite the Go neurons and inhibit the No-Go neurons. Consequently, positive action values will influence the decision more than negative action values. By contrast, when the motivation is low, the Go neurons tend to be excited to a smaller degree, but the No-Go neurons will be released from inhibition, such that negative values are weighted stronger.
Much research has also focused on how the synapses of Go and No-Go neurons are modified by experience. A systematic investigation revealed that bursts of activity of dopaminergic neurons encode outcome prediction errors, which measure the difference between outcome (typically rewards) obtained and expected [9, 10]. Note that we use the phrases ‘outcome prediction error’ and ‘reinforcement’ instead of the more common ‘reward prediction error’ and ‘reward’ respectively. This use of language emphasizes that in our theory, the feedback signal is informative of both positive and negative events and that not only rewards but any outcome will be compared with predictions. That perspective is well supported by experimental results; see Discussion for a review of evidence for negative prediction errors (e.g. pauses in dopaminergic firing) caused by negative experiences.
Such bursts of dopaminergic activity produce distinct changes in the synaptic weights of Go and No-Go neurons [11]. Several computational models have attempted to describe the learning process of the synapses of Go and No-Go neurons [12–15]. Among these models, the OpAL model provided simple and analytically tractable rules describing the changes in weights of Go and No-Go neurons as a function of outcome prediction errors [8]. However, no-one so far examined how the basal ganglia might estimate payoff and cost if they are both associated with the same action.
The goal of this paper is to show how the Go and No-Go neurons can learn the payoffs and costs of individual actions through local synaptic plasticity rules. We argue that the payoffs and costs of individual actions are not necessarily correlated (for instance, two actions might have comparable benefits, but very different costs), and strive to construct a model that is able to represent those independent dimensions of reinforcement for every single action. Ultimately, we confront the resulting model with experimental results.
Instead of constructing a new set of learning rules from scratch, we will employ the theory of striatal learning described in [16], which has been shown to account for diverse observations. That theory was originally developed to explain how the mean and the spread of the reinforcement signal could be learned by the basal ganglia network. In this article, we will prove that if the weights of Go and No-Go neurons change according to these rules, they can eventually represent payoff and cost. In summary, we show that a set of learning rules, originally constructed to estimate statistical properties of the reinforcement signal, can be reinterpreted as rules to estimate payoffs and costs. We thus extend both the interpretation of the striatal pathways of Collins and Frank and the striatal learning rules of Mikhael and Bogacz to ultimately obtain a consistent theory of learning the payoffs and costs of actions.
According to the experimental and modeling work mentioned above, dopaminergic activity encodes both information about motivational state and the outcome prediction error. However, if the dopaminergic neurons carried both signals, the striatal neurons would need a way to decode each signal and react appropriately, i.e. change their activity according to the motivation signal, and change the synaptic weights according to the prediction error. The prominent suggestion that motivation might be encoded in the average or tonic dopamine level, and outcome prediction errors in the burst or phasic activity [17] is hotly debated; it seems to be contradicted by the observation of fast-changing dopaminergic activity that encodes motivation [18–20]. Note, though, that these apparently divergent views could potentially be reconciled–see e.g. [21]. Anyhow, the motivation and teaching signals could both be provided by other means. For example, the activity of striatal cholinergic neurons may inform what the dopaminergic neurons encode at the moment [20]. In this paper, we assume that striatal neurons can read out both motivation and teaching signals encoded by dopaminergic neurons, and we leave the details of the mechanisms by which they can be distinguished to future work.
Inspired by the OpAL model [8], we assume that synaptic weights within the Go pathway encode positive consequences of actions, that is the positive reinforcement caused by food, drink or other appetitive stimuli obtained through actions. More precisely, we claim that the typical payoff of a particular action a in a particular situation s is encoded in the strength of the connections from the cortical neurons selective for the situation to the striatal Go neurons selective for the action. We denote these weights by G(s, a) (see Fig 1), and propose that after learning, the weights G represent the mean payoff for an action. Mathematically, the collective strength of the weights G corresponds to a single, non-negative number. The negative consequences, on the other hand, are encoded in the synaptic connections of striatal No-Go neurons. Negative consequences should be understood as the negative reinforcement induced by aversive stimuli such as pain, effort or disgust. We denote their weights by N(s, a), and propose that after learning, they represent the mean cost of an action. Just as with G, we mathematically represent the collective strength of the weights N by a single, non-negative number.
To learn the positive and negative consequences of actions respectively, the striatal neurons can take advantage of the fact that these consequences typically occur in different moments in time. Let us consider a situation in which an animal performs an action that involves an effort in order to obtain a reward: Fig 2a sketches a task in which a rat is given the opportunity to press a lever in order to obtain a food pellet. Due to the effort, the instantaneous reinforcement during the course of this action is negative at first, while pressing the lever. Then, it turns positive at the time the payoff is received. Fig 2b sketches the resulting changes in the synaptic weights. The leftmost display shows the initial weights. While making an effort to perform an action, the outcome prediction error is negative. Similarly as in previous models [8, 12], we assume that the negative prediction error results in a strengthening of N (compare the red arrows in the middle and the left displays in Fig 2b). This allows the weights N to encode negative consequences. Later, reception of the payoff causes a positive prediction error, which strengthens G. This leads the weights G to encode the positive consequences. Here, we assumed that–at baseline dopamine level–positive prediction errors trigger more plasticity in the Go pathway than in the No-Go pathway, while negative prediction errors affect the No-Go pathway more than the Go pathway. In Discussion, we will review data suggesting that the properties of D1 and D2 receptors allow this assumption. Generally, if an experience involves both positive and negative consequences, both weights are increased during the experience (compare the right and the left displays in Fig 2b).
To mathematically implement these ideas, we need to model the weighs of the Go pathway G(s, a), the weighs of the No-Go pathway N(s, a), and the prediction error. The outcome prediction error, which we denote by δ, quantifies the difference between the expected reinforcement and the received reinforcement r after executing action a in situation s. If r is negative, we shall speak of cost, and when r is positive, we shall speak of payoff or reward. The expected reinforcement, on the other hand, directly corresponds to the expected payoffs and costs, which–according to our theory–are represented by the synaptic weights G and N. We take the expected reinforcement to be the average over the expected payoff and the expected cost. Altogether, we model the outcome prediction error as
δ = r - 1 2 ( G ( s , a ) - N ( s , a ) ) . (1)
It should be clarified that this definition of the prediction error differs from one in the original model [16], in that we introduced here a factor 1/2. This factor allows G and N to converge to the exact payoffs and cost, and not to values proportional payoffs and costs, and hence increases the clarity of the exposition. However, since value cannot be measured directly, the overall scaling of values through this factor is not observable, but a mere convention.
Equipped with the quantities δ, G and N, we can formulate our theory of learning payoff and cost. To present the theory, we simply describe how the collective connection strengths G(s, a) and N(s, a) change when a prediction error δ is received after executing action a in situation s; we use ΔG(s, a) and ΔN(s, a) to denote the changes in relevant connection strengths. Note that any update only applies if the resulting weights are still positive—if an update would render a weight negative, that weight is set to zero instead. In all other cases, we follow Mikhael and Bogacz [16] in prescribing
Δ G ( s , a )= α f ϵ ( δ ) - λ G ( s , a ) (2)
Δ N ( s , a )= α f ϵ ( - δ ) - λ N ( s , a ) , (3)
where α is the learning rate, ϵ is the slope parameter and λ the decay rate. The slope parameter ϵ controls the strength of the nonlinearity exhibited by the function fϵ, which we introduce in Fig 3d and 3e. The nonlinearity of the function fϵ accounts for the fact that positive and negative prediction errors affect the weights differently. From here on, we drop the dependency of G and N on a and s to simplify notation. The dependency is still implicitly assumed unless stated otherwise.
There is a normative intuition for each term in the rules Eqs 2 and 3. These intuitions are most easily gained by following through a couple of steps to reconstruct the rules from scratch. We may start by observing that several models of learning in Go and No-Go neurons assume the effect of the prediction error on G to be opposite to its effect on N [7, 8]. We thus propose that ΔG and ΔN might simply be proportional to the prediction error and its negative, respectively. To see whether this proposal works, we formulate it mathematically and simulate the learning of an alternating sequence of costs −n and payoffs p. Fig 3a shows both the mathematical formulation and the simulation. There is a problem: the strengthening of N due to negative prediction error, caused by the cost, is always immediately reversed by the following positive prediction error caused by the payoff. The same is true for the changes in G. As illustrated by the simulation, there is no net effect of learning.
To overcome this problem, we proceed by damping the impact of negative prediction errors (which are usually caused by costs) on G, and the impact of positive prediction errors on N, by introducing a nonlinear transformation of the prediction errors. This transformation is given in form of a piecewise-linear function fϵ, defined and depicted in panels d and e of Fig 3. The transformation leaves positive prediction errors invariant (fϵ (δ) is just the identity for δ > 0) but reduces the impact of negative prediction errors by scaling them down (for δ < 0, fϵ (δ) is linear with slope ϵ < 1). Hence, fϵ introduces a pathway-specific imbalance between learning from positive prediction errors and learning from negative prediction errors (which, as we point in Discussion, is in accordance with the properties of dopaminergic receptors on these pathways). For the construction at hand, it is also logical, since costs should not alter the estimate G of the payoffs and vice versa. Such damping can be achieved by replacing the simple proportionality to δ in the first proposal by a nonlinear dependence, mediated by the functions depicted in Fig 3e. We update our mathematical formulation accordingly, and again simulate the effects of the previously used reinforcement sequence—both these steps are illustrated in Fig 3b. The simulation shows that, while producing the appropriate tendencies, these rules cause unconstrained, ongoing strengthening of both connections. Such dynamics are neither biologically plausible nor useful to infer the actual payoff and cost.
Finally, to stop unconstrained strengthening and stabilize the weighs, we balance growth with decay. Adding decay terms to the mathematical formulation of the rules yields their final form Eqs 2 and 3. The simulation in Fig 3c suggests that the construction was successful: the final version of the rules allows the weights to converge to p and n respectively.
After providing an intuitive understanding of the learning rules and their mathematical formulation, we proceed to a more rigorous analytical treatment. We saw the potential of Mikhael and Bogacz’ [16] rules to learn payoffs and costs. Appropriate choice of parameters is key to unlock that potential, and we shall now investigate how that choice must be made. In particular, we will derive certain relations between parameters that must be satisfied for payoff and cost to be learned.
Originally, the rules Eqs 2 and 3 were meant to describe learning of reinforcements statistics. Mikhael and Bogacz [16] showed that after learning, particular combinations of G and N will encode the mean E R and the mean spread E | R − E R | of the received reinforcements. For further reference, we denote these important statistics by q ≔ E R and s ≔ E | R − q |. How are the mean and the mean spread of received reinforcements related to payoff and cost? Consider the reinforcement statistics of an action that reliably requires effort (corresponding to negative reinforcement) to produce a payoff (which corresponds to positive reinforcement). Assume that these reinforcements are clearly negative and positive respectively, such that one will not be confused for the other even in the presence of noise. Repeat that action multiple times, and record all received reinforcements, the costs as well as the payoffs. Finally, analyze how all these received reinforcements are distributed. If an effort was required to earn the payoff, the distribution of reinforcements will turn out bimodal, as schematically shown in Fig 4. It features two peaks, one centered around the mean payoff p, and one centered around the mean cost −n, respectively. Fig 4 also shows the mean q and the mean spread s of that distribution. We observe that payoffs and costs are both exactly one mean spread s away from the center q of the distribution—the payoff above, and the cost below. This implies that there is, at least in this representative case, a strong connection between payoffs and costs and the reinforcement statistics:
p= q + s (4)
− n= q − s (5)
This connection allows us to set up conditions for the result of learning: if G and N are to represent payoff and cost, they must approach q + s and −q + s respectively. Equivalently, we can ask for 1/2(G − N) and 1/2(G + N) to approach q and s in the course of learning.
After revealing the link between reinforcement statistics and payoff and cost, we are ready to derive the relations necessary to learn the latter. To that end, we first determine the connection strengths G and N that result from training on stochastic reinforcements. Such uncertain reinforcements are sampled at random from a fixed distribution. Then, we implement the newly identified conditions, demanding for 1/2(G − N) to approximate q and 1/2(G + N) to approximate s after training is finished. From these conditions, we will be able to derive the desired parameter relations.
Working through these steps is simpler after changing variables from G and N to Q ≔ 1/2(G − N) and S ≔ 1/2(G + N) right away. We saw that the new variables Q and S have a clear computational interpretation: if learning goes as planned, Q and S track the mean q and the mean spread s of the experienced reinforcement. To determine how Q and S change due to prediction errors δ, we simply add and subtract the update rules Eqs 2 and 3. Certain convenient properties of the nonlinear functions fϵ help to further simplify the resulting equations: Fig 3f shows that subtracting and adding functions depicted in Fig 3e give functions proportional to identity and absolute value, respectively. Explicitly, fϵ(x) − fϵ(−x) = (1 + ϵ)x and fϵ(x) + fϵ(−x) = (1 − ϵ) |x|. Exploiting these properties, we obtain
Δ Q = α Q δ − λ Q (6)
Δ S = α S | δ | − λ S . (7)
Here, for brevity of notation, we introduced the effective learning rates αQ = α(1 + ϵ)/2 and αS = α(1 − ϵ)/2. Note that the changes of Q and S are proportional either to the prediction error itself or to its absolute value, in contrast to the changes of G and N.
Now, let us determine the strengths of the weights G and N, or equivalently of the variables Q and S, after many encounters with an action. When learning the reinforcements of a previously unknown action, Q and S typically change a lot during the first trials. These changes then get smaller and smaller as more experience is integrated—the learning curve plateaus. After enough trials, Q and S stop changing systematically, and start to merely fluctuate about some constant values, which we denote by Q* and S* and refer to as equilibrium points. In mathematical terms, directed learning stops when we may expect Q and S to remain unchanged by another trial, i.e. when E ( Δ Q ) = E ( Δ S ) = 0. If that stage is reached, the equilibrium points can be inferred by computing the mean value of the fluctuating variables: Q * = E Q and S * = E S. With these identities and the learning rules Eqs 6 and 7, we can determine the equilibrium points Q* and S*:
0=E Δ Q = E [ α Q ( R − Q ) − λ Q ] = α Q ( q − Q * ) − λ Q * (8)
0=E Δ S = E [ α S | R − Q | − λ S ] = α S E | R − Q | − λ S * . (9)
To solve these equations, we shall make the additional assumption that the fluctuations of Q about Q* are small. This assumption is justified whenever α is sufficiently small, and allows us to approximate E | R − Q | ≈ E | R − Q * |. Collecting all those intermediate results, we may solve Eqs 8 and 9 for the equilibrium points. The solutions read
Q *= c Q q (10)
S *≈ c S E | R − c Q q | , (11)
with cQ = αQ/(αQ + λ) and cS = αS/λ. Those are the approximate values of Q and S after learning.
Next, we need to implement the conditions we inferred from Fig 4. Thanks to our choice of variables, this simply amounts to requiring that Q converge to the mean reinforcement q, and S to the mean spread s, i.e. requiring Q* = q and S* = s. Inserting the approximate values from Eqs 10 and 11 produced by the learning rules, we obtain
c Q q= q (12)
c S E | R − c Q q |= s (13)
These equations are central to this publication. Their left-hand side represents the result of learning according to Mikhael and Bogacz’ [16] rules. Their right-hand side specifies what needs to be learned if G and N really represented payoffs and costs, as Collins and Frank hypothesized [8]. Equating the left-hand and the right-hand side amounts to merging both theories. It allows us to determine how the parameters would be related if both theories were exactly true: for Eqs 12 and 13 to hold, α, λ and ϵ must take values such that cQ = 1 and cS = 1.
This result evokes several questions: Is it at all possible to satisfy the derived conditions? What do the conditions mean with respect to the parameters α, λ and ϵ? And finally, is there a practical way to determine sets of parameters α, λ and ϵ which—at least approximately—satisfy the conditions? We discuss each of these questions in the following paragraphs.
Firstly, is it possible to satisfy cQ = 1 and cS = 1 exactly? Examining the definition cQ = αQ/(αQ + λ) quickly reveals that letting cQ → 1 would amount to letting λ → 0. To see why this is the case, consider that cQ → 1 amounts to λ/αQ → 0. However, αQ is an effective learning rate, and so must take values smaller then one. Thus, we really need to let λ → 0. Now, we derived above that after learning, S will fluctuate about its equilibrium point S * ≈ c S E | R − c Q q | with cS = αS/λ. In order to keep the equilibrium point S* finite as λ → 0, we would therefore be forced to have αS → 0 also. This, though, would pose a real problem: αS is the effective learning rate for S—having it vanish would imply stopping learning in S all together. We must conclude that strict satisfaction of the constraints cQ = 1 and cS = 1 is not compatible with non-vanishing learning rates that lead to a finite equilibrium. Specifically, cQ = 1 can only ever hold approximately if the spread s is to be learned in finite time. Nevertheless, no such problem arises when cS is set to 1 exactly.
Now, what do the constraints cQ ≈ 1 and cS = 1 mean in terms of the parameters α, λ and ϵ? In the previous paragraph, we saw that cQ ≈ 1 is equivalent to λ/αQ ≈ 0. Since both λ (a decay constant) and αQ (an effective learning rate) are inherently positive, we may rewrite this as λ/αQ ≪ 1. Inserting the definition αQ = α(1 + ϵ)/2 immediately yields
2 λ ≪ α ( 1 + ϵ ) (14)
The other condition, cS = 1, is easily translated analogously. We need only use the definitions cS = αS/λ and αS = α(1 − ϵ)/2 to obtain
2 λ = α ( 1 − ϵ ) . (15)
Eqs 14 and 15 constitute the exact relations between the parameters α, λ and ϵ that need to hold for payoffs and costs to be estimated accurately. They cannot be further simplified, but we may use them to gain some more insight into the required magnitudes of the individual parameters: by substituting 2λ according to Eq 15 on the right-hand side of Eq 14, one obtains a condition of the form 1 − ϵ ≪ 1 + ϵ. Now, given that the intended range for ϵ is [0, 1], one quickly reaches the conclusion that ϵ ≈ 1. Reinserting this into Eq 14 yields λ ≪ α. In conclusion, we found that it is necessary (though not sufficient) for accurate learning of payoffs and costs to maintain a small, but non vanishing nonlinearity ϵ in the transmission of the prediction error signal, as well as a non vanishing decay rate λ, which is much smaller than the learning rate α.
Finally, how can such parameters α, λ and ϵ practically be determined? To implement the conditions cQ ≈ 1 and cS = 1, one can for instance express λ and ϵ in terms of α, cQ and cS. It is tedious, but without conceptual difficulty to invert the definitions of cQ and cS in order to yield ϵ = (1 − cS(1/cQ − 1))/(1 + cS(1/cQ − 1)) and λ = α(1 − ϵ)/(2cS). Then, one chooses α freely at one’s convenience, and cQ and cS close (or, in case of cS, equal) to one. Importantly, cQ must be chosen smaller then one to result in a positive λ. From these choices, one finally obtains ϵ and λ to work with the chosen α. Our simulations suggest that even values such as cQ = 0.7 and cS = 0.9, in combination with a learning rate of, say α = 0.3, are close enough to one to allow reasonably accurate estimations of payoff and cost. This can be seen in Fig 3: the simulations shown in there used those exact settings, which equivalently means that ϵ = 0.443 and λ = 0.093.
In summary, we used a statistical argument–the connection between payoffs and costs and the reinforcement statistics–to determine conditions under which payoffs and costs can be learned with the update rules Eqs 2 and 3.
In the preceding section, we derived relations that are necessary for successful learning of payoff and cost. If rewards are awarded stochastically, those relations are also sufficient for successful learning. But what happens to the weighs G and N if the received reinforcements follow a strong pattern? Assume, for instance, that an action reliably yields a fixed cost −n followed by a fixed payoff p. Under which additional conditions do G and N then still reflect the magnitudes of payoff and cost after learning?
To answer that question, we must again determine the connection strengths that result from experiencing the action time and again. Now, we do not have to rely on a probabilistic treatment—when the pattern of the reinforcements is fully known, it is possible to determine the evolution of G and N exactly. As in the previous section, we will concentrate on the result of learning rather than on its dynamics. Here, this amounts to determine the fixed points of the learning rules. These fixed points are simply those values of G and N (or equivalently of the alternative variables Q and S we defined above) that are invariant under the updates caused by the action. We denote the fixed points by G* and N*, or Q* and S*. During learning, the variables converge to their respective fixed points and cease to change notably once they arrive in their vicinity.
First, we focus on determining the fixed point of Q. Note that each encounter with the action yields two updates of Q: one due to the cost and one due to the payoff. Mathematically, we can formulate this as
Q after action = Q before action + ( Δ Q ) cost + ( Δ Q ) payoff . (16)
To find Q*, demand that these successive updates have no net effect on Q: If Qafter action equals Qbefore action, then Qbefore action can rightfully be called fixed point. If this is so, the two updates must have canceled each other:
( Δ Q ) cost + ( Δ Q ) payoff = 0 (17)
This condition, in combination with the update rules Eqs 2 and 3, allows to determine Q* in terms of p, n and the parameters α, ϵ and λ. First, we use the update rule Eq 6 for Q to write (ΔQ)cost as
( Δ Q ) cost= α Q ( r cost − Q before action ) − λ Q before action= α Q ( − n − Q before action ) − λ Q before action .
Then, one uses the rule again to write (ΔQ)payoff as
( Δ Q ) payoff= α Q ( r payoff − Q after cost ) − λ Q after cost= α Q ( p − Q after cost ) − λ Q after cost= α Q ( p − ( Q before action + ( Δ Q ) cost ) ) − λ ( Q before action + ( Δ Q ) cost ) .
Finally, one substitutes (ΔQ)cost from above into this expression, and then inserts (ΔQ)cost and (ΔQ)payoff into Eq 17. Solving the equation for Qbefore action, which in case of Eq 17 is identical to Q*, yields
Q * = 1 2 − α Q − λ ( n ( α Q + λ − 1 ) + p ) , (18)
where αQ = α(1 + ϵ)/2. Now, recall that the definition of Q in terms of G and N is Q = 1/2(G − N), and that true payoffs and costs of in this model are p and n. If G and N represented the true payoffs and costs after learning, it must be true that G* ≈ p and N* ≈ n, and thereby
1 2 − α Q − λ ( n ( α Q + λ − 1 ) + p ) ≈ 1 2 ( p − n ) . (19)
Just as Eqs 12 and 13, this equation is an interface between the results of Mikhael and Bogacz’ [16] update rules on the left-hand side and the requirement that Go and No-Go weights encode payoffs and costs on the right-hand side. For both sides to agree, we must have
α Q + λ ≈ 0 . (20)
This is a novel condition for learning the correct magnitudes of payoffs and costs from a deterministic reinforcement pattern. The definition of αQ and the previously derived conditions in Eqs 14 and 15 may be used to transform this novel condition into the simpler form α ≪ 1.
Next, we repeat the same analysis for S. Since we search for additional conditions on the parameters, we are free to use the original conditions in Eqs 14 and 15 to simplify our calculations. The only complication we encounter is the appearance of Q in the update rules of S, which we resolve by substituting Q with Q*, acknowledging that the fixed points of S and Q depend on each other. We arrive at
S * ≈ 1 2 ( p + n ) . (21)
Again, using the definition S = 1/2(G + N) allows comparing the result of learning with the strengths required to represent payoffs and costs. We immediately find that G* ≈ p and N* ≈ n already hold. Thus, Eq 20 is the only additional condition for successful learning of payoff and cost from reinforcements that follow a strong pattern.
From the results presented in this section, we conclude that the learning rules Eqs 2 and 3 facilitate learning of the magnitudes of fixed payoffs and costs that occur reliably one after the other. However, we also saw that this is only true if Eq 20 holds in addition to the conditions that we derived in the previous section.
The analysis above revealed the conditions under which the striatal plasticity rules Eqs 2 and 3 could learn the magnitudes of the payoffs and costs of actions. We identified the conditions in two different paradigms: first, we investigated learning from purely stochastic reinforcements sampled from a fixed distribution. Then, we considered a deterministic pattern of reinforcements. We obtained two key results:
Consider a reinforcement distribution—obtained from multiple encounters with an action—that is shaped by payoffs and cost, as the one shown in Fig 4. If trained on reinforcements sampled from that distribution, the plasticity rules Eqs 2 and 3 will enable learning of the mean payoffs and costs if
2 λ≪ α ( 1 + ϵ ) (22)
2 λ= α ( 1 − ϵ ) (23)
hold. These conditions imply–but do not follow from–a non-vanishing but small nonlinearity in the transmission of the prediction error, and a non-vanishing but small decay of the connection weights. Here, a small decay is characterized by a decay rate λ which is small compared to the learning rate α.
If trained on a pattern of reinforcements that alternates between payoffs of magnitude p and costs of magnitude n, the plasticity rules Eqs 2 and 3 will capture those exact payoffs and costs if, in addition to Eqs 22 and 23,
α ≪ 1 (24)
holds. In words, unbiased learning of payoffs and costs in deterministic scenarios explicitly requires a small learning rate α.
The previous sections revealed what to expect from training the learning rules Eqs 2 and 3 on certain types of reinforcement. Specifically, we investigated the connection strengths G and N after many experiences of either totally predictable or totally random reinforcements. In this section, we aim to confirm and extend those results using numerical simulations rather than analytic methods.
Fig 5 shows the results of simulating the gradual change of connection weights in four different tasks. In all those simulations, G and N change according to the learning rules Eqs 2 and 3. The parameters we used roughly fulfill the conditions Eqs 12 and 13 for learning of the correct magnitudes of payoffs and costs, but are also chosen to facilitate quick convergence. The values presented in Fig 5a mirror that compromise.
The simulation in Fig 5a is based on a repeating an action that reliably results in a cost −n, followed by a payoff p. An analytic treatment of that case can be found in the previous sections. Both weights constantly oscillate due to the alternation of payoff and costs. This oscillating behavior is superimposed with learning curves that take the weights from their initial values towards the magnitudes of the payoffs and costs respectively. After 30 trials, G and N represent good approximations of p and n. Fig 5b is similar to Fig 5a, with a slight variation: Just as in Fig 5a, payoffs and costs alternate reliably. But while the cost is again held constant at −n, this time the payoff P is sampled from a fixed distribution (a normal distribution with mean p and non-vanishing variance) in each trial. Thus, the task includes both stochastic and deterministic components: each repetition of an action results in a fixed cost, which is followed by an uncertain reinforcement. The depicted simulations show that under such conditions, N eventually represents the cost n, while G converges towards the mean payoff p = E P.
Finally, Fig 5c and 5d contain simulations of repeated actions with reinforcements drawn completely at random from fixed distributions. In Fig 5c, the obtained reinforcements are valued either p or −n, with probabilities 1/2 each. In Fig 5d, reinforcements are sampled from a normal distribution with mean μr = 1/2(p − n) and standard deviation of σ r = 1 / 2 π / 2 ( p + n ). We simulate the experience resulting from such actions by sampling reinforcements from a fixed distribution on each trial. The stochastic nature of this procedure causes the evolution of the weights G and N to be different each time the simulation is run. To overcome that effect and segregate random fluctuations from reproducible effects, we collect and average a large number of runs. Each row in Fig 5b–5d contains both a single run of the simulation and an average of 500 successive runs. In the above sections, we proved that in purely stochastic tasks, the weights would approximate key statistics of the reinforcement distribution after convergence. Those statistics are indeed approximated in the simulations, confirming the results of the analytic treatment above.
In the previous sections, we focused on the change of the synaptic weights associated with a single action during the accumulation of experience. In this section, we redirect our attention. Instead of considering one action during learning, we now consider multiple actions after learning, and ask: can effects of dopamine depletion on choice behavior be explained in terms of payoffs versus costs?
In a classic experiment illustrated in Fig 6a, rats were given a choice between pressing a lever in order to obtain a nutritious pellet and freely available lab chow [22]. Normal animals were willing to work for pellets, but after blocking D2 receptors with the drug haloperidol they were not any more willing to make an effort and preferred a less valuable but free option. Collins and Frank [8] provided a mechanical explanation for this surprising effect. The theory proposed in this paper accounts for it in a conceptually similar but slightly simpler way. Here, we explain our modeling of the experiment and then describe the simulations—the differences to the account of OpAL model are pointed out in Discussion.
To model the experiment, we need to specify how the striatal weights G and N and the motivation signal transmitted by dopamine affect the output of the basal ganglia system, and how that output then affects choice. We refer to the output of the basal ganglia as the thalamic activity, denoted by T. T depends on the cortico-striatal weights G and N, and dopaminergic motivation signal denoted by D. Even though this relationship might admittedly be complex, we restrict ourselves to just capture the signs of the dependencies by using a linear approximation:
T = D G − ( 1 − D ) N (25)
In the above equation, the first term DG corresponds to the input from the striatal Go neurons. This term is positive because the projection from striatal Go neurons to the thalamus involves double inhibitory connections (see Fig 1) resulting in an overall excitatory effect. The activity of the Go neurons depends on synaptic weights G. We assume that their gain is modulated by the dopaminergic input D, extrapolated from the observation of an increased slope of the firing-input relationship in the presence of dopamine in pyramidal neurons expressing D1 receptors [23]. These data are replotted in Fig 7a. The second term −(1 − D)N corresponds to the input from the striatal No-Go neurons. It has a negative sign because the projection form the No-Go neurons to the thalamus includes three inhibitory connections. The activity of the striatal No-Go neurons depends on their synaptic weights N, and we assume that their gain is reduced by dopamine, so the synaptic input is scaled by (1 − D). This assumption is based on data showing that agonists reduce the slope of the firing-input relationship of striatal No-Go neurons [24], which are replotted in Fig 7b. Those assumptions about the impact of dopamine on the activity of striatal neurons are backed up by detailed modeling studies [25, 26], which predict precisely that dopamine enhances activity in the Go- and inhibits activity in the No-Go pathway. In Eq 25, we further assume that D ∈ [0, 1] and that the value of D = 0.5 corresponds to a baseline level of dopamine for which both striatal populations equally affect the thalamic activity.
Although arising from a slightly different induction, the action value defined by Eq 25 is directly proportional to the action value proposed by Collins and Frank, which is defined by Eq 4 of their publication [8]: Q ∝ βGG − βNN. One easily verifies the direct proportionality of the two expressions by rewriting
D = 1 / 2 ( 1 + ( β G − β N ) / ( β G + β N ) ) .
How does thalamic activity affect choice? Again, we use a very simple dependency to capture the key aspects of that relationship: In our model of the experiment, we calculate the thalamic activity for each option. Then, we add some random noise independently to each option. Finally, all options with negative noisy thalamic activity are discarded, and the option with the highest noisy thalamic activity is chosen. If the noisy thalamic activity is negative for all available options, no choice will be made; the model defaults to staying inactive.
Often in similar situations, the softmax rule is the preferred choice procedure. According to that rule, one should first transform the set of different action values (or thalamic activities in this case) into a probability distribution over the available actions, by use of the softmax function. Then, one should sample an action from that distribution, and declare it the choice of that trial. Collins and Frank’s OpAL model [8] exemplifies the use of the softmax rule.
We deliberately decided against this conventional approach and in favor of the above-described procedure to accommodate a certain feature of the data presented in [22]: The group with D2 antagonist differed from the control group not only in their willingness to work for food but also in their overall food consumption. The rats with D2 antagonist consumed less food in total (see Fig 8c). We can hope to capture this effect with our model, since it allows for the possibility to make no choice at all, and thus consume neither of the food items. A softmax decision rule, on the other hand, forces a choice on each trial, and must therefore always lead to the same number of consumed food items.
Finally, how does the drug haloperidol affect the thalamic activity, and hence choice? Haloperidol is a D2 antagonist; it blocks the D2 receptors on the medial spiny neurons of the No-Go pathway. This blocking reduces the (inhibiting) impact of dopamine on the activity N of that pathway. To account for this in our model, we introduce another factor κN ∈ [0, 1] into our expression for the thalamic activity:
T = D G − ( ( 1 − κ N D ) N . (26)
The parameter κN controls the how much dopamine affects the activity of the No-Go pathway N, and is hence suitable to model D2-blocking: κN = 1 recovers the normal thalamic activity given in Eq 25, while κN = 0 (total blocking) fully removes the impact of dopamine on the indirect pathway, leading to completely uninhibited activity N. In the control group of the experiment, κN is set to 1 (no medication is administered, no blocking happens). In the group that received the medication, κN is a free parameter that must be fitted to the data. The best fit featured κN = 0.7507, corresponding to blocking of D2 receptors with an efficiency of roughly 25%.
Fig 6b illustrates how the model can account for the behaviour when the dopamine level has a normal baseline value. In the figure, the strength of the cortico-striatal connections is denoted by the labels and the thickness of arrows. Pressing the lever gives a high payoff, so the weights of Go neurons selective for this action are strong, but it also has a substantial cost, so the No-Go weights are also present. On the other hand, the free food is not particularly nutritious so the Go weights are weak, and there is no cost, so the No-Go weight is negligible. When no medication is administered, the positive and negative consequences are weighted equally, so the thalamic neurons selective for pressing the lever have overall slightly higher activity, which ultimately leads to a higher likelihood for this action to be chosen over the free option. By contrast, Fig 6c shows that when the D2 receptors are blocked, costs are weighted more than payoffs, and the thalamic activity associated with pressing the lever decreases. Approaching free food has only negligible cost; therefore, the activity of thalamic neurons selective for this option is now higher, and this action is overall more likely to be chosen.
A quantitative fit of our model to Salamone et al.’s experimental results [22] is illustrated in Fig 8. The panels on the left side in Fig 8 summarize experimental data: the top-left display corresponds to a condition in which both high-valued pellets and the low-valued lab chow were freely available. In this case, the animals preferred pellets irrespective of the dopamine level. The bottom-left panel corresponds to the condition in which the animal had to press a lever in order to obtain a pellet, and as mentioned before, after injections of a D2 antagonist they started to prefer the lab chow.
In our model of the experiment, we run through a sequence of trials mimicking those illustrated in Fig 6: on each trial, the model makes a choice between two actions—pressing a lever or approaching lab chow—or remains inactive. Before the main experiments, the animals were trained to press a lever to obtain rewards and were exposed to the lab chow [22]. To parallel this in simulations, the model was first trained such that it experienced each action a number of times, received corresponding payoffs and costs, and updated its weights according to Eqs 2 and 3. The weights resulting from that learning are reported in Fig 6b and 6c. Then, the model was tested with and without blocking, e.g. with κN a variable and κN fixed to one. As described in Materials and Methods, the parameters of the model were optimized to match experimentally observed behavior. As shown in the right displays in Fig 8, the model was able to reproduce the observed pattern of behavior. This illustrates the model’s ability to capture both learning about payoffs and costs associated with individual actions and the effects of the dopamine level on choices.
Above, we dedicated a whole section to derive conditions for the parameters of the learning rules Eqs 2 and 3 to guarantee correct (i.e. unbiased) estimation of payoffs and costs. We also pointed out that these conditions cannot be satisfied exactly even in theory; in fact, our own simulations throughout this paper yield parameter settings that significantly violate the conditions. The proposed biological implementation of the rules, certainly imperfect and subject to unpredictable influences, is yet less likely to feature parameters close to the constraint surface. How robust is the presented learning algorithm under parameter detuning? How much variation around the conditions can the rules take without breaking? Here, we first describe the effect of parameter detuning on the values to which Go and No-Go weights converge. Then, we argue that the algorithm will still produce useful results even under substantial detuning of the parameters.
We are interested in the coding of payoffs and costs after learning, and should therefore investigate the equilibrium values G* and N* of G and N. Those equilibrium values may be obtained via combination of the equilibrium values of Q* and S* given in Eqs 8 and 11:
G *= Q * + S * ≈ c Q q + c s E | R − c q q | ≈ c Q q + c s s (27)
− N *= Q * − S * ≈ c Q q − c s E | R − c q q | ≈ c Q q − c s s . (28)
Here, we assumed that the average spread around cQ q is approximately equal to the average spread around q, which is a good approximation if the spread of a distribution is comparable to the mean. Next, we can use the relation of payoffs p and costs n to the statistics q and s of the reinforcement distribution they generate. These relations are given in Eqs 4 and 5; inverting and inserting those yields
G *≈ 1 2 ( c Q + c S ) p − 1 2 ( c Q − c S ) n (29)
N *≈ − 1 2 ( c Q − c S ) p + 1 2 ( c Q + c S ) n . (30)
We observe that as long as cQ = cS, the Go and No-Go weights converge to the vicinity of values proportional to the payoffs and costs. Thus, as long as cQ = cS, the payoffs and costs are encoded separately in the two pathways.
Expressed in terms of the elementary parameters α, λ and ϵ, and solved for ϵ, this condition becomes
ϵ = ( 2 λ / α) 2 + 1 − 2 λ / α . (31)
A second solution of the condition exists; however, it yields ϵ < 0, which is biologically implausible. Hence, we ignore that second solution and focus our attention on Eq 31: if λ/α is very small (i.e. if decay is weak relative to learning), then ϵ approaches one, rendering the learning rules approximately linear. If, on the other hand, λ/α is very large (i.e. decay is very strong compared to learning), then ϵ approaches zero, rendering the learning rules maximally non-linear. This relationship between ϵ and λ is not surprising; in fact, we have seen in Fig 2 that decay is necessary to balance the unconstrained strengthening of the weighs that results from introducing the nonlinearity (compare Fig 2b and Fig 2c). Eq 31 makes this manifest: the stronger the nonlinearity (i.e. the closer ϵ gets to zero), the stronger the decay relative to learning–and vice versa.
Now, after investigating the effect of detuning on G* and N*, let us explore the effect of detuning on the thalamic activity T, which is the relevant output of our model as far as action selection is concerned. Substituting the above equations into the definition of thalamic activity in Eq 25 we obtain:
T = p ( 1 2 c Q − 1 2 c S + D c S ) − n ( 1 2 c Q + 1 2 c S − D c S ) (32)
When cQ = cS ≠ 1, the thalamic activity becomes scaled by a constant cS, but as this scaling constant is the same for all actions, the network can still select actions on the basis of payoffs and costs modulated by motivation signal D, in the same way as described in the previous subsection. Importantly, the effect of dopamine–to emphasize the payoff when increased, and emphasize the cost when decreased–is present as long as cS > 0 even if cQ ≠ cS. These signature effects of the proposed mechanism are thus robust even under significant detuning. However, the disadvantage of setting parameters such that cQ ≠ cS is that the dopaminergic motivation signal D would have a relatively smaller effect on changing the weighting between payoffs and costs; for example the payoffs or costs could no longer be ignored by setting D to its extreme values of 0 or 1. From this analysis, we may conclude that while action selection is quite robust under violation of the derived conditions, dopaminergic regulation works most effectively if the conditions are met approximately.
So far, we assumed that the outcome prediction is computed by the same striatal neurons that encode the payoffs and costs of actions. Only one network was involved: that which is responsible for the choice of action. We refer to such a network as ‘actor’ in the remainder of this exposition. In this section, we look at how the theory described above generalizes to the actor-critic framework [27]. That framework assumes that the outcome prediction is not computed by the actor, but by a separate group of striatal patch neurons called the ‘critic’. More formally, the purpose of that critic is to learn the value V of the current state.
One way to generalize our theory in this direction is to keep the actor network unaltered, while supplementing it with a similar critic network that learns by the very similar rules Eqs 2 and 3:
Δ G c r i t i c ( s )=α f ϵ ( δ ) − λ G c r i t i c ( s ) (33)
Δ N c r i t i c ( s )=α f ϵ ( − δ ) − λ N c r i t i c ( s ) (34)
The crucial difference between the actor and the critic is that the critic network is not selective for the action, but only for the situation (note that Gcritic(s) and Ncritic(s) depend on s, but not on a, as opposed to Gactor (s, a) and Nactor (s, a)). It thus learns the value of a situation irrespective of the actions chosen. Importantly, the critic is in charge of supplying the outcome predictions. Those predictions are compared to the actual outcomes to produce the outcome prediction errors δ from which both networks learn.
We take the state value to be encoded in the difference of Gcritic(s) and Ncritic(s): Vcritic(s) = 1/2(Gcritic(s) − Ncritic(s)). The change of the state value on each trial can be obtained by subtracting Eqs 33 and 34:
Δ V c r i t i c ( s ) = α ( 1 + ε ) 2 δ − λ V c r i t i c ( s ) (35)
The prediction error δ—which teaches the actor as well—is the difference between the obtained reinforcement r and the reinforcement prediction by the critic:
δ = r − V c r i t i c ( s ) (36)
What would be learned with that architecture? If the same action is selected on each trial, the actor will learn in exactly the same way as the critic. Then, the prediction error in the actor-critic model is the same as in the actor-only model described above, and the weights of the actor in the actor-critic model converge to exactly the same values as for the actor-only model. However, this reasoning does not seem to apply if more than one action is available: empirically, animals then select the actions that maximize their rewards in their own perception. In the process of learning, they will likely sample all available actions.
If such behavior generates input for an actor-critic model, the critic will integrate the experience of all those trials, and will thus represent a mixture of the expected reinforcements associated with the available actions. This generally interferes with correct learning of the payoffs and costs of the different actions. However, there is a caveat: one of the available actions will eventually prove most useful; as soon as the animal has determined that best action, it will select it in the majority of cases. That, in turn, forces the critic into mainly representing the expected reinforcement of this best action. As a final consequence, also payoff and cost of that best action are inferred correctly.
We confirmed this empirically for the model specified above: in Fig 9, we present simulations of a task in which the subject must choose between two actions. Both actions reliably yield a constant cost followed by a constant payoff each time they are selected. One of the actions is unambiguously superior to the other: its payoff is larger and its cost is lower.
Both an actor-only model and an actor-critic model interacted with that task. On each trial, an action was selected by sampling from a softmax distribution over all available actions: the probability of choosing action a in situation s was proportional to exp (βQ(s, a)), where Q(s, a) = 1/2(G(s, a) − N(s, a)) was the action value, and β was the softmax temperature. Fig 9 shows the temporal evolution of the involved synaptic weights over the course of learning. Fig 9a and 9b depict the actor-only evolution of the weights G and N that encode the payoffs and costs of actions 1 and 2, respectively. For both actions, payoffs and costs are learned correctly. Learning is notably slower for action 1. This is easily explained: action 1 is the worse of the two options and thus chosen much less frequent. In contrast, the actor-critic driven evolution of the same weights presented in Fig 9d and 9e leads to a correct estimate of the payoff and cost only for the superior action 1. Learning is impaired for the inferior action 2, as anticipated in the qualitative discussion above. The state value, presented in Fig 9c, provides further confidence in the validity of that discussion: Instead of encoding a mixture of the values of all available actions, it converges to the value of the superior action, indicated by the higher of the two dashed lines.
What have we learned in this section? We set out to analyze an actor-critic formulation of our model, where the feedback signal that teaches the actor is computed by a different network called the critic. We found that our formulation (which is by no means the only possible one) enables the actor to learn accurate estimates of the payoffs and costs of the most advantageous action from the critic’s feedback. The payoffs and costs of the other actions were not estimated as accurately, which was due to a sampling bias towards more rewarding options. This does not necessarily compromise behavior–after all, one may trust the model to provide accurate information on the actions that are most frequently picked, and thus to be helpful in the majority of cases. However, we believe that a more sophisticated actor-critic variant of our model could conceivably provide good estimates of the payoffs and costs of all actions. The development of this improved actor-critic variant is left to future work; here we merely demonstrate that our model is not meant to compete with actor-critic models, but rather to complement them.
This article describes how the positive and negative consequences of actions can be separately learned on the basis of a single teaching signal encoding outcome prediction error. In this section, we relate the theory with data and other models, state experimental predictions, and highlight the directions in which the theory needs to be developed further.
The model described in this paper was shown in simulations to avoid actions that require effort when the motivational signal was reduced. The unwillingness to make an effort for reward in dopamine-depleted state has also been observed in other paradigms: During a choice in a T-maze, dopamine-depleted animals were less likely to go to an arm with more pellets behind the barrier, but rather chose the arm with easily accessible but fewer pellets [28]. Parkinson’s patients were not willing to exert as much physical effort by squeezing a handle in order to obtain reward as healthy controls, especially if they were off medications [29]. These effects can be explained in an analogous way [8] by assuming that in the dopamine-depleted state the effort of crossing the barrier or squeezing a handle is weighted more, resulting in lower activity of thalamic neurons selective for this option. Both in OpAL and the model proposed here, reducing the dopamine level reduces the tendency to choose actions involving costs, and thus changes preferences.
Let us now consider how the weight changes in our model relate to known data on synaptic plasticity in the striatum. Fig 10b illustrates the weight changes when an animal performs an action involving a cost n in order to obtain a payoff p (Fig 10a), e.g. pressing a lever in order to obtain a pellet. The direction of changes in G and N depending on the sign of δ are consistent with the changes of synaptic weights of Go and No-Go neurons observed at different dopamine concentrations. Fig 10c shows experimentally observed changes in synaptic strengths when the level of dopamine is low (displays with white background) and in the presence of agonists (blue background) [11]. Note that the directions of change match those in the corresponding displays above, in Fig 10b.
These directions of changes in striatal weights are also consistent with other models of the basal ganglia [8, 12], but the unique prediction of the rules described in this paper is that the increase in dopaminergic teaching signal should mainly affect changes in G, while the decrease in dopamine should primarily affect N. Thus, the dopamine receptors on the Go and No-Go neurons should be most sensitive to increases and decreases in dopamine level respectively. This matches the properties of these receptors. The D2 receptors on No-Go neurons have a higher affinity and therefore are sensitive to low levels of dopamine compared to D1 receptors on Go neurons [31]. This property is illustrated in Fig 10d where the green and red curves show the probabilities of D1 and D2 receptors being occupied as a function of dopamine concentration. The blue dashed lines indicate the levels of dopamine in the striatum predicted to result from the spontaneous firing of dopaminergic neurons [32]. At these levels, most D1 receptors are deactivated. Thus the D1 receptor activation will change when the dopamine goes up, but not when it goes down, as indicated by the black arrows. This is consistent with the stronger impact of positive prediction errors on the weight changes of the Go neurons implemented in Eq 2. By contrast, the D2 receptors are activated at baseline dopamine levels, so their activation is affected by the decreases in dopamine level but little by increases, in agreement with a stronger impact of positive prediction errors on the No-Go neurons implemented in Eq 3.
Our model further requires decay of relevant weights whenever prediction errors are absent. In terms of neural implementation, this translates into mild LTD resulting from co-activation of the pre- and post-synaptic cells at baseline dopamine levels. Recently, this effect has been observed at cortico-striatal synapses in vivo [33]: in anesthetized rats, presynaptic activity followed by postsynaptic activity caused LTD in the absence of induced dopaminergic response.
In summary, the plasticity rules allowing learning positive and negative consequences are consistent with the observed plasticity and the receptor properties.
Recently, there has been a debate concerning the fundamental concept of basal ganglia function, i.e. the relationship between the Go and No-Go neurons: on one hand they have the opposite effects on a tendency to make movements [2], but on the other hand they are co-activated during action selection [34]. The presented theory is consistent with both observations: It assumes that Go and No-Go neurons have opposite effects on movement initiation. But during action selection, the basal ganglia need to calculate the utility which combines information encoded by both populations, so may require their co-activation.
The proposed model assumes that while an animal makes an effort, the outcome prediction error should be negative, thus the dopamine level should decrease. However, at the time of lever pressing the system needs to be energized to perform a movement, so one could expect an increased level of dopamine. Furthermore, voltammetry studies measuring dopamine concentration in the striatum did not observe a decrease in dopamine level during lever pressing [35]. Nevertheless a recent study recording activity of single dopaminergic neurons that provided a better temporal resolution reported that dopaminergic neurons increased the activity before movement, and then decreased it below baseline during movement [32]. The increase before movement may be related with energizing system for movement, while the decrease during movement may be related with representing effort.
In addition to effort, other negative experiences lead to phasic decreases in dopaminergic activity as well: the unexpected experience of pain [36], aversive stimuli such as air puffs [37] and, for humans, monetary losses (literal costs) [38] all coincide with decreased activity of dopamine neurons. This supports the general idea that the No-Go pathway encode costs of all kinds.
A direct test of the proposed model could involve the recording of the activity of Go and No-Go neurons (e.g. with photometry) during a task in which an animal learns the payoffs and costs associated with an action. Assuming that G and N are reflected in the activity of the Go and No-Go neurons while the animal evaluates an action (i.e. just before its selection), one could analyze the changes in the activity of Go and No-Go neurons across trials. One could compare if they follow the pattern predicted by the rules given in this paper, or rather by other rules proposed to describe learning in striatal neurons [7, 8, 14].
Just as the OpAL model [8], the theory proposes that the positive and negative consequences are separately encoded by the Go and No-Go neurons which are differentially modulated by dopamine. The theory predicts that agonists specific to just one of the striatal populations change the effect of consequences encoded by this population without changing the impact of the other population. For instance, a D1 antagonist would suppress the reception of dopamine in the direct pathway. There, dopamine increases activity. Hence, the D1 antagonist would diminish the impact of the direct pathway, and therefore of learned positive consequences, on choices. However, it would not change the impact of the indirect pathway, i.e. the impact of learned negative consequences. This prediction could be tested in an experiment involving choices between options with both payoff and cost. Consider, for instance, the decision between a neutral option (p = 1, n = 1) and a high-payoff option (p = 2, n = 1). Since a D1 antagonist decreases the impact of payoffs on decisions, it should decrease the preference for the high-payoff option. On the other hand, the avoidance of a high-cost option (p = 1, n = 2) over the neutral option should not be affected by the D1 antagonist, since it does not affect the impact of costs on decisions.
It could also be worthwhile to investigate whether changing the influence of positive and negative consequences on choice can not only be achieved by pharmacological manipulations, but also by changing a behavioral context such as hunger, or reward rate which has been shown to affect the average dopamine level [19].
The theory assumes that the synaptic plasticity rules include a decay term proportional to the value of the synaptic weights themselves. Decay terms are also present in other models of learning in basal ganglia [15, 39, 40]. This class of models predicts that the synaptic weights of striatal neurons which are already high increase less during potentiation than the smaller weights (an opposite prediction is made by the OpAL model [8], where the weights scale the prediction error in the update rule). This prediction could be tested by observing the Excitatory Post-Synaptic Currents (EPSCs) evoked at individual spines. The class of model including decay predicts that the spines with smaller evoked EPSCs before inducing plasticity should be more likely to potentiate.
The proposed model builds on the seminal work of Collins and Frank [8], who proposed that the Go and No-Go neurons learn the tendency to execute and inhibit movements, and how the level of dopamine changes the influence of the Go and No-Go pathways on choice. The key new feature of the present model is the ability to learn both payoffs and costs associated with a single action. We demonstrated above that when the model repeatedly selects an action resulting first in a cost and then in the payoff, G and N—under certain conditions that we specified—converge to the magnitudes of that payoff and cost. This is not so in the original OpAL model, as we shall show in a brief analysis.
Collins and Frank [8] demonstrated that when the environment is stationary and prediction error δ converges to zero, then the weights G and N in the OpAL model converge to bounded values. However, we will show that Go and No-Go weights converge to zero when an action that results first in a cost and then in the payoff is repeatedly selected.
The OpAL model is based on the actor-critic framework; hence, the prediction error is defined as in Eq 36. The weights of the critic are modified simply as ΔV = αδ. The weights of the actor are modified according to the following equations [8]:
Δ G=α G δ (37)
Δ N=− α N δ (38)
Fig 11 shows how the weights change in a simulation of the OpAL model. The weights of the critic approach a value close to the average of payoff and cost. Let us consider what happens in the model once the critic weight stops changing between trials (i.e. from ∼10th trial onward in Fig 11). The weight of the critic still changes within a trial, i.e. decreases when cost is incurred and increases after a payoff. This happens because the prediction error oscillates around 0, i.e. it is equal to δ = −d while incurring a cost and δ = d while receiving a payoff, where d is a constant. If so, let us consider how a Go weight changes within a trial. According to Eq 37 the weight changes as follows:
G after cost=G before action − α G before action d (39)
G after payoff=G after cost + α G after cost d (40)
Substituting Eq 39 into Eq 40 we obtain:
G after payoff=G before action − α G before action d + α ( G before action − α G before action d ) d =G before action − α 2 G before action d 2 (41)
We see that within a trial a Go weight decays proportionally to is value, resulting in an exponential decay across trials seen in Fig 11. Analogous calculations show that the No-Go weight decays in the same way. We conclude that the OpAL model is unable to estimate positive and negative consequences for actions which result in both payoffs and costs. It is worth noting that the decay of actor weights to zero demonstrated above is specific to the version of basal ganglia model proposed by Collins and Frank [8], but would not be present in another version of the model [39] where the learning rules include a special term preventing the weights from approaching zero. On the other hand, nothing in the above calculation depended on G, N and V updating at the same learning rate α–the derivation can be carried out in exactly the same way assuming αV ≠ αN ≠ αG. Hence, we may summarise that even such generalised OpAL models must fail to learn payoffs and costs of actions, irrespective of the specific parameter values unless further terms are added to the learning rules. Our analysis suggests that learning payoffs and costs can be enabled by different effective learning rates after positive versus negative feedback for Go and No-Go synapses, which in our model is achieved by setting ϵ < 1.
To interpret this result, note that we do not claim that the OpAL model is not capable of optimizing the policy. It is set up as a policy improving algorithm, and might even reflect the payoffs and costs of actions in the weights G and N in certain situations. However, as we have shown there is also situations in which OpAL is not able to encode the payoffs and costs. In contrast, we showed above the model presented in this paper does encode payoffs and costs in any situation, given a suitable set of parameters and enough time to learn.
The model described in this paper has been shown to account for the effects of dopamine depletion on the willingness to make effort, which have also been simulated with the OpAL model. To simulate the effects of dopamine depletion on the choice between an arm of a T-maze with more pellets behind a barrier and an arm with fewer pellets, [8] trained a model on three separate actions: eating in the left arm, eating in the right arm, and crossing a barrier. In this way, it was ensured that each action had just payoff or just cost, and the model could learn them. Subsequently, during choice, the model was deciding between a combination of two actions (e.g. crossing a barrier and eating in the left arm) and the other action. By contrast, the model proposed in this paper was choosing just between the two options available to an animal in an analogous task (Fig 6), because it was able to learn both payoffs and costs associated with each option. This is a useful ability, as most real-world actions have both payoffs and costs.
In the original paper introducing the plasticity rules [16], it was proposed that the rules allow the Go and No-Go neurons to encode reinforcement variability because when an action results in variable reinforcements, both G and N increase during learning. It was further proposed that the tonic level of dopamine controls the tendency to make risky choices, as observed in experiments [41], because it leads to emphasizing potential gains, and under-weighting potential losses. However, here it is proposed that the striatal learning rules primarily sub-serve a function more fundamental for survival, i.e. learning payoffs and costs of actions. From this perspective, the influence of dopamine level on the tendency to make risky choices arises as a by-product of a system primarily optimized to weight payoffs and costs according to the current motivational state.
There are multiple directions in which the presented theory could be extended. For example, the theory has to be integrated with the models of action selection in the basal ganglia to describe how the circuit selects the action with the best trade-off of payoffs and costs. Furthermore, the theory may be extended to describe the dependence of the dopaminergic teaching signal on the motivational state. Learning experiments in which an animal may be deprived of physiologically required substances suggest that both terms in the outcome prediction error encoded by dopamine (i.e. the reinforcement and the expected outcome) are scaled by motivation [42]. It would be interesting to incorporate such scaling in our model, where the direct pathway, as well as the indirect pathway, contribute to the outcome estimate, which is then compared to the experienced reinforcement to compute the prediction error. If dopaminergic modulation is taken into account also at this stage, the dopaminergic motivation signal should affect the outcome estimate, and hence influence learning.
A limitation to our current model is the rudimentary form of the basal ganglia output, given in Eq 25. It is known that the effect of dopamine on the activity in the two pathways is not linear (as assumed in this paper), but exhibits saturation effects. The fact that the reception of dopamine is nonlinear plays a crucial role in the learning part of our model (the piecewise linear functions fϵ introduce exactly that nonlinearity), and could also be implemented at the decision-making stage, if the activity of Go and No-Go neurons (combined in Eq 25) depended nonlinearly on the dopamine level. In such more elaborate formulation, the fine-tuning of the baseline dopamine level then becomes critical. Including nonlinear effects of dopamine on activity during choice would allow studying interactions between learning and decision making, which would both be affected by the position of the baseline and the strength of the nonlinearity.
It is intriguing to ask whether the evaluation of actions combining separately encoded positive and negative consequences is also performed by areas beyond the basal ganglia. Indeed, positive and negative associations are encoded by different populations of neurons in the amygdala [43]. Moreover, an imaging study [44] suggests that costs and payoffs are predicted by the amygdala and the ventral striatum respectively, and ultimately compared in the prefrontal cortex. Furthermore, different cortical regions preferentially project to Go or No-Go neurons [45], raising the possibility that the positive and negative consequences are also encoded separately in the cortex. Therefore, it seems promising to investigate if similar plasticity rules could also describe learning beyond the basal ganglia.
During simulations of an experiment by Salamone et al. [22], the model received payoff pchow = 1 for approaching the lab chow, and payoff ppellet for choosing a pellet. The model was simulated in two conditions differing in the cost of choosing a pellet which was equal to npellet = 0 in the free-pellet condition, and to npellet = nlever in a condition requiring lever pressing to obtain a pellet. There was no cost of choosing lab chow (nchow = 0) in either condition.
For each condition, the model was simulated in two operational modes: in the control state, the coupling κN of dopamine to the D2-expressing neurons was fixed at κN = 1 during choice (making manifest the assumed fully functioning dopaminergic modulation in the control group). Conversely, in the state corresponding to the presence of the D2-antagonist haloperidol, κN was treated as a variable valued in [0, 1], now allowing for impaired dopaminergic regulation. The level of dopamine D was kept fixed at D = 0.5 throughout, assuming largely an unaltered baseline level for both groups.
For each condition and state, the behavior of Nrats was simulated. Each simulation consisted of 180 training and 180 testing trials (as each animal in the experiment of [22] was tested for 30 minutes, so 180 trials correspond to an assumption that a single trial took 10s). At the start of each simulation, the weights were initialized to Gpellet = Npellet = Gchow = Nchow = 0.1. During each training trial, the model experienced choosing a pellet as well as approaching the lab chow. In detail, it received the cost npellet, modified the weights Gpellet and Npellet, then received the payoff ppellet and modified the weight again, and analogously for the lab chow. During each testing trial, the thalamic activity for each option was calculated from Eq 25), and Gaussian noise with standard deviation σ was added. An option with the highest thalamic activity was selected, and if this activity was positive, the action was executed, resulting in the corresponding cost and payoff and weight modification. If thalamic activity for both options was negative, no action was executed and no weights were updated.
The values of model parameters: ppellet, nlever, κN, σ were optimized to match the choices made by the animals. In particular, for each set of parameters, the model was simulated Nrats = 100 times, and the average number of choices c i , j , k s i m of option i in dopamine state j and experimental condition k was computed. The mismatch with corresponding consumption in experiment c i , j , k e x p was quantified by a normalized summed squared error:
C o s t = ∑ k = 1 2 ∑ j = 1 2 ∑ i = 1 2 ( c i , j , k s i m Z k s i m − c i , j , k e x p Z k e x p) 2 (42)
In the above equation Z k d a t a s e t is a normalization term equal to the total number of choices or consumption in a particular condition:
Z k d a t a s e t = ∑ j = 1 2 ∑ i = 1 2 c i , j , k d a t a s e t (43)
The values of parameters minimizing the cost function were sought using the Simplex optimization algorithm implemented in Matlab, and the following values were found: ppellet = 15.511751, nlever = 14.510517, κN = 0.7507 and σ = 1.066246. Subsequently, the model with these optimized parameters was simulated with Nrats = 6, which was the number of animals tested by [22]. The resulting mean number of choices across animals are shown in Fig 8.
|
10.1371/journal.pntd.0000933 | Activities of Rifampin, Rifapentine and Clarithromycin Alone and in Combination against Mycobacterium ulcerans Disease in Mice | Treatment of Mycobacterium ulcerans disease, or Buruli ulcer (BU), has shifted from surgery to treatment with streptomycin(STR)+rifampin(RIF) since 2004 based on studies in a mouse model and clinical trials. We tested two entirely oral regimens for BU treatment, rifampin(RIF)+clarithromycin(CLR) and rifapentine(RPT)+clarithromycin(CLR) in the mouse model.
BALB/c mice were infected in the right hind footpad with M. ulcerans strain 1059 and treated daily (5 days/week) for 4 weeks, beginning 11 days after infection. Treatment groups included an untreated control, STR+RIF as a positive control, and test regimens of RIF, RPT, STR and CLR given alone and the RIF+CLR and RPT+CLR combinations. The relative efficacy of the drug treatments was compared on the basis of footpad CFU counts and median time to footpad swelling. Except for CLR, which was bacteriostatic, treatment with all other drugs reduced CFU counts by approximately 2 or 3 log10. Median time to footpad swelling after infection was 5.5, 16, 17, 23.5 and 36.5 weeks in mice receiving no treatment, CLR alone, RIF+CLR, RIF alone, and STR alone, respectively. At the end of follow-up, 39 weeks after infection, only 48%, 26.4% and 16.3% of mice treated with RPT+CLR, RPT alone and STR+RIF had developed swollen footpads. An in vitro checkerboard assay showed the interaction of CLR and RIF to be indifferent. However, in mice, co-administration with CLR resulted in a roughly 25% decrease in the maximal serum concentration (Cmax) and area under the serum concentration-time curve (AUC) of each rifamycin. Delaying the administration of CLR by one hour restored Cmax and AUC values of RIF to levels obtained with RIF alone.
These results suggest that an entirely oral daily regimen of RPT+CLR may be at least as effective as the currently recommended combination of injected STR+oral RIF.
| Buruli ulcer (BU) is found throughout the world but is particularly prevalent in West Africa. Until 2004, treatment for this disfiguring disease was surgical excision followed by skin grafting, procedures often requiring months of hospitalization. More recently, an 8-week regimen of oral rifampin and streptomycin administered by injection has become the standard of care recommended by the World Health Organization. However, daily injections require sterile needles and syringes to prevent spread of blood borne pathogens and streptomycin has potentially serious side effects, most notably hearing loss. We tested an entirely oral regimen, substituting the long acting rifapentine for rifampin and clarithromycin for streptomycin. We also evaluated each drug separately. We found that rifapentine alone is as good as rifampin plus streptomycin, but the simultaneous addition of effective clarithromycin doses, at least in the mouse, reduces the activity of both rifampin and rifapentine, making it difficult to assess the efficacy of the oral regimens in the model. Studies of serum drug concentrations indicated that separating treatment times by one hour or reducing the clarithromycin dose to one active in humans should overcome this issue in experimental and clinical BU treatment, respectively.
| Mycobacterium ulcerans disease, also known as Buruli ulcer (BU), is the third most prevalent disease caused by mycobacteria [1]. It is characterized by deep and necrotizing skin ulcers with undermined edges resulting from the secretion by M. ulcerans of an immunosuppressive macrolide toxin, termed mycolactone [2]. It is predominantly found in scattered foci in tropical riverine and marshy regions throughout the world. In certain parts of Africa its prevalence may exceed 150/100,000 individuals [3].
Until 2004, the recommended treatment for BU was surgical excision and skin grafting [1]. However, experimental studies using the mouse footpad model demonstrated that the combination of rifampin (RIF) and an aminoglycoside was bactericidal for M. ulcerans [4], [5], [6], [7]. Based on these findings and subsequent studies in humans [8], [9] the daily administration of the streptomycin-rifampin (STR+RIF) combination for 2 months was recommended by the World Health Organization (WHO) for the treatment of BU [10]. Depending on the size, severity and location of the ulcer, additional surgical intervention with skin grafting was also recommended. Treatment with STR requires intramuscular injection, which is difficult and expensive to implement in resource-poor countries since it requires use of sterile needles and syringes to avoid infection with blood borne pathogens. Therefore, the development of an entirely oral regimen is desirable [11].
In vitro, M. ulcerans is susceptible to a limited number of oral antibiotics including fluoroquinolones and macrolides [12], [13],[14],[15],[16]. However, in the mouse model, the combination of clarithromycin (CLR), a bacteriostatic or weakly bactericidal drug against M. ulcerans, and RIF, which also has limited bactericidal activity [4], [6] has not consistently shown efficacy similar to the standard STR+RIF regimen. Three mouse studies assessing the bactericidal activity and the relapse rate after treatment completion yielded conflicting results. In the first study [5], the oral combination was less effective than the standard aminoglycoside plus RIF combination whereas in the second and third [12], [13] studies, both combinations appeared to be as effective as the STR+RIF controls. There is, therefore, a need to directly address the issue of oral antibiotic treatment of BU, both with RIF and CLR and with new treatment regimens including anti-BU drugs that may have improved activity. Rifapentine (RPT), a rifamycin derivative with a much longer half-life than RIF could be an ideal substitute. In the murine model of tuberculosis when substituted for RIF at 10 mg/kg in a daily regimen in combination with isoniazid and pyrazinamide, it shortened the duration of treatment necessary to achieve cure [17], [18]. Such regimens are currently under evaluation in at least 3 Phase II trials for tuberculosis treatment. In a murine model of M. ulcerans disease, daily RPT at the lower dose of 5 mg/kg has also been shown to be as active as, or even more active than, daily RIF at 10 mg/kg [13].
In this study, we hypothesized that the use of daily RPT along with CLR would increase the efficacy of the rifamycin-CLR combination and help in the development of an entirely oral regimen for treatment of BU. We first demonstrated that there were no negative in vitro interactions of CLR and RIF (as a representative rifamycin) and then compared the efficacy of the RIF+CLR regimen to that of the RPT+CLR regimen using the STR+RIF standard regimen as control to determine whether daily RPT is a better substitute for daily RIF in the treatment of M. ulcerans disease in this murine model and whether daily RPT+CLR is also a better substitute for the standard daily STR+RIF combination.
STR and RIF were purchased from Sigma (St. Louis, MO) and RPT was a gift from sanofi-aventis pharmaceuticals (Paris, France). CLR was a gift from Abbott Laboratories (Abbott Park, U.S.A.). Stock solutions of RIF, RPT and CLR were prepared in sterile 0.05% agarose solution and STR was prepared in sterile normal saline. All stock solutions were prepared weekly and were stored at 4°C. All antimicrobials were administered orally (by gavage) using an esophageal cannula, except STR which was given by subcutaneous injections.
A recent isolate of M. ulcerans from a Ghanaian patient, strain Mu1059 [19] provided by Dr. Pamela Small, was used for the study.
To determine whether the RIF and CLR interaction is synergistic, indifferent or antagonistic, serial two-fold concentrations ranging from 0.125 to 2 µg/ml of both drugs alone and in combination were prepared in 7H11 agar+Oleic Acid-Albumin-Dextrose-Catalase (OADC) supplement. Eight-week-old colonies of Mu1059 from 7H11 agar plates were suspended in phosphate-buffered saline (PBS), briefly vortexed, and kept undisturbed for 30 minutes to allow larger particles to settle. The optical density at 600 nm of this suspension was adjusted to 1, and 500 µl of the appropriate dilutions were plated in duplicate on antibiotic-containing plates and control plates without antibiotic. Plates were incubated at 32°C, and final CFU counts were performed after 12 weeks. The MIC was defined as the lowest drug concentration to inhibit growth of at least 99% of CFU on drug-free control plates. The fractional inhibitory concentration (FIC) value of individual drugs was then calculated using the MIC of the drug alone and MIC of the drug in combination. The sums of the two FIC values were combined to give the ΣFIC value which was then used to determine whether synergism (ΣFIC≤0.5), indifference ΣFIC (>0.5 to ≤4) or antagonism (ΣFIC>4) occurred between the antibacterial agents. All calculations were performed in accordance with current accepted standards [20], [21], [22].
For each infection, an aliquot of a twice-mouse-passaged Mu1059 strain stored at −80°C was thawed and inoculated in mouse footpads. Once the footpads were swollen to a lesion index of 2–3 (defined as inflammatory footpad/hind foot swelling) [5], mice were sacrificed and footpad tissue was harvested, minced and suspended in sterile PBS. The solution was vortexed briefly, allowed to stand for 30 minutes, and the supernatant was used for footpad infection. Prior to infection, the inoculum was checked qualitatively for acid-fast bacilli, serially diluted, and plated for CFU counts on Middlebrook selective 7H11 plates (Becton-Dickinson, Sparks, MD).
The kinetic method developed by Shepard for assessing the activity of anti-leprosy drugs [4], [5], [6], [23], was used to assess drug activity. In brief, 320 female BALB/c mice aged 4-to-6 weeks (Charles River, Wilmington, MA) were infected in the right hind footpad with 0.03 ml of the M. ulcerans suspension. After infection, mice were randomized to one of two control groups or one of six test groups. The control groups included untreated negative controls (n = 50), and mice treated with STR+RIF as positive controls (n = 55). The test groups included mice treated with each antibiotic alone, i.e., CLR (n = 30), STR (n = 25), RIF (n = 25), and RPT (n = 25), and the two-drug combinations RIF+CLR (n = 55) and RPT+CLR (n = 55). Ten mice from the untreated group were sacrificed the day after infection (D1) and 11 days later at treatment initiation (D11) to establish baseline CFU counts in the footpads. All mice were treated for 4 weeks, 5 days per week. The drugs were given at the following doses that are equivalent (similar AUC) to the human doses [25], [26]: RIF 10 mg/kg, RPT 10 mg/kg, STR 150 mg/kg and CLR 100 mg/kg. On treatment completion, 5 mice from each group were sacrificed for quantitative CFU counts in the footpads and all of the remaining mice were kept without treatment to determine the time to footpad swelling.
For quantitative footpad CFU counts, each footpad was harvested after having been thoroughly disinfected with soap and sterile PBS followed by 70% alcohol swabs. The footpad tissue was homogenized by fine mincing and suspended in 2 ml sterile PBS. Appropriate dilutions were plated on selective 7H11 plates and incubated at 32°C for 12 weeks before CFU were enumerated. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal procedures were approved by the Johns Hopkins Animal Care and Use Committee (protocol MO08M240) and conducted according to relevant national and international guidelines.
The activity of each treatment was assessed in terms of CFU counts on treatment completion and median time to footpad swelling in treated mice compared with untreated control mice. CFU counts were performed by harvesting and homogenizing the footpad as described above and suspending each footpad in 2 ml PBS. Serial 10-fold dilutions were prepared and 0.5 ml of appropriate dilutions were plated in duplicate on 7H11 selective plates. The plates were then incubated at 32°C for 12 weeks before the CFU counts were made. Median time to footpad swelling was assessed by checking the footpads of mice every week for 39 weeks after infection. If the median time to footpad swelling in treated mice exceeded that in untreated mice by no more than the duration of the treatment, i.e. 4 weeks, then the treatment was considered to be bacteriostatic. Longer median time to swelling was indicative of bactericidal activity or prolonged post-antibiotic effect. Absence of swelling at the end of the follow-up period was indicative of sterilizing potential.
Because we observed a negative antimicrobial interaction between both rifamycin derivatives and CLR in vivo, a series of single-dose pharmacokinetic (PK) studies were performed in BALB/c mice. In the first study mice were co-administered 100 mg/kg of CLR and RIF 10 mg/kg. One sample for PK analysis was collected per mouse at 1, 2, 4, 6, 9 or 16 hrs after dosing. From these data, composite concentration-time curves were developed and compared. Because RIF serum concentrations appeared to be diminished when the two drugs were dosed together, we conducted a second study in which mice received 10 mg/kg of RIF alone, 10 mg/kg of RIF followed by 100 mg/kg of CLR one hour later, or 10 mg/kg of RIF co-administered with a lower dose of CLR (10 mg/kg). In a third study we substituted RPT for RIF and assessed RPT serum concentrations in mice receiving 10 mg/kg of RPT alone, 10 mg/kg of RPT co-administered with 100 mg/kg of CLR, or 10 mg/kg of RPT followed 1 hour later by 100 mg/kg of CLR. In a fourth study we evaluated RPT serum concentrations after RPT 10 mg/kg was co-administered with CLR at 10 mg/kg. Serum samples were frozen at −80°C and shipped overnight on dry ice to the Infectious Disease Pharmacokinetics Laboratory, National Jewish Medical and Research Center, Denver, CO. Drug concentrations were determined using validated HPLC methods. PK parameters were calculated using non compartmental methods with Phoenix WinNonlin software, version 6.1.0 (Pharsight, Cary, NC).
Survival analysis, with footpad swelling as the measurement, was performed using the Kaplan-Meier method [27]. The log rank test was used to determine the level of statistical significance when comparing survival curves of the different treatment groups with the control group. p values were two-tailed, and a value of p<0.05 was considered statistically significant. CFU counts were log-transformed before analysis. Culture-negative footpads were assigned a log value of 0. Group means for experimental treatment groups were compared with that of the standard treatment control by one-way analysis of variance with Dunnett's post-test. Paired t-tests were also used to compare groups of equal size. All analyses were performed with GraphPad Prism version 4.01 (GraphPad, San Diego, CA).
The results of the checkerboard study are shown in table 1. They indicated that the interaction (ΣFIC = 0.75) between the two drugs was neither synergistic nor antagonistic and therefore was termed indifferent using the current guidelines [20], [21], [22].
The initial footpad suspension used for the inoculum contained 5.76 log10 CFU per ml or 4.24 log10 in the 0.03 ml that was inoculated per footpad. The next day (D1) 10 mice were sacrificed and the mean CFU count per footpad was 3.29±0.41 log10. On initiation of treatment, 11 days after infection (D11), the mean CFU count in the 10 mice that were sacrificed was 3.35±0.16 log10 CFU, indicating that there was no substantial multiplication in the footpads during the first 11 days.
On treatment completion (Figure 1), 4 weeks later, the mean log10 CFU count was 5.01±0.62 in untreated control mice (W4 UT), demonstrating that M.ulcerans had multiplied well, increasing by about 2 log10 in the footpads during the 4 weeks following treatment initiation, and suggesting a division time close to 4 days. In the positive control mice treated with STR+RIF, the mean log10 CFU count was 0.76±0.52, with one footpad out of the 5 harvested footpads culture-negative, underscoring the potent bactericidal activity of the STR+RIF combination against actively multiplying M. ulcerans. Among the test mice, the mean log10 CFU count was 3.54±0.18 in mice treated with CLR alone, a value similar to the 3.35±0.16 log10 value on treatment initiation, confirming the bacteriostatic activity of CLR against actively multiplying M. ulcerans. For other antibiotics alone or in combination, the mean (including footpads with negative culture) log10 CFU counts were significantly reduced (p<0.01) compared to the baseline value (Figure 1) but were not significantly different from each other except that mice treated with STR+RIF had a lower mean CFU count compared to RIF alone by paired t-test analysis (p = 0.0335, though not significant after adjustment for multiple comparisons): 0.82±0.58 for STR alone (no CFU was isolated from 1 of the 5 mice); 1.33±0.24 for RIF alone and 1.37±1.15 for RIF+CLR (no CFU was isolated from 1 of the 5 mice); 0.48±0.56 for RPT alone (no CFU was isolated from 2 of the 5 mice) and 0.20±0.31 for RPT+CLR (no CFU was isolated from 3 of the 5 mice).
After completing 4 weeks of treatment, mice were monitored on a weekly basis for footpad swelling. Time to median swelling in untreated mice was 5 weeks after infection (Figure 2). In accordance with the CFU counts on treatment completion, mice treated with CLR were the first to reach footpad swelling. But the median time to swelling was 16 weeks after infection, well beyond the 9 weeks that would have been expected after 4 weeks of treatment with a purely bacteriostatic drug added to the 5 weeks time to swelling in untreated control mice. CLR treatment is thus accompanied by a prolonged delay in footpad swelling possibly due to a significant post-antibiotic effect. Mice treated by CLR were followed by mice treated by RIF alone and STR alone, with median time to footpad swelling of 23.5 and 34 weeks, respectively. Only 26.3% of mice treated with RPT alone and 11.4% of the positive controls treated with STR+RIF developed footpad swelling at the end of the 8-month follow-up period after treatment completion, emphasizing the potent sterilizing effect of both regimens. The difference between RPT alone and STR+RIF was not statistically significant (p = 0.33).
Surprisingly, as illustrated in Figure 2B, the time to footpad swelling was much shorter in mice treated with RIF+CLR (p<0.008) or RPT+CLR (p = 0.116) than in mice treated with RIF alone or RPT alone, respectively, suggesting antimicrobial or pharmacological antagonism between rifamycins and CLR in the mouse. Despite this antagonism, however, the RPT+CLR regimen caused a significantly greater (p = 0.0007) delay in footpad swelling than did RIF+CLR. Because the checkerboard study did not reveal antagonism between the two antimicrobials, the antagonism was likely to be due to pharmacokinetic drug-drug interaction as demonstrated below.
In the first PK study, as in the second and third PK studies, the experiments were performed as single and first-dose assessments, RIF (10 mg/kg) was either administered alone or co-administered with CLR (100 mg/kg). The mean AUC0–16h and Cmax of RIF were 118.63±18 µg*hr/ml and 11.97±1.3 µg/ml, respectively, when RIF was administered alone, and 92.5±27 µg*hr/ml and 8.48±0.54 µg/ml, respectively, when RIF was co-administered with CLR (Figure 3A), suggesting that CLR co-administration led to diminished RIF concentrations. In a second PK study, RIF (10 mg/kg) was administered alone, with CLR (100 mg/kg) given 1 hr later, or co-administered together with CLR at a lower dose of 10 mg/kg. The mean AUC0–21h and Cmax of RIF were 123±21 µg*hr/ml and 15.7±4.2 µg/ml, respectively, when RIF was administered alone, 133±32 µg*hr/ml and 16.7±2.5 µg/ml, respectively, when CLR (100 mg/kg) was administered 1 hr after RIF, and 125±22 µg*hr/ml and 15.5±1.2 µg/ml, respectively when RIF and CLR were co-administered at an equal dose of 10 mg/kg (Figure 3B).
Similar observations were made in mice given RPT and CLR. The AUC0–24h and Cmax of RPT were moderately decreased from 317.24±25 µg*hr/ml and 18.11±1.0 µg/ml, respectively, when RPT was given alone to 241.09±0.37 µg*hr/ml and 13.07±1.6µg/ml, respectively, when RPT was co-administered with CLR (100 mg/kg). Delaying administration of CLR (100mg/kg) by 1 hr resulted in a RPT AUC0–24hof 279.70±0.47 µg*hr/ml and Cmax of 15.14±1.7 µg/ml (figure 4A). Reducing the dose of CLR to 10 mg/kg when co-administered with 10 mg/kg RPT resulted in RPT AUC0–24hand Cmax values of 302.51±28 µg*hr/ml and 17.52±1.1 µg/ml, respectively, similar to those of 324.23±44 µg*hr/ml and 19.24±1.6 µg/ml, respectively, obtained with RPT alone (Figure 4B).
The main result of the present work is that RPT alone administered 5 days a week at a dose of 10mg/kg is at least as active in terms of bactericidal effect and as active in terms of relapse prevention as the standard combination of STR+RIF against experimental M.ulcerans disease in the mouse. Such a result is extremely promising for the future of M.ulcerans disease treatment because it suggests that an entirely oral treatment may be as active as the present regimen containing parenteral STR. However, RPT cannot be administered alone because of the risk of drug resistance resulting from monotherapy, and CLR is the oral companion drug of choice to combine with a rifamycin [28]. As CLR alone exhibited clear-cut bacteriostatic activity, an additive effect of the combination RPT+CLR and even RIF+CLR was expected. Unfortunately the co-administration of a rifamycin and CLR, both drugs given orally at doses equivalent to human doses on the basis of serum AUC, was less effective than each rifamycin alone in mice infected with M. ulcerans. As no antagonistic effect between RIF and CLR was exhibited in vitro in the checkerboard assay, the lesser in vivo effectiveness of the combination could not be related to a negative antimicrobial drug-drug interaction. Rather, it appears that the negative drug-drug interaction was pharmacokinetic in nature. Indeed, co-administration of a 10-mg/kg dose of RIF with a 100 mg/kg dose of CLR resulted in a 22% reduction of RIF AUC and a 29% reduction of RIF Cmax compared to administration of RIF alone. Similarly, RPT 10 mg/kg given together with 100 mg/kg CLR resulted in a 24% and 28% reduction of the RPT AUC and Cmax, respectively, compared to RPT administered alone. When administration of 100mg/kg of CLR was delayed by one hour from RIF administration, the pharmacokinetic interaction became insignificant. Similarly, there were also no negative pharmacokinetic drug-drug interactions when mice were co-administered 10 mg/kg of RIF and 10 mg/kg of CLR. These results indicate that the co-administration of CLR and RIF negatively interacts with the blood levels of rifamycins in mice probably by interfering with their absorption or by another mechanism and that this drug interaction is dose-dependent. However, a recent study in humans showed that concomitant CLR did not impact the absorption rate constant, the Cmax, or the Tmax of RIF at steady state, indicating that, at clinically relevant doses (7.5 mg/kg of CLR and 10 mg/kg of RIF), CLR does not negatively affect the levels of rifamycins in humans [29].
Our findings illustrate the difficulties in designing experiments in the murine model that aim to instruct treatment of a human infectious disease and in interpreting their results.
In order to adequately assess in mice the antimicrobial potential of a given drug, that drug should be given at doses deemed equivalent to human doses. As, drugs are usually metabolized much more rapidly in mice than in humans, the drug doses in mice have to be increased to obtain similar drug exposure in mice as in humans [25], [30]. That is the case for CLR [30], [31]. But the dose of 100 mg/kg that is adequate in mice to assess the antimicrobial activity of CLR when the drug is used alone presents a problem when it is co-administered with 10 mg/kg of RIF, most likely by reducing the absorption of RIF. Therefore the fact that combinations of CLR and a rifamycin were less active than the corresponding rifamycin alone should be considered an experimental artifact, and both drugs should be administered separately, with an interval of no less than one hour between them. Interestingly, the same phenomenon is observed in the experimental chemotherapy of tuberculosis for which RIF should be administered at least one hour before isoniazid and pyrazinamide to prevent a negative pharmacokinetic interaction in mice [32], [33].
Although the negative pharmacokinetic interactions prevented a reliable assessment of the antimicrobial activity of the RIF+CLR and RPT+CLR combinations against experimental M.ulcerans disease in mice, they did not prevent assessment of each drug alone in reference to the positive controls receiving STR+RIF. Besides the promising potency of RPT, our study also emphasizes the peculiar prolonged delay in footpad swelling resulting from treatment with CLR in mice infected with M. ulcerans. Whatever its antimicrobial or immunomodulatory [34] nature, this delay in footpad swelling is favorable and supports the use of CLR in the treatment of Buruli ulcer.
Finally, it is important to note that, in our experimental model, drug activity was assessed during a 4-week period during which untreated animals had a 2 log10 increase in CFU counts in their footpads. As expected, CLR exhibited bacteriostatic activity whereas other drugs exhibited bactericidal activity, especially RPT. But, very interestingly, even the most active drugs and drug regimens did not reduce the CFU counts by more than 3 log10 in 4 weeks. This was much less than the 5–6 log10 reduction in the CFU counts observed by Ji et al. [12], [13], [14] when treatment was initiated at the plateau phase of growth, i.e., when the organisms were no longer actively multiplying likely because of immune containment. In our experimental model because antibiotic treatment was initiated during the incubation phase of the disease, the reduction in the CFU counts and the time to foot pad swelling are measuring only the antimicrobial activity. When treatment is initiated at the plateau phase of growth, its effect is likely a mixture of antimicrobial activity, immune containment, and shutting down the enzymes involved in mycolactone production. It does not facilitate the assessment of the respective antimicrobial value of each drug regimen, even though it might better recapitulate the response of patients to antibiotic therapy. In the chemotherapy of BU, as in the chemotherapy of tuberculosis, the drug activity against actively multiplying organisms, usually termed bactericidal activity, is very different from the drug activity against organisms that are no longer actively multiplying, i.e., sterilizing activity. Consequently, the occurrence, magnitude, and duration of the antimicrobial effect depend on the experimental model used. The information provided by each model is therefore different yet complementary, and not at all contradictory.
|
10.1371/journal.ppat.1006908 | A protein coevolution method uncovers critical features of the Hepatitis C Virus fusion mechanism | Amino-acid coevolution can be referred to mutational compensatory patterns preserving the function of a protein. Viral envelope glycoproteins, which mediate entry of enveloped viruses into their host cells, are shaped by coevolution signals that confer to viruses the plasticity to evade neutralizing antibodies without altering viral entry mechanisms. The functions and structures of the two envelope glycoproteins of the Hepatitis C Virus (HCV), E1 and E2, are poorly described. Especially, how these two proteins mediate the HCV fusion process between the viral and the cell membrane remains elusive. Here, as a proof of concept, we aimed to take advantage of an original coevolution method recently developed to shed light on the HCV fusion mechanism. When first applied to the well-characterized Dengue Virus (DENV) envelope glycoproteins, coevolution analysis was able to predict important structural features and rearrangements of these viral protein complexes. When applied to HCV E1E2, computational coevolution analysis predicted that E1 and E2 refold interdependently during fusion through rearrangements of the E2 Back Layer (BL). Consistently, a soluble BL-derived polypeptide inhibited HCV infection of hepatoma cell lines, primary human hepatocytes and humanized liver mice. We showed that this polypeptide specifically inhibited HCV fusogenic rearrangements, hence supporting the critical role of this domain during HCV fusion. By combining coevolution analysis and in vitro assays, we also uncovered functionally-significant coevolving signals between E1 and E2 BL/Stem regions that govern HCV fusion, demonstrating the accuracy of our coevolution predictions. Altogether, our work shed light on important structural features of the HCV fusion mechanism and contributes to advance our functional understanding of this process. This study also provides an important proof of concept that coevolution can be employed to explore viral protein mediated-processes, and can guide the development of innovative translational strategies against challenging human-tropic viruses.
| Several virus-mediated molecular processes remain poorly described, which dampen the development of potent anti-viral therapies. Hence, new experimental strategies need to be undertaken to improve and accelerate our understanding of these processes. Here, as a proof of concept, we employ amino-acid coevolution as a tool to gain insights into the structural rearrangements of Hepatitis C Virus (HCV) envelope glycoproteins E1 and E2 during virus fusion with the cell membrane, and provide a basis for the inhibition of this process. Our coevolution analysis predicted that a specific domain of E2, the Back Layer (BL) is involved into significant conformational changes with E1 during the fusion of the HCV membrane with the cellular membrane. Consistently, a recombinant, soluble form of the BL was able to inhibit E1E2 fusogenic rearrangements and HCV infection. Moreover, predicted coevolution networks involving E1 and BL residues, as well as E1 and BL-adjacent residues, were found to modulate virus fusion. Our data shows that coevolution analysis is a powerful and underused approach that can provide significant insights into the functions and structural rearrangements of viral proteins. Importantly, this approach can also provide structural and molecular basis for the design of effective anti-viral drugs, and opens new perspectives to rapidly identify effective antiviral strategies against emerging and re-emerging viral pathogens.
| Flaviviridae such as Hepatitis C Virus (HCV), Dengue Virus (DENV), Zika Virus (ZIKV) or West Nile Virus (WNV) are cause of several acute and chronic diseases worldwide. The continuous investigation of the molecular processes by which these RNA viruses infect and replicate into their host is critical to develop innovative anti-viral strategies and anticipate viral resistance to pre-existing drugs.
The high potency of viral genomes to mutate is often considered as a major limitation for the development of effective anti-viral strategies. Nevertheless, the high-mutation rate of RNA viruses represents a unique opportunity to decrypt viral protein functions and structures. Highly-evolving viral genomes are shaped by important evolutionary constraints to maintain genetic structure and proper protein folding. Amino-acid coevolution, which refers to mutations of different residues at a similar time frame, mirrors such constraints. Hence, the identification and characterization of coevolution signals imprinted within viral protein sequences can provide unique insights into viral protein functions and conformational changes, and ultimately guide the design of original anti-viral strategies.
Virus entry is a conserved, critical step during the viral life cycle and represents a valuable target for the development of antivirals and vaccines. The entry process of HCV is orchestrated by two envelope glycoproteins, E1 and E2, which are incorporated onto the virion surface. During entry, E1E2 mediate viral particle attachment to cell surface receptors and induce the merging (called fusion) of endosomal and virus membranes at acidic pH, thus leading to release of viral RNA into the cytosol [1]. Flaviviruses such as DENV, ZIKV or WNV harbor two envelope glycoproteins: E and PrM. E is a class-II fusion protein composed of three distinct domains (domain I, II and III; or DI, DII and DIII respectively) and carries both binding and membrane fusion properties [2]. Crystal structure of E at different pH allowed to draw a fusion model during which initial E dimers change conformation and fold back as trimer structures to induce membrane merging [3]. In contrast, how HCV E1 and E2 mediate membrane fusion remains poorly defined. Our understanding of the HCV fusion process is strongly dampened by the absence of a well-defined pre- and post-fusion full-length E1E2 crystal structure. Few studies have attempted to computationally model pre-fusion E1E2 complex [4,5] but their impact remain limited as they have to rely on partial structural and functional information that are often collected in a non-heterodimer context.
The structure of a large region of the E2 ectodomain (E2 core) [6,7] exhibits a globular, non-extended fold divided into two distinct sheets: a front sheet composed of a front layer and a central Ig-fold domain, and a back sheet (or back layer, BL). Although the central Ig-fold domain represents a common structure among class-II fusion proteins, the BL harbors an original structure, thus undermining the possibility that HCV E2 is a classical fusion protein. It has been suggested through the resolution of the bovine viral diarrhea pestivirus (BVDV-1) E2 glycoprotein structure that HCV E1 may represent the HCV fusion protein [8–10]. Several studies have identified a hydrophobic region in E1 (CSALYVGDLC) that could represent the putative HCV fusion peptide [11–16]. Another study also suggested that E1 proteins form trimeric structure at the virus surface [17]. However, the recent crystal structure of the N-terminal domain of HCV E1 ectodomain does not harbor the expected truncated class-II fusion protein fold [18], suggesting that HCV fusion might be a unique process.
Mutagenesis studies have shown that both E1 and E2 domains, as well as E1-E2 dialogs, are involved in the HCV fusion process [13,14,19–21]. Thus, rather than being mediated by a single glycoprotein, HCV fusion appears to be mediated by complex intra- and inter-molecular E1-E2 dialogs that shape structural and conformational rearrangements of the heterodimer complex. Consequently, the characterization of interplays between E1 and E2 is critical to decipher the HCV fusion mechanism.
Here, using HCV fusion as a model of study, we aimed to provide a proof of concept that amino-acid coevolution and protein evolutionary constraints can shed light on viral protein functions and rearrangements. We hypothesized that detection of E1-E2 coevolution patterns can uncover their functional interplays as well as critical features of the HCV fusion mechanism.
We recently developed an original computational method, Blocks In Sequences (BIS), that can robustly detect coevolution signals within conserved cellular and viral proteins using a limited number of protein sequences [22–24]. Taking advantage of this methodology, we aim at establishing a map of E1-E2 coevolution patterns and test whether coevolution analysis can be employed to gain mechanistic insight into poorly characterized viral processes such as HCV fusion. BIS was able to accurately predict features of DENV glycoproteins structural organization onto viral particle, as well as E fusogenic conformational changes. When applied to HCV E1E2, BIS suggested that HCV E2 BL is a critical modulator of HCV fusion. Consistently, a soluble form of the E2 BL was able to inhibit HCV fusion. Moreover, coevolution signals between E1 and E2 BL/Stem predicted by BIS were found to regulate virus fusion in vitro. Beyond providing novel insights into the HCV fusion mechanism, our work also demonstrates that coevolution analysis can shed light on viral-mediated processes and can open avenues for the accelerated design of innovative anti-viral compounds against challenging human tropic-viruses.
We have previously reported that BIS, a combinatorial-based coevolution analysis method (S1 Fig), can accurately detect coevolution signals within a wide range of well-characterized cellular and viral proteins [22–24]. As BIS has not been previously tested for its ability to predict viral envelope glycoproteins structural organization and rearrangements, we first performed a coevolution analysis of the well-characterized DENV envelope glycoproteins, E and PrM [2,3]. During virus maturation, M protein (a mainly transmembranous protein) is associated to Pr, a peptide that protect the E fusion peptide and is cleaved prior viral budding [25]. Briefly, the PrM-E complex protrude as trimer at the surface of immature viral particles in the endoplasmic reticulum, a neutral pH compartment. Immature particle then navigates toward the trans-Golgi network, a more acidic compartment, where PrM-E complex form dimeric structures that lie down onto the surface of the particles. Pr, which initially concealed E fusion peptide, is then cleaved by the Furin. This cleavage achieves the maturation of viral particles that are then released into the extracellular compartment. As M is mostly a transmembrane protein with only very partial structural information available, we aimed to determine whether BIS can recapitulate the diversity of E-Pr structural organization at the surface of immature viral particles. BIS analysis of 17 DENV PrM-E serotype 2 sequences led to the identification of 14 groups of coevolving residues (possibly organized in blocks of consecutive amino-acids), further referred as clusters (S1 Table). Among those, three clusters (cluster 2, 7 and 9) displayed a strong statistical significance (with associated p values of 8e-5, 8e-5 and 7e-3 respectively) and involved coevolving blocks between E and Pr. When E is assembled as dimer at low pH condition, cluster 2 coevolving block positions supported the close proximity of E DIII and Pr (Fig 1A and 1B-left). Similarly, cluster 7 also recapitulated the close proximity between Pr and E domain DII on a trimeric E-Pr structure that form at neutral pH in the endoplasmic reticulum (Fig 1A and 1B-center). On a trimeric Pr-E structure (one dimer + one monomer of PrM-E) found at low pH when particles mature in the trans-Golgi, cluster 9 also supported a close proximity between Pr and E DII, but also between E dimers (Fig 1A and 1B-right). Two other clusters (cluster 3 and 8) of strong statistical significance (with respective p values of 1.34e-5 and 4.11e-5) were identified by BIS but did not involve coevolving residues located within E protein (S1 Table). Although cluster 8 was composed of two coevolving blocks located within M protein only, clusters 3 involved coevolving residues located within both M and Pr. As cluster 3, cluster 2 and 9 also supported the existence of interactions between Pr and M (Fig 1A and 1B). As M and Pr are the two cleavage products of a single PrM protein, these three clusters hence represent additional evidence of BIS ability to recapitulate biologically significant protein interactions. Taken together, our results demonstrate that BIS has the ability to accurately predict the tridimensional assembly of two viral proteins within different conformational states.
As DENV E has been demonstrated to mediate DENV viral fusion, we then aimed to study the intra-protein coevolution signals within DENV E only (Fig 2A; see new E numbering by BIS in comparison to Fig 1A), and determined whether coevolution signals can also predict E fusogenic rearrangements. Using 17 different DENV E serotype 2 sequences, BIS identified 12 clusters (S2 Table). Among them, nine clusters (clusters 2–8 and 11,12) displayed associated p-values ranging between 7e-3 to 2e-7 and two clusters (clusters 9 and 10) exhibited associated p-values of 0.058 (S2 Table). Cluster 1 and 2 were either conserved (p-value = 1) or too large respectively to be considered. Several clusters (3,4,6 and 7) displayed small blocks located within a single region of E both in the linear protein sequence and on the tridimensional structure, suggesting that coevolution signals might contribute to the structural organization of secondary protein sub-domains (such as internal loops) (S2 Fig). DENV E DI and DII are composed of two or three sub-domains that are distant on the linear protein sequence but form single structured domains in the protein tertiary structure. Cluster 8 blocks were mostly located within the two sub-domains of DII and were consistent with the tridimensional organization of this protein domain (Fig 2B). Cluster 8 blocks located within the second sub-domain of DII (DII-2) of each E monomer were in close contact on the dimeric E structure (especially at the level of the DII α-helix), hence suggesting that coevolution signals can predict point of contacts between E monomers once organized as dimer (Fig 2B). Despite lower statistical significance (p<0.06), cluster 9 and 10 coevolving blocks also supported E structural organization as these blocks were distant on the linear structure but close on the E dimer structure (S2 Fig). Finally, three clusters (5, 11 and 12) displayed coevolving blocks that were both distant on the linear and tridimensional E structure. During fusion, E DIII folds-over toward DII, and DII becomes at close proximity with the E transmembrane domain [3] (Fig 2C). Cluster 5 and 11 coevolving blocks organization were consistent with these structural rearrangements (Fig 2C), suggesting that BIS can recapitulate viral glycoprotein fusogenic conformational changes.
Following validation of BIS ability to model viral envelope glycoprotein structural rearrangements, we then applied the BIS methodology to HCV E1E2. We analyzed independently using BIS ten groups of E1E2 sequences from different genotypes (gt) or sub-types (1a, 1b, 1 = 1a+1b, 2a, 2b, 2 = 2a+2b, 3, 4a, 5a and 6a) (S3 Table). Interestingly, most of the identified clusters involved residues in both E1 and in E2 suggesting the existence of conserved tight dialogs between E1 and E2 proteins (Fig 3A). Only a few number of statistically significant clusters were found among genotype 4a to 6a sequences, due to the low number of sequences available and to their very low genetic divergence.
Unlike Dengue E and PrM, the full panel of HCV E2 conformations still remain unknown and E2core only represent a single of these possible conformations, in a given biochemical context. When BIS coevolution analysis is applied to protein(s) for which only a fraction of its/their conformations have been characterized (which is the case for HCV E1 and E2), BIS coevolution clusters can thus only suggest, but not contradict a given conformational hypothesis, this unless the full panel of a protein conformations is known. Consequently, in vitro and/or in vivo experiments are then critical to ascertain the functional significance of a given conformational hypothesis.
Given this particular experimental context and in order to identify putative E1E2 rearrangements during HCV fusion, we thus adopted the following approach. First, we aimed to assign to each E1E2 cluster a given function by mapping clusters blocks with residues previously identified in the literature to impact E1E2 folding/heterodimerization, binding or fusion. Second, we sought to identify among BIS clusters classified as “fusion clusters” a putative protein rearrangement supported by several fusion clusters across multiple genotypes, prior experimental challenge through in vitro experiments.
We first focused our efforts on analyzing gt1a clusters. Detailed analysis of these clusters (S4 Table; S3 Fig; note that BIS numbered E1 and E2 residues by considering the first amino acid of E1 as residue #1) showed E1 as coevolving systematically with all the E2 domains. Plotting the gt1a coevolving blocks onto reference sequences (S4 Fig) revealed a strong correlation between blocks and residues previously identified in the literature to be important for heterodimer folding or viral binding site conformation (grouped both under the term “structural”) or fusion. This correlation allowed us to propose functions (structural, fusion, or multifunctional clusters) for most gt1a clusters (S5 and S6 Tables). When plotted on the E2core structure (Fig 3B), most fusion clusters involved blocks located within the E2 BL (Fig 3C and 3D) in contrast to structural or multifunctional clusters for which blocks were broadly distributed across E2 (S5 Fig). Interestingly, some of the fusion clusters involved distant blocks in both E1 and E2 (clusters 5,7,10; Fig 3C and 3D), highlighting that E1 terminal regions and the E2 BL could be in close proximity during fusogenic rearrangements. In addition, BIS also suggested an association between fusogenic rearrangements and a potential packing of E2 domains (clusters 8,10; Fig 3C and 3D). Thus, BIS proposed that interdependent rearrangements of E1 and BL could represent a hallmark of E1E2 fusogenic conformational changes.
Analysis of clusters from another HCV genotype (gt2) provided similar findings as fusion clusters also involved the BL (cluster 10,13) as well as distant blocks on E1 (cluster 6,10) and E2 (cluster 10,16) (Fig 3E; S6–S8 Figs; S7–S9 Tables). Statistically significant coevolution networks between E1 and BL were also found among genotype 3 sequences (S9 Fig). Similar networks were also found among sequences of genotype 4 to 6, but displayed poor statistical significance for the reason described above. Genotype 3 to 6a cluster positions are available through the webpages indicated in the data availability statement.
HCV E1 and E2 transmembrane were previously shown to be critical for E1E2 heterodimerization and correct E1E2 functions [26–28]. Consistently, BIS was also able to identify several coevolution clusters between the transmembrane of E1 and E2 using sequences of genotype 1 and 2 (S10 Table), hence strengthening the functional significance of BIS analysis.
In parallel, BIS also revealed several gt1a and gt2 structural and multifunctional clusters as supportive of the E2core central scaffold structure (S5 and S8 Figs), reinforcing BIS as a relevant method to model viral protein conformations. The detailed analysis of all the clusters of gt1a and 2, regardless of their attributed function, can be found in S1 and S2 Texts respectively.
Altogether, BIS coevolution analysis of E1E2 sequences suggested that E2 may adopt a pre-fusion structure distinct from E2core as well as yet unreported molecular rearrangements that could occur during fusion. We hence hypothesized that a movement of the BL (green; Fig 4A), through dialogs with E1, could mediate the evolution from a potential stretched E2 pre-fusion structure toward a domain-packed E2 post-fusion structure (Fig 4B).
To challenge the potential role of the BL in E1E2 rearrangements, we generated a soluble 9kDa 6His-tagged BL domain (71 aa; named BLd-H77; Fig 4C; S10A Fig) from the H77 gt1a strain, detectable through Coomassie blue staining and Western immunoblotting (Fig 4D and 4E). Non-reducing SDS-Page electrophoresis and Dynamic Light Scattering (DLS) analysis (S10B and S10C Fig) confirmed the homogeneity of the peptide in solution and suggested that BLd-H77 fold as a monomer. The far UV circular dichroism (CD) spectrum of BLd-H77 eluted from size exclusion chromatography displays the molar ellipticity per residue expected for a protein folded mainly in α-helix (S10D Fig). We next assessed its effect on HCV infection. Interestingly, BLd-H77 was able to inhibit, in a dose-dependent manner, infection of Huh7.5 cells by replicative hepatoma cell line-derived HCV particles (HCVcc) harboring envelope glycoproteins of gt1a (H77/JFH-1) but also gt2a (JC1) (Fig 4F; S11A Fig). Although BLd-H77 was able to slightly inhibit HCVcc infection when pre-incubated with cells prior infection, it showed a strong potency to inhibit infection when present during the first four hours of infection (Fig 4G) thus suggesting that BLd-H77 might likely act on early steps of the virus life cycle. Consistently, BLd-H77 was able to efficiently inhibit infection of Huh7.5 by non-replicative retroviral pseudoparticles harboring HCV E1E2 (HCVpp) from different genotypes (Fig 4H and 4I; S11B Fig). This inhibition was specific as BLd-H77 was not able to inhibit infection by pseudoparticles harboring VSVG envelope (VSVGpp). Time-course experiments using BLd-H77 (Fig 4I) as well as two additional entry inhibitors, Bafilomycin A1 that acts on cell endosome acidification [29]) and an anti-E2 neutralizing antibody (that binds to E1E2 complexes [30]), demonstrated that BLd-H77 is an entry inhibitor that likely acts on viral particles (S11C Fig) but not on cells. No effect of BLd-H77 on HCV cell entry receptors expression could be observed (S11D Fig). Moreover, BLd-H77 was also able to inhibit cell-to-cell transmission (S11E Fig) in addition to cell-free infection (Fig 4F and 4G). Altogether, these results highlighted that BLd-H77 likely inhibits a conserved mechanism during HCV entry without affecting cell susceptibility for infection. Importantly, BLd-H77 was able to inhibit HCVcc and primary human hepatocytes-derived HCVcc virus (or HCVpc) infection of primary human hepatocytes (PHH) and Huh7.5 respectively (Fig 4J and 4K; S11F Fig). Finally, we assessed the ability of BLd-H77 to inhibit infection in vivo. BLd-H77 showed a potency to inhibit HCVcc JC1 infection over time in humanized liver mice treated with 150 μg of BLd-H77 under a prophylactic protocol (Fig 4L). Despite their uneven infectivity [31], our results also suggested that BLd-H77 is able to inhibit patient-derived HCV particles infection in humanized mice p = 0.02 for all quantifiable values equal and above the detection limit) as well as in PHH (S11G and S11H Fig). BLd-H77 had no impact on human hepatocyte viability in mice as assessed by serum albumin concentration over the course of infection (S11I Fig). Altogether, our results indicate that BLd-H77 is able to inhibit entry of different types of HCV particles in vitro and in vivo and thus target a strongly conserved virus entry mechanism.
We then sought to elucidate how BLd-H77 blocks HCV entry. By pre-incubating viral particles with BLd-H77 and diluting the mix prior to infection to reach a BLd-H77 concentration below its efficient neutralizing activity (determined in Fig 4F and 4H), we showed that BLd-H77 could irreversibly neutralize HCV particles regardless of mix dilution, hence suggesting that BLd-H77 can bind native viral particles prior viral entry (Fig 5A). To assess the presence of an interaction between HCV particles and the BLd-H77, we constructed a transmembrane form of BLd-H77 (called BLd-tm) (Fig 5B). Following lentiviral transduction, BLd-tm expression was detectable at Huh7.5-BLd-tm surface (S12A Fig), and did not impact HCV receptor expression (S12B Fig). BLd-tm expression at Huh7.5 surface, but not expression of a similar construct encoding for an anchored HIV-1 fusion inhibitor (namely C46), inhibited HCVcc propagation both in a cell-free and cell-to-cell transmission manner (Fig 5C and 5D; S12C Fig). HCVpp entry, but not VSVpp entry, (Fig 5E) was inhibited following infection of Huh7.5-BLd-tm, hence highlighting that BLd-H77 specifically inhibits HCV entry likely through binding of E1E2 glycoproteins. Consistently, more HCVpp were detected at Huh7.5-BLd-tm cell surface 4h post infection in comparison to Huh7.5 (S12D Fig), suggesting a potential containment of HCV particles by BLd-tm at the cell surface. The ability of recombinant soluble E2 (sE2) to bind more efficiently Huh7.5-BLd-tm cells than Huh7.5 cells in a dose dependent manner (Fig 5F) further suggested that virus entry is inhibited through an interaction between E2 and BLd-tm. In order to explore more precisely a putative interaction between E2 and BLd-H77, we designed an ELISA assay to quantify the ability of sE2 to be captured by coated BLd-H77. Our result showed a significant ability of sE2 to bind coated-BLd-H77 and coated-anti-E2 antibody AR3B [32] but not a coated-mouse IgG isotype (Fig 5G; S12E Fig). Altogether, these results strong suggest that BLd-H77 inhibit HCV entry by binding to E2 glycoprotein.
Next, we explored which step of HCV entry is inhibited by BLd-H77. BLd-H77 had no significant effect on attachment of HCVcc (HCVcc JC1, Fig 5H), HCVpp-H77 particles or soluble E2 (S13A Fig) on Huh7.5 cells. Using a highly specific and previously established HCVpp binding assay [21], we confirmed that BLd-H77 does not abrogate HCVpp binding to either human CD81 or SR-BI when used at a highly neutralizing concentration (S13B–S13D Fig). Moreover, BLd-H77 neutralizing activity was not competing with the activity of a neutralizing anti-E2 antibody known to inhibit viral particle binding [30], and their use in combination showed a synergistic neutralization effect (S13E Fig). Consistently, BLd-H77 could bind viral particles following their binding at the cell surface, and was shown to have its more potent neutralizing activity during post-binding steps (Fig 5I). Using a cell-cell fusion assay, we showed that BLd-H77 could strongly inhibit cell-to-cell fusion in comparison to control envelope glycoproteins (Fig 5J; S14A Fig). Importantly, cell-to-cell fusion was only inhibited when cells were incubated with BLd-H77 before low pH exposure that activate membrane fusion (S14B Fig), underlining BLd-H77 ability to specifically binds E1E2 pre-fusion conformations. Finally, using a HCVpp fusion assay with liposomes, which are devoid of any receptors or cell factors, the BLd-H77 inhibited fusion in a dose-dependent manner (Fig 5K, S14C Fig). Altogether, these results suggest that BLd-H77 specifically blocks E1E2 fusogenic rearrangements and the formation of post-fusion structures through binding to E2 protein, in accordance with BIS predictions.
Beside highlighting potential E1E2 rearrangements during fusion, BIS can identify pairs of residues that need to mutate in concert to guarantee structural compensations and proper viral fitness. Indeed, we have previously demonstrated how HCV entry depends on strain-specific dialogs between particular E1 and E2 domains [21]. Thus, we aimed at addressing whether BIS is able to unveil specific dialogs between residues of E1 and the E2 BL that modulate HCV fusion. The BIS predictions identified the multifunctional gt2 cluster 5 (S7–S9 Tables) as an interesting candidate for supporting E1 and E2 dialogs, BL movements and transition from E1E2 pre-fusion to post-fusion states. This cluster, similar to gt1a fusion cluster 5, linked two central blocks in E1 (residues 104, 105, and 109) and one block in E2 BL (residues 427 to 436; orange; Fig 6A). In order to challenge its potential role during fusion, we used cluster 5 blocks to guide the rational design of E1E2 chimeric constructs. The E1 region containing two cluster 5 blocks (Fig 6B; Region 1) and E2 BL regions containing the other cluster 5 block (Fig 6B; Region 2) or not (Fig 6B; Region 3, non-coevolving cluster 5 block as a control of specificity) were swapped between two E1E2 heterodimers from different gt2 strains, one allowing efficient HCVpp entry (J6) and another one mediating sub-optimal HCVpp entry (2b1) (S15A and S15B Fig). All chimeras were similarly expressed and incorporated (S15C Fig). Although J6 chimera carrying 2b1 cluster 5 regions (J6-1/2) displayed a poor entry efficiency, similar to 2b1 parental glycoproteins, a 2b1 chimera carrying J6 cluster 5 regions (2b1-1/2) exhibited >10-fold improved entry ability (Fig 6C). In contrast, 2b1 chimera carrying J6 Region 1 and 3 (2b1-1/3; Fig 6C) were not optimal for entry, hence suggesting that the E1 cluster 5 blocks and the N-terminal half of the BL domain are involved in a dialog regulating virus entry. Production and titration of HCVcc particles harboring these different chimeric envelopes confirmed these results (Fig 6D, S15D Fig). Interestingly, 2b1-1/2 also displayed improved ability to mediate cell-to-cell fusion (Fig 6E) as well as higher fusion efficiency at neutral pH than at acidic pH (Fig 6F), suggesting that this chimera exhibited an E1E2 conformation already primed for fusion at neutral pH. In contrast, J6-1/2 chimera did not increase J6 fusion efficiency and abrogated E1E2 sensitivity to low pH at levels similar to those of 2b1 (Fig 6E and 6F). Altogether, consistently with BIS predictions, these results suggest that conserved interplays between the central region of E1 and the N-amino-terminus region of the BL likely govern E1E2 fusogenic conformational states. Our results also support the maintenance of such dialogs through coevolution as they appeared to be mediated by genotype-specific regions of E1 and the BL.
To extend our transfer between the bioinformatics identification of coevolving amino acid clusters to the functional linkage of these domains, we tested the ability of BIS to pinpoint specific amino acids located in other regions than BL and furthermore, in the context of another genotype than genotype 2 (studied above). For this purpose, we sought to challenge a gt1a fusion cluster identified by BIS (Fusion cluster 5, S4–S6 Tables), which involved four residues within E1 central region (position 78 and 113–115) and a domain of 10 amino acids (452–461) within the Stem region. We employed two poorly divergent genotype 1a E1E2 sequences, H77 and A40, that displayed two E1 (SI/GM; position 112 and 117) and one E2 (D/N; position 462) amino acid differences located at the borders of the gt1a fusion cluster 5 blocks (Fig 7A; S16A Fig). Unlike J6 and 2b1, the level of functionality of H77 and A40 were relatively close (2.4x104 and 1.6x104 GFP i.u. per ml respectively) despite being significantly different (Fig 7B), hence making it challenging to predict the influence of residue swaps on envelope functionality. H77 chimera harboring both swapped E1 and E2 A40 residues significantly impacted HCVpp infectivity, although swapped E1 or E2 residues alone did not impact H77 functionality (Fig 7B) despite similar E1E2 expression and incorporation (Fig 7C). Importantly, H77 chimeras harboring only the E1 or E2 A40 residues showed defect for cell-cell fusion compared to H77, although fusion ability of the H77 chimera harboring both the E1 and E2 mutations were enhanced (Fig 7D). Altogether, consistently with BIS predictions, our results suggest that these E1 and E2 residues −and to larger extend the E1 central region and the E2 stem− are part of a coevolving network that regulates the fusogenic properties of gt1a viral envelope.
Flaviviridae are cause of many health concerns worldwide. A better understanding of the molecular processes regulating the life cycle of these viruses is critical for the design of potent anti-viral therapies. By taking advantage of the high-mutation rate of these viruses, coevolution analysis represents a valuable approach to decrypt viral protein functional rearrangements and provide basis for their inhibition. Here, we employed a recent coevolution analysis method, BIS [22–24] to provide a proof of concept of such approach.
Coevolution signals detected within DENV E and Pr recapitulated several structural features of DENV E/E-Pr protein complexes in different conformational states, hence highlighting the structural accuracy of BIS predictions. Coevolution analysis of HCV E1E2 sequences from several genotypes and sub-types led to the identification of several coevolution signals in HCV E1E2 and suggested that E1 and E2 are strong coevolving partners that refold interdependently during fusion (Fig 8). Importantly, the E2 BL emerged as a key element of these rearrangements that could mediate the transition of E1E2 complex from a pre-fusion to a post-fusion conformation. We propose that during this transition, the endogenous E2 BL packs with the front sheet of E2. Thus, a recombinant soluble BLd-H77 could compete with the endogenous E2 BL and block HCV fusogenic rearrangements, then leading to inhibition of membrane fusion (Fig 8). Such competition is consistent with the idea that E2 may harbor a stretched pre-fusion structure exposing internal epitopes. The existence of other E2 structures is also supported by the ability of neutralizing antibodies to target an epitope that is not exposed in the E2core structure (i.e. aa305-324 [33]). Whether the entire BLd-H77 or specific BL amino acids regions are sufficient to inhibit HCV fusion remain to be determined. However, a recent study aiming to screen E2-derived peptide inhibiting HCV infection did not identify peptides within the BL [34], suggesting that a large fraction of the BL, instead of a specific amino acid region, is likely to be required to inhibit HCV entry. This hypothesis reinforces the idea of a physical, but not functional, competition between the endogenous BL and BLd-H77. Our work also suggests that the E2 BL acts in close collaboration with E1 and that these domains are probably in close proximity in E1E2 heterodimer. Indeed, E2core structure is highly concealed by glycans at the exception of the BL [7]. This, combined with the fact that E1 and BL do not seem to represent preferential targets of anti-HCV neutralizing antibodies, suggests that the BL region and E1 may conceal respective epitopes. The critical interplay between BL and E1 during fusion was also demonstrated at the amino acid level, as BIS was able to accurately identify coevolution signals between specific residues of E1 and the BL that tightly regulate HCV fusion (Fig 8).
By employing an original multi-disciplinary approach combining computational analysis and experimental assays, our work sheds light on potentially important structural and functional features of the HCV fusion mechanism. Although the clear roles of E1 and E2 during the HCV fusion process still remain to be better defined through the structural resolution of the E1E2 heterodimer at different pH, our work initiates a path toward an experimentally-supported model for HCV-cell fusion.
In this putative model, E1 would play the role of a fusion protein protruding onto the virus surface whereas E2 would be a receptor binding protein and a fusion chaperone concealing E1 epitopes. Although E2core is a truncated E2 protein and may not exhibit full E2 properties, E2core does not respond to pH variation [6]. This, combined with the fact that E2 does not harbor fusion protein structural features, could suggest that E2 needs to be associated with E1 to undergo conformational changes and to chaperone E1 fusion-promoting rearrangements. Interestingly, non-conserved E1 residues swapped between the J6 and 2b1 envelope (Fig 6) were located at a very close proximity upstream of the putative E1 fusion peptide (CSALYVGDLC) [11–16]. This result may suggest that the interaction of these E1 residues with E2 BL could participate to critical E1-E2 rearrangements leading to the fusion peptide insertion into the endosomal membrane. Following receptor priming [35] and insertion of a putative E1 fusion peptide into the endosomal membrane, HCV fusion could be triggered by a fold-over of the C-terminal domain of E1 toward its N-terminal domain. This rearrangement could be mediated by E2 BL movements, or alternatively, could promote E2 BL movements. Overall, this interdependent refolding would result in the packing of the BL and of the E2 C-terminal region toward the E2 front sheet, and to the fusion of the host and viral membranes (Fig 8).
The Ig-fold β-sandwich structure of E2core has been proposed to display similarities with domain III or B from class II fusion proteins [7]. By chaperoning E1 fold-over and membrane fusion, our results support that E2 function could be related to a domain III-like, as recently suggested [36]. Further structural studies aiming to resolve the full structure of the entire HCV heterodimer in pre- and post-fusion conformations will be needed to decipher the integral fusogenic rearrangements. Despite the fact that the structural differences detected for pestivirus E2 vs. HCV E2 glycoproteins makes unlikely that these viruses harbor a similar fusion mechanism, it is possible that their E1 proteins could be derived from a common ancestor and represent so far a potential new class of fusion protein as suggested by others [9,18]. Thus, although the resolution of the E2core structure has undermined the role of E2 as a fusion protein, HCV fusion is likely mediated by two interdependent partners that display original structures and conformational changes. To our knowledge, our data provide unique evidence that a component (in this case, the BL) derived from a receptor-binding protein with no fusion peptide can modulate viral fusion rearrangements, hence highlighting HCV fusion as a unique mechanism among known enveloped viruses.
A primary limitation of current coevolution analysis approaches relies on the availability of a large number of sufficiently divergent evolutionarily-related sequences [37]. Such sets of sequences constitute the bottleneck for today’s coevolution analysis methods. We have shown previously that BIS can overcome these limitations and can address coevolution of conserved sequences such as viral genotype sequences [22–24]. Consistently, we failed to detect coevolution signals within our sets of E1E2 sequences when employing other existing coevolution methods (such as DCA, PSICOV and EVcouplings [38–40]), methods that we previously discussed in [23].
In strong contrast to our previous work [23] that was critical to understand the computational power of coevolution analysis applied to viral sequences, our current study goes beyond the speculative and computational predictions and is fundamental is several ways. First, exploring coevolution signals when no or little structural and functional information are available remain highly challenging and hamper the delineation of undescribed viral protein-mediated processes. As coevolution signals imply structural or functional associations between coevolving residues [22,23], our work shows that the BIS method is able to successfully highlight E1E2 contacts that likely orchestrate structural rearrangements of the heterodimer complex. In this respect, our work constitutes an important demonstration of BIS ability to decode structural features of major conformational change in proteins families characterized by few and conserved sequences. Hence, our work provides evidence that computational analysis of coevolution with BIS can be fruitfully used to find direct (and possibly indirect) contacts between proteins where the three-dimensional structure and rearrangements of protein complexes are not known. Second, it demonstrates that coevolution analysis can highlight the existence of conformational changes in proteins through pairs of coevolving residues that are not in contact in the known crystal structure. Many of the coevolution analysis tools developed in recent years (such as DCA and DCA-like approaches) are detecting “direct contacts” and justify their success by using the 3D crystal as their evidence of true positive predictions. In our study, we show that this idea only represents a part of the truth. Proteins are more complex systems, undergoing different structural conformations during their lifetime, and that evolutionary signals code not only for direct interaction but also for intermediate folding states and alternative structural conformations. Third, we advance in the comprehension of how computational techniques can be used to help revealing protein “contacts” that are biologically interesting, within and among structures. Finally, we provide experimental evidence of the biological significance of the coevolution signals for viral genotype sequences, hence allowing for a rapid identification of critical residue contacts regulating protein functions and conformations.
Beyond uncovering important features of the HCV fusion mechanism, our study provides altogether a proof of concept that coevolution can be successfully harnessed to decrypt viral protein rearrangements and interactions, as well as to expand our knowledge of viral proteins-mediated biological processes. Virus entry mechanisms and envelope glycoproteins conserved epitopes represent valuable targets for the development of drugs and innovative vaccine strategies against challenging human pathogens [41–46]. By unscrambling key protein interactions or rearrangements, our work demonstrates that coevolution predictions can be of considerable value for stimulating a fast-tracked design and screening of innovative translational approaches and antiviral strategies unimpeded by virus plasticity.
Experiments were performed in accordance with the EU guidelines (Directive 2010/63/EU) on approval of the protocols by the local ethical committee (Authorization Agreement C2EA-15; Ethic commitee: Comité d’Evaluation Commun au Centre Léon Bérard, à l’Animalerie de transit de l’ENS, au PBES et au laboratoire P4 (CECCAPP), Lyon, France.
Human Huh-7.5 (kind gift of C. Rice, Rockefeller University, NY), BRL3A rat hepatoma (ATCC CRL-1442), and 293T kidney (ATCC CRL-1573) cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal calf serum (Invitrogen). Primary human hepatocytes (PHH; BD Biosciences) were centrifuged using a F-12 HAM medium (Sigma) and seeded overnight in collagen-coated 48 well plate (1x105 cells/well) into Gentest seeding medium (BD Biosciences) completed with 5% FCS. The next morning, PHH were washed and cultured with the culture medium for PHH William’s E medium (W4128, Sigma) supplemented with 7,5% BSA, 1% ITS (insulin transferrin selenium, Gibco), 10-7M of Dexamethasone (Sigma), 1% of non-essential amino acids (Gibco), 1% of Glutamax (Gibco) and 1% Penicillin-Streptomycin solution (Gibco). Sera containing HCV particles were obtained from a gt1b infected patient (Hôpitaux Universitaires de Strasbourg, Strasbourg, France) and were subsequently amplified in uPA-SCID humanized liver mice. Viral loads (RNA copies number/ml) were determined by RT-qPCR using a clinical diagnostic kit (Abbot).
The rat anti-E2 clone 3/11 [47], the mouse anti-E2 AP33 (Clayton et al., 2002) and the conformational mouse anti-HCV E2 H53 [48] are kind gifts from J. McKeating (University of Birmingham, UK), Arvin Patel (MRC—University of Glasgow Centre for Virus Research, Glasgow, UK) and J. Dubuisson (Institut Pasteur de Lille, FR) respectively. AR3B [32] and AR4A [30] antibody are a kind gift from Mansun Law (The Script research institute, San Diego, USA). MLV Capsid was detected by a goat anti-MLV-CA antibody anti-p30 (Viromed). HCVcc foci forming units were stained with a mouse anti-HCV NS5A antibody 9E10 [49], a kind gift of C. Rice (Rockefeller University, NY, USA). Human CD81 were detected with JS81 mAb (BD Biosciences), human SRB1 with CLA-1 mAb (BD Pharmingen), human Claudin-1 with the MAB4618 mAb (R&D Syst.) and human Occludin with an anti-Occludin mAb targeting the C-terminal region of the protein (Laboratories Inc.). A Mouse anti-Human IgG2 antibody targeting human IgG2 hinge region (Novus Biological) was used to quantify BLd-tm expression at cell surface. A rabbit 6His-tag antibody (Pierce antibody) was used to detect soluble E2 and BLd-H77 6His-tag.
In order to construct a relevant BLd soluble peptide derived from a genotype 1a sequence, we took into account the BL borders previously suggested (Fig 3B, see also reference [7]), but also aimed to refine these borders using the locations of BIS coevolution networks of genotype 1a. Indeed, BIS suggested that a 15 amino acids extension from residue 390 to 405, classified previously as an undefined domain [7] between the central Ig scaffold and the E2 BL, contained blocks (S4 Fig, block 7–1) that coevolve with another region of the BL (block 7–2), forming altogether a fusion cluster (cluster 7). Hence, in order to rigorously challenge BIS predictions that suggested a potential redefinition of the BL, we constructed a soluble BLd peptide from residue 390 to 460. Hence, DNA sequence of HCV E2 coding for the residues 390 to 640 was amplified from a genotype 1a envelope H77 (AF009606) cDNA sequence. This sequence was subcloned into a phCMV plasmid to fuse the last 18 amino acids from the C-term part of HCV core that act as a signal peptide, and a CH3-terminal 6 Histidine-tag was added. The resulting peptide of 77 amino acids (71+6) was then named as BLd-H77.
BLd-H77 was expressed in 293T following transient transfection and purified from cell culture supernatant (OptiMEM) by fast protein liquid chromatography on a Superdex G-75 gel filtration column (GE Healthcare). BLd-H77 was re-suspended in 1XPBS. The concentration of purified BLd-H77 peptides was determined by absorption at 205 nm. The mass of BLd-H77 peptide was measured by ESI mass spectrometry using a Finnigan LCQ ion trap mass spectrometer (Thermo Electron Corporation). Analysis by SDS-PAGE electrophoresis was performed using a standard Tris–glycine system and 11% acrylamide gels, in reducing or non-reducing condition. Electrophoresis was followed by either coomasie blue staining or western immunoblotting with an anti His tag antibody.
An anchored form of BLd-H77 was engineered by adding a hinge region (human IgG2) and the transmembrane domain of the CD34 protein to the E2 BLd. The construct, named BLd-tm, was then inserted into Gae-SFFV-IRES backbone harboring selectable marker gene P140K MGMT. A similar construct, but encoding for a HIV-1 gp41 fusion inhibitor peptide [50], namely C46, was used as a control. Construct details are available upon request. Lentiviral vectors transducing BLd-tm or C46 were produced from 293T cells. Stable expression in Huh-7.5 was obtained by transduction with vector particle-containing supernatants of 293T producer cells, followed by selection of O6-benzylguanine and BCNU. BLd-tm expression in Huh-7 cells was quantified by FACS analysis using a mouse anti human IgG2 and an anti-mouse APC antibody.
BRL cells expressing CD81 or SRBI were washed and stained for 1h at 4°C with a mouse anti-human IgG2 antibody (for detection of C46 and BLd-tm), an anti-CD81 (JS81) and with an anti-human SR-BI (CLA-1) respectively. Cells were then washed and incubated with a secondary anti-mouse or rat APC antibody for 1h at 4°C. Cell surface expression levels were then quantified by flow cytometry (FACS CANTO II–BD Biosciences). For HCV receptor detection on Huh7.5 and Huh7.5-BLd-tm cells, cells were fixed with 2% formaldehyde for 20 min at room temperature and washed. For Occludin staining, cells were permeabilized with Perm/Wash Buffer (BD Biosciences) for 15 min at 4°C prior staining. Human CD81 was stained with JS81 mAb, human SRBI with CLA-1 mAb, human Claudin-1 with the MAB4618 and human Occludin with an anti-Occludin mAb targeting the human Occludin C-terminal region. To characterize the effect of BLd-H77 on HCV receptor expression, Huh7.5 were incubated overnight with BLd-H77 (50μg/ml) prior staining.
HCVpp were produced as previously described [21,51] from 293T cells cotransfected with a murine leukemia virus (MLV) Gag-Pol packaging construct, an MLV-based transfer vector encoding the green fluorescent protein, and E1E2 envelope expression constructs H77 (AF009606), A40 (unreferenced), UKN1B 12.6 (AY734975), UKN2A 2.4 (AY734979), JFH-1(AB047639), J6 (AF177036), UKN2B 2.8 (AY734983), UKN3A 1.9(AY734985), UKN4 21.16 (AY734987), UKN5 14.4 (AY785283), HK 6A-2.1 (FJ230883) or control envelope HA-NA (CY077420) and VSV-G (AJ318514). All chimeric J6/2b1 E1E2 heterodimers were constructed by molecular cloning, PCR and/or digestion between the genotype 2a envelope J6 (AF177036) and a genotype 2b envelope UKN-2b1 (unreferenced). All chimeric H77/A40 E1E2 heterodimers were constructed using a similar strategy between the genotype 1a envelope H77 (AF009606) and A40 (unreferenced, but previously employed [21]). 72 to 96h following infection, percentage of infected cells was quantified by FACS Canto II or LSRII (BD Biosciences) to quantify GFP expression. For BLd-H77 dose-dependent neutralization assay, HCVpp H77 or VSVGpp were pre-incubated with different concentrations of BLd-H77 or with PBS for 1h at room temperature and were then used to infect Huh7.5. For the time-course neutralization assay, Huh7.5 were infected with HCVpp H77 for 4h prior washing. PBS, BLd-H77 (50 μg/ml), Bafylomycin A1 (20nM) or AR4A (25 μg/ml) were added into cell supernatant for 1h prior infection, during infection or after infection. Percentage of infected cells was determined 72h following infection. For HCVpp co-neutralization assay, HCVpp H77 pseudoparticles were pre-incubated for 1h at room temperature with BLd-H77 alone (35 or 50 μg/ml), AR4A alone (2 or 17 μg/ml) or with both BLd-H77 and AR4A (35 and 2 μg/ml, or 35 and 17 μg/ml). Pre-mixes were then used to infect Huh7.5 and media was changed 6h post infection. GFP intracellular levels were quantified 4 days post infection by flow cytometry. For HCVpp post-binding neutralization assay, HCVpp-H77 pseudoparticles were incubated with Huh7.5 in presence of BLd-H77 (50 μg/ml) or AR4A (25μg/ml) for 1h at 4°C (binding), for 4h at 37°C following binding (entry), or for 72h following entry (post-entry). As control, Huh7.5 were incubated and infected with HCVpp-H77 in a similar manner but not treated with BLd-H77 or AR4A.GFP intracellular levels and related percentage of infection were then quantified by flow cytometry. For HCVpp containment assay on Huh7.5-BLd-tm, Huh7.5 or Huh7.5-BLd-tm were infected or not with HCVpp H77 for 5h. Then, cells were washed and E2 cell surface expression was determined by flow cytometry following staining using the anti-E2 H53 antibody and a secondary anti-mouse APC antibody. Huh7.5 and Huh7.5-BLd-tm were infected similarly and GFP expression of infected cells was analyzed 72h post-infection.
Transfected 293T cells were lysed and nuclei were removed by centrifugation at 12 000 rpm for 10 min. HCVpps were purified and concentrated from the cell culture medium by ultracentrifugation at 82,000xg for 1h 45 min through a 20% sucrose cushion. Cell lysates and viral pellets were subjected to western blot analysis using 3/11 anti-E2 antibody and an anti-MLV-CA antibody as described previously [21].
Plasmid pFK H77/JFH1/HQL (kind gift of R. Bartenschlager), termed as H77/JFH-1, displaying HCV genome with adaptive mutations (Y835H in NS2, K1402Q in NS3, and V2440L in NS5A) and harboring H77 sequence derived from the BLd-H77peptide, as well as plasmid pFKi389-Venus-Jc1, termed as Jc1, (an intra-genotypic recombinant between J6-CF sequence (AF177036) and JFH1 sequence) were used to produce and electroporate into Huh7.5 the respective H77/JFH-1 and JC1 viral RNAs as described previously [21]. Huh7.5 cells, Huh7.5-BLd-tm or Huh7.5-C46 were infected with different dilutions of culture supernatants harvested at 24h, 48h and 72h post electroporation. Four days post-infection, cells were fixed with EtOH 100% and foci forming units (FFUs) were visualized after NS5A immunostaining as described previously [21]. For BLd-H77 dose-dependent neutralization assay, HCVcc particles were pre-incubated with different concentrations of BLd-H77 or with PBS for 1h at room temperature and were then used to infect Huh7.5. For time-course neutralization assay, Huh7.5 were infected with HCVcc for 4h prior washing. PBS or BLd-H77 (35 μg/ml) were added into cell supernatant for 1h prior infection, during infection or after infection. FFU/ ml were determined 4 days post-infection. To construct HCVcc particles harboring 2b1, J6-1/2, 2b1-1/2 or 2b1-1/3 envelope, we inserted by molecular cloning the related envelope into the pFKi389-Venus-Jc1 molecular clone, that initially encodes for J6 envelope. Viral RNAs were electroporated into Huh7.5 as described above. At 72h post electroporation, cell culture supernatants were tittered and used to infect naïve Huh7.5. Number of foci forming units per ml were determined 4 days post infection as described above. In parallel, viral RNA were extracted from electroporated cell culture supernatants at 72h post (ZR viral RNA kit, Zymo). HCV viral RNA copy number was quantified by one-step reverse transcription-PCR (RT-PCR) using MultiCode-RTx Real-Time PCR (Luminex) according to manufacturer’instructions and run on a Step One Plus quantitative PCR machine (Life Technologies). Data were analyzed using the MultiCode Analysis Software v1.6.5 (Luminex). The following primers were used for the detection of HCV RNA: GCTCACGGACCTTTCA (sense) and GGCTCCATCTTAGCCC (antisense).
H77/JFH1 virus was used to infect Huh7.5 and Huh7.5-BLd-tm (m.o.i. 0,1). At day 1, 3 and 5 post infection, cells were fixed with 2% formaldehyde for 20 min at room temperature, washed and permeabilized with Perm/Wash Buffer (BD Biosciences) for 15 min at 4°C. NS5A expression levels were then quantified using anti-NS5A antibody 9E10 by flow cytometry (FACS CANTO II–BD Biosciences). In parallel, cell supernatants were harvested at each time point, filtered and used to infect naïve Huh7.5. 72h post infections, infectious titers (FFU/ml) of cell supernatant were determined as described above.
HCVpp H77 and H77/JFH-1 HCVcc particles were preincubated for 1h at room temperature with PBS or BLd-H77 (50 μg/ml or 35 μg/ml respectively). Then, BLd-H77 and concentrated viral particles were diluted (1/5) with cell culture media or not prior infection of Huh7.5. 4 days post infection, percentages of GFP-positive cells were determined by flow cytometry and FFUs/ml were determined by NS5A immunostaining as described above.
H77/JFH-1 were used to infect Huh7.5 for 4h at 37°C (m.o.i. 0,05). After washing, cells were incubated with anti-E2 antibody AP33 (25μg/ml) alone, or mixed with BLd-H77 (35μg/ml), or with PBS. 72h post infection, cells were fixed and numbers of cell per foci for each condition were quantified through NS5A immunostaining as described above.
PHH were washed and infected by JC1 HCVcc virus at different m.o.i. (0,005; 0,001; 0,05; 0,1). 4 days post infection, cell culture supernatants were harvested and used to infect naïve Huh7.5. Infectious titers were revealed through NS5A immunostaining 4 days post infection as described above. Indirect titrations were performed as infected PHH were poorly detectable through NS5A immunostaining, making the indirect titration the only accurate method to quantify the amount of infectious viral particles (and not physical viral particles) in a PHH-cell culture supernatant. For neutralization assay, JC1 virus or JC1-derived HCVpc were pre-incubated with BLd-H77 (10, 20 and 40 μg/ml) or with PBS for 1h prior PHH (m.o.i. 0,05) or Huh7.5 infection respectively (m.o.i. 0,01 and 0,02). 4 days post infection, infectious titers of PHH cell culture supernatants were determined by infecting Huh7.5 as described above. In parallel, infectious titers of JC1-derived HCVpc following Huh7.5 infection were quantified as described above. Sera containing HCV particles were used to infect PHH at a m.o.i. of 0,1 following BLd-H77 (30 μg/ml) or PBS preincubation for 1h at room temperature. 4 days post infection, cell culture supernatants were harvested and viral loads (RNA copies number/ml) were determined by RT-qPCR using a clinical diagnostic kit (Abbot Real Time™ HCV assay) with a limit of quantification (LOQ) of 12 IU/mL (i.e. 51.6 HCV RNA copies/ml). Given a serum dilution of 1:100 in PBS, LOQ = 5160 HCV RNA copies/ml.
Retroviral vectors expressing human CD81 (NM_004356) and SR-BI (Z22555) were described previously [52]. Retroviral vectors containing these cDNAs were produced from 293T cells as VSV-G pseudotyped particles as described previously [53,54]. Stable expression of either receptor in BRL cells was obtained as described previously [52].
Binding assays were performed as described previously [21]. Briefly, 50μl of concentrated virus (100x) or 100 ul of concentrated soluble E2 (100x) were pre-incubated with BLd-H77 (50 μg/ml) or with PBS for 1h at room temperature. Then, pseudoparticles or soluble E2 were mixed with Huh7.5, BRL, BRL-CD81 or BRL-human SR-BI in presence of 0.1% sodium azide for 1h at 37°C. Cells were then washed with PBFA (PBS, 2% fetal bovine serum, and 0.1% sodium azide). Bound viruses were detected using the mouse H53 anti-HCV E2 antibody and soluble E2 was detected using either the H53 antibody or a rabbit 6His-tag antibody for 1h at 4°C. After washing, primary antibodies were quantified by flow cytometry (FACS Canto II, BD Biosciences) using APC goat anti-mouse immunoglobulin-G. In parallel, concentrated HCVpp were pre-incubated with 50 μg/ml of BLd-H77 and used to infect Huh7.5 in order to verify the neutralizing effect of BLd-H77 on the entry of concentrated HCVpp. For BLd-H77 binding assay, BLd-H77 (50 μg/ml) or PBS were mixed with Huh7.5 for 1h at 37°C prior 6His-tag staining using a rabbit 6His-tag antibody.
Concentrated soluble E2 (100x; 100 or 250 μl) were mixed with equivalent number of Huh7.5 or Huh7.5-BLd-tm (2x105 or 5x105 cells) in presence of 0.1% sodium azide for 1h at 37°C. After washing, bound soluble E2 were detected using anti-E2 H53 antibody as described above. Levels of binding enhancement were determined relatively to the basal E2 binding on naïve Huh7.5. For HCVpp binding enhancement assay, 1x105 Huh7.5 cells or Huh7.5-BLd-tm were infected with HCVpp for 4 hours. Cells were then trypsinized and washed with PBFA (PBS, 2% fetal bovine serum, and 0.1% sodium azide). Bound viruses were quantified by flow cytometry (FACS Canto II, BD Biosciences) following cell staining with the mouse H53 anti-HCV E2 and an APC goat anti-mouse immunoglobulin-G.
JC1 HCVcc particles (5x104 i.u.) were pre-incubated with BLd-H77 (35 μg/ml), Heparine (250 μg/ml) or with PBS for one hour at 37°C. Viral particles were then mixed with 1x105 Huh7.5 cells for 2h at 4°C. After 3 washings, cells were lysed using the RLT Buffer (QIAGEN) complemented with β-mercaptoethanol. Total RNAs were then extracted using the RNeasy mini kit (QIAGEN) as recommended by the manufacturer. Extracted RNAs were reverse transcribed using the iScript cDNA synthesis kit (Bio-Rad) and HCV viral RNA (5-CTTCACGCAGAAAGCGCCTA and 5-CAAGCGCCCTATCAGGCAGT) and human GAPDH (5- GAAGGTGAAGGTCGGAGTC and 5- GAAGATGGTGATGGGATTTC) were then quantified by qPCR using the FastStart Universal SYBR Green Master kit (Roche Applied Science) on an Step One Plus quantitative PCR machine (Life Technologies). Cell-associated viral RNA copies were normalized on human GAPDH expression for each sample.
96-well plates (Corning) were coated overnight with different amounts of a mouse IgG isotype (10ng and 100ng; Abcam), anti-E2 antibody AR3B [32] (10 and 100ng), and BLd-H77 (10, 100 and 250ng). The next day, following a one-hour incubation step with a SuperBlock blocking buffer (Thermo Scientific) to prevent non-specific binding, each coating condition was incubated with 10ng of sE2 or not. Interactions were revealed using a rat anti-E2 antibody 3/11 [47] or a rat IgG isotype, then followed by an incubation with an anti-rat HRP antibody (Biorad). Optical density signals at 450 nm were then assessed using a TriStar Multimode Microplate reader (Berthold).
FRG (Fah–/–Rag2–/–Il2rg–/–) mice (mixed background: C57BL/6 and 129Sv) were housed in our animal facility (Plateau de Biologie Experimentale de la Souris, PBES, Lyon, France). Because of their lethal phenotype, mice are maintained on 8 mg/l of NTBC (nitro-4-trifluoro-methylbenzoylcyclohexanedione) in the drinking water. 48h prior the engraftment, adult (6–10 weeks old) mice were injected intravenously with 2x109 p.f.u. of an adenoviral vector encoding the uPA transgene. 7x105 to 1x106 PHH (BD Biosciences) were injected intrasplenically as previously described [55]. Immediately after engraftment, the NTBC was progressively withdrawn as follow: 2 days at 10% of colony maintenance concentration, then 2 days at 5%, 2 days at 2.5%, then the NTBC was completely removed. During the phase without NTBC, mice were weighted every two days. After 2–6 weeks, mice with clinical symptoms (lethargy, hunched posture) or severe weight loss (>15%) were put again on NTBC for 3 days before second withdrawal (cycling). Cycling was repeated until clinical symptoms resolved. In order to prevent the development of a murine hepatocellular carcinoma, highly reconstituted mice selected for infectious experiments were subjected to further NTBC treatment (3–4 weeks w/o NTBC and 3 days with 100%).
Blood from transplanted mice and controls were collected every 2–4 weeks after engraftment by retro orbital puncture. Sera were sent to a diagnostic laboratory for quantification of human Albumin (Cobas C501 analyzer, ROCHE).
Highly reconstituted mice (HSA >15mg/ml) were infected with JC1 HCVcc particles inocula (105 i.u.) or with patient-derived HCV particles (5x104 i.u.) via intraperitoneal route. Mice sera were collected at day 7 and day 14 post-infection by retro-orbital bleeding. At day 21, mice were sacrificed, sera were collected and levels of human albumin were determined (Cobas C501 analyzer, ROCHE). JC1 infectious titers in mice sera were determined through infection of Huh7.5 with different dilutions of sera as described above. In parallel, for HCV particle containing sera, viral load was determined by RT-qPCR using a clinical diagnostic kit (Abbot Real Time™ HCV assay) with a limit of quantification (LOQ) of 12 IU/mL (i.e 51.6 HCV RNA copies/ml). Given a serum dilution of 1:100 in PBS, LOQ = 5160 HCV RNA copies/ml.
A first cohort of 11 mice was attributed for an in vivo inhibition assay of JC1 infection (9 mice + 2 negative controls; one non-infected and one non-engrafted). 7 mice were treated under a prophylactic protocol with PBS (4 mice) or with 30 μg (3 mice) of BLd-H77. Mice were treated via intra-peritoneal route one day prior infection, and at day 1, 7 and/or 14 post-infection infection. For all the mice, sera were harvested and viral titers were quantified at day 7, 14 and during sacrifice 21 post infection. A second independent cohort of 7 mice was attributed for another in vivo inhibition assay of JC1 infection (5 mice + 2 negative control). Here, JC1 infection was challenged with a dose of 150μg of BLd-H77 (2 mice) or with PBS (3 mice) under a prophylactic protocol similar as described above. A cohort of 14 mice (12 mice + 2 negative control) was attributed for an in vivo inhibition assay of serum-derived HCV particles infection, challenged with one dose of BLd-H77 (150 μg, 5 mice) or with PBS (7 mice) under a prophylactic protocol similarly to what has been described above. Serum-derived HCV particles viral load were determined as described above. Following sacrifice, the level of human Albumin was quantified for each untreated and treated mice in order to ensure that HCV infection inhibition was not due to a decrease of liver humanization caused by BLd-H77.
Cell-cell fusion assays were performed as described previously [13,19]. Briefly, HEK-293T cells (2.5x105 cells/well seeded in six-well tissue culture dishes 24 h before transfection) were co-transfected using calcium phosphate reagent with a HCV (H77, J6, 2b1 or J6/2b1 chimera) or VSV-G envelope encoding-plasmids and with an HIV-1 LTR (long terminal repeat) luciferase reporter plasmid. After 12h, transfected HEK-293T cells were detached with versene (0.53 mM EDTA; Invitrogen) and co-cultured (5x104 cells/well) with Huh-7-Tat indicator cells (5.104cells/well) in a 24-well plate. After 24 h, the cells were washed with serum free DMEM, incubated for 3 min in either pH 7 or pH 5 buffer (130 mM NaCl, 15 mM sodium citrate, 10 mM Mes and 5 mM Hepes) and then washed three times with serum free DMEM. The luciferase activity was measured 72 h later using a luciferase assay kit according to the manufacturer’s instructions (Promega). For fusion neutralization assay, coculture were washed, pre-incubated with BLd-H77 (50μg/ml) or with PBS for 1h at 37°C prior a second washing and exposure to pH buffer. Alternatively, co-cultured cells pre-treated by BLd-H77 or PBS were incubated for an extension of 24h following pH shock by BLd-H77 (50μg/ml) or PBS.
HCVpp/liposome lipid mixing was performed as previously described [21,56]. R18-labeled liposomes were obtained by mixing octa-decyl rhodamine B chloride (R18; Molecular Probes) and lipids (phosphatidylcholine and cholesterol; Aventi) and mixed with 40 μl of concentrated HCV pseudoparticles (non-enveloped HCVpp, HCVpp-H77 and HCVpp harboring the fusion defective envelope E140/E2H77 previously characterized [21]) or retroviral particles pseudotyped with Influenza Hemagglutinin-Neuraminidase (HANApp), all diluted in PBS (pH 7.4) within a 37°C thermostable 96-well plate. After pH decrease to 5 (acidification), dequenching of R18 due to lipid mixing between HCVpp and liposomes were recorded on a micro-plate fluorometer (InfiniteM1000 Tecan Group Ltd) for a period of 5 to 20min with an excitation wavelength (λexc) at 560 nm and an emission wavelength (λem) at 590 nm. Maximal R18 quenching was measured after the disruption of liposomes by the addition of 0.1% TritonX-100. For fusion-neutralization assays, pseudoparticles were pre-incubated with different doses of BLd-H77 or with PBS for 1h prior incubation with liposomes and acidification.
Structural analysis of Dengue E pre-fusion structure (PDB 1K4R), E-Pr complex structure (3C6E), E post-fusion structure (1OK8) and E2core structure (4MWF) were realized using Chimera software [57] (UCSF).
GraphPad Prism software (version 6) was used for statistical analysis. Statistics were calculated using Student’ t test and/or two-way ANOVA when appropriate. (*p<0.05, **p<0.01, ***p<0.001).
Protein amino acid sequence alignments were realized using ClusterX2.1 and rendered as Post-script format.
The BIS methodology and related-coevolution signal analysis are described in details in reference [22], [23] and [24]. Applicability of BIS for detecting coevolution signals in viral sequences is specifically described in reference [23].
Literature describing amino acids mutations within E1E2 sequences impacting E1E2 folding/heterodimerization, E1E2 binding to cellular receptors or E1E2 fusion was used as references to attribute functions to clusters mapping specific regions within E1E2 sequences. The references used were the following: Folding/Heterodimerization [19,27,28,58–60], Viral binding site conformation [19,20,26,60–66] and Fusion [11,13,19–21,27,67].
|
10.1371/journal.ppat.1004131 | Human Cytomegalovirus Fcγ Binding Proteins gp34 and gp68 Antagonize Fcγ Receptors I, II and III | Human cytomegalovirus (HCMV) establishes lifelong infection with recurrent episodes of virus production and shedding despite the presence of adaptive immunological memory responses including HCMV immune immunoglobulin G (IgG). Very little is known how HCMV evades from humoral and cellular IgG-dependent immune responses, the latter being executed by cells expressing surface receptors for the Fc domain of IgG (FcγRs). Remarkably, HCMV expresses the RL11-encoded gp34 and UL119-118-encoded gp68 type I transmembrane glycoproteins which bind Fcγ with nanomolar affinity. Using a newly developed FcγR activation assay, we tested if the HCMV-encoded Fcγ binding proteins (HCMV FcγRs) interfere with individual host FcγRs. In absence of gp34 or/and gp68, HCMV elicited a much stronger activation of FcγRIIIA/CD16, FcγRIIA/CD32A and FcγRI/CD64 by polyclonal HCMV-immune IgG as compared to wildtype HCMV. gp34 and gp68 co-expression culminates in the late phase of HCMV replication coinciding with the emergence of surface HCMV antigens triggering FcγRIII/CD16 responses by polyclonal HCMV-immune IgG. The gp34- and gp68-dependent inhibition of HCMV immune IgG was fully reproduced when testing the activation of primary human NK cells. Their broad antagonistic function towards FcγRIIIA, FcγRIIA and FcγRI activation was also recapitulated in a gain-of-function approach based on humanized monoclonal antibodies (trastuzumab, rituximab) and isotypes of different IgG subclasses. Surface immune-precipitation showed that both HCMV-encoded Fcγ binding proteins have the capacity to bind trastuzumab antibody-HER2 antigen complexes demonstrating simultaneous linkage of immune IgG with antigen and the HCMV inhibitors on the plasma membrane. Our studies reveal a novel strategy by which viral FcγRs can compete for immune complexes against various Fc receptors on immune cells, dampening their activation and antiviral immunity.
| Herpes viruses persist lifelong continuously alternating between latency and virus production and transmission. The latter events occur despite the presence of immune IgG antibodies. IgG acts by neutralization of virions and activation of immune cells bearing one or more surface receptors, called FcγRs, recognizing the constant Fc domain of IgG. Activating FcγRs induce a wide range of immune responses, including antibody dependent cellular cytotoxicity (ADCC) of virus-infected cells by natural killer (NK) cells, cytokine secretion and the uptake of immune complexes to enhance antigen presentation to T cells. We demonstrate that the HCMV glycoproteins RL11/gp34 and UL119-118/gp68 block IgG-mediated activation of FcγRs. A novel reporter cell-based assay was used to test FcγRs individually and assess their relative susceptibility to each antagonist. This approach revealed that gp34 and gp68 block triggering of activating FcγRs, i.e. FcγRI (CD64), FcγRII (CD32A) and FcγRIII (CD16). Co-immunoprecipitation showed the formation of ternary complexes containing IgG, IgG-bound antigen and the viral antagonists on the cell surface. Assigning the redundant abilities of HCMV to hinder IgG effector responses to the viral Fc binding proteins, we discuss gp34 and gp68 as potential culprits which might contribute to the limited efficacy of therapeutic IgG against HCMV.
| Human cytomegalovirus (HCMV) constitutes the prototypical human pathogenic β-herpesvirus found worldwide with high immunoglobulin G (IgG) seroprevalence rates of 50–98% [1]. Despite the expression of a very large antigenic proteome of approximately 750 translational products [2], HCMV avoids sterile immunity and invariably persists lifelong in the human host in a latent state with periodic phases of reactivation and virus shedding. While infection of immune competent individuals is usually subclinical, HCMV causes severe symptoms in immunocompromised individuals and congenitally infected newborns [1], [3]. Cytomegalovirus immune control is organized in a hierarchical as well as redundant manner, with crucial roles for natural killer (NK) cells as well as T lymphocytes [4]. HCMV expresses a large set of immune evasion genes that impair recognition of infected cells by CD8+, CD4+ and NK effector cells and thus facilitate virus persistence, spread and superinfection [5]–[7] while cellular immune responses are nevertheless indispensable for CMV immune surveillance. Experimental and clinical evidence suggest that cytomegalovirus can persist for the lifetime by effectively defending itself from both cellular and humoral immunity. In the absence of either viral immune evasion genes or subsets of immune cells, the balance of pathogenesis versus clearance of the virus can be tilted. For example, B cell deficient mice exhibit a much higher susceptibility during recurrent mouse cytomegalovirus (MCMV) infection compared to control mice, reflected by 100–1,000-fold increased titers in the absence of CMV-specific IgG [8]. Moreover, adoptive transfer of memory B cells into naïve Rag−/− mice is sufficient for long term protection from lethal MCMV disease [9], and passive immunization with immune IgG reduces MCMV-induced pathology in newborn mice [10]. In clinical settings, HCMV-immune IgG preparations are used with varying degrees of success. Human intravenous hyperimmune immunoglobulin against HCMV (e.g. Cytotect) significantly lowers the risk of congenital CMV infection and disease at birth when given to primary HCMV-infected pregnant women [11]. Nevertheless, meta-analyses of clinical studies with solid organ transplant recipients as well as patients undergoing hematopoietic stem cell transplantation document little if any benefit of IgG prophylaxis against HCMV disease [12]–[14].
IgG antibodies have two functional domains: the fragment antigen binding (Fab) that contains the paratope recognizing the respective epitope of the antigen and the fragment crystallisable (Fc) which recruits IgG effector functions. Receptors for the Fc domain of IgG (FcγRs) are expressed on immune cells to connect the humoral and cellular branches of immunity. Upon IgG binding and receptor activation, FcγRs trigger a diversity of effector responses including antibody-dependent cellular cytotoxicity (ADCC), phagocytosis, endocytosis of immune complexes and cytokine production. Importantly, the set of human FcγRs includes different activating members, i.e. FcγRI (CD64), FcγRIIA (CD32A), FcγRIIC (CD32C) and FcγRIIIA (CD16) which differ in immune cell distribution, affinity for distinct IgG subclasses [15] and effector functions elicited upon activation [16]–[19].
Fcγ binding activity on the surface of HCMV-infected cells has long been reported [20], but the consequences of Fcγ binding to immune responses are unknown [21]. The HCMV Fcγ binding proteins gp34 and gp68 are type I transmembrane glycoproteins encoded by independent genes, RL11 and UL119-UL118, respectively, which are fully dispensable for HCMV replication in vitro [22], [23]. Both HCMV proteins show cell surface disposition and exquisite ligand specificity for human IgG but no other Ig classes (e.g. IgA or IgM) [22]. Minimal sequence relatedness in their extracellular domains with particular immunoglobulin supergene family domains present in FcγRI and FcγRII/III suggests differing binding characteristics from those of host FcγRs [22]. In contrast to host FcγRs, both gp34 and gp68 recognize Fcγ in a manner independent of N-linked glycosylation, further corroborating a binding mode to Fcγ which is distinguishable from host FcγRs [24].
Cytomegaloviruses frequently reactivate and can super-infect despite the presence of relatively high levels of HCMV-specific IgG [25]–[27] which raises the apparent question by which mechanism HCMV avoids antibody-mediated immune control. One conceivable possibility is that HCMV-encoded FcγRs gp34 and gp68 compete with cellular FcγRs. To test this hypothesis, we had to established a suitable FcγR activation assay that allows a comprehensive analysis of the activation status of individual FcγRs [28]. We compared cells infected with viruses lacking gp34 and/or gp68 upon opsonization with graded doses of polyclonal HCMV-immune IgG. As a control, we included the human α-herpesvirus herpes simplex virus 1 (HSV-1), which expresses a viral FcγR composed of the glycoproteins gE and gI that is known to protect infected cells from ADCC elicited by HSV-immune IgG [29], [30]. Our approach identifies FcγRI, FcγRIIA and FcγRIII as principal targets of both HCMV gp34 and gp68, while the prototypic HSV-1 FcγR gE was found to inhibit only FcγRIIA and FcγRIII. This is the first experimental proof of CMV-encoded glycoproteins interfering with IgG-mediated immunity.
To assess the relative surface density of viral Fcγ receptors on the plasma membrane of HSV and HCMV-infected cells, Fcγ binding was evaluated by flow cytometry using FITC-labeled Fcγ fragment. As expected, Fcγ-FITC surface binding was observed for HSV wt virus- and the gE revertant virus-infected cells, but not for cells infected with ΔgE HSV (Figure 1A). MRC-5 cells infected with either of two HCMV HB5 single vFcγR deletion mutants, HB5Δgp68 or HB5ΔIRLΔgp34 [22], were decorated at the cell surface with diminished levels of Fcγ-FITC compared to wt-infected control cells (Figure 1A). Cells infected with a HCMV mutant lacking both gp68 and gp34 [31] showed only very low Fc-binding when compared to mock-infected fibroblasts (Figure 1A). Together with our previous experiments documenting Fcγ binding upon ectopic expression of gp34 and gp68 using recombinant vaccinia viruses (rVACV) [22] these data define gp34 and gp68 to be sufficient and essential for Fc binding by HCMV-infected cells.
In productively infected cells, herpesvirus gene expression is regulated in a cascade fashion. Viral proteins encoded by genes of the early phase of infection are required for viral DNA replication, which is a prerequisite for the subsequent expression of structural virion proteins during the late phase of gene expression. To assign the immune-dominant HCMV and HSV surface antigens recognized by opsonizing IgG to the temporal class of genes, we applied the novel reporter cell system allowing quantification of host FcγR activation [28]. This assay is based on co-cultivation of antigen-bearing cells with reporter cells stably expressing FcγR-ζ chain chimeric receptors which produce mouse IL-2 upon recognition of immune IgG, provided that the opsonizing antibody is able to activate the particular FcγR [28]. Importantly, BW5147:FcγR-ζ reporter cells are neither activated by cells lacking the appropriate antigen (e.g., non-infected cells) nor by antigen-expressing cells which had been cultivated in absence of IgG or in presence of non-immune IgG, proving strict antigen- and immune IgG specificity of the assay (Figure S1 and Reference [28]). Late phase gene expression was blocked using 250 µg/ml phosphonoacetic acid (PAA) which blocks the viral DNA polymerase and is the active component of the clinically approved anti-HCMV drug Foscarnet. At 72 and 48 hpi, resp., infected cells were opsonized with graded concentrations of intravenous immunoglobulin (IVIG) Cytotect as a source of human HCMV- and HSV-immune IgG. As expected, immune IgG did not induce receptor activation (IL-2 response) in the presence of mock-infected cells (Figure 1B). While late antigens of HCMV efficiently triggered FcγRIII reporter cells, infected cells arrested in the early phase of replication elicited very poor if any responses (Figure 1B). On the contrary, early antigens of HSV-1 were sufficient to efficiently trigger FcγRIII and HSV late antigens only slightly increased FcγR-ζ activation (Figure 1C). Despite the fact that Cytotect is prepared from donors selected for particularly high HCMV IgG titers [11], [32] we found - in agreement with earlier studies [28] - that Cytotect contains higher titers of HSV-immune IgG activating FcγRIII compared to FcγRIII-reactive HCMV-immune IgG. Due to this fact, IVIG dilutions used for HSV-1 and HCMV experiments had to be chosen differently. The relatively poor responses triggered by IgG opsonized HCMV-infected cells was compatible with our hypothesis that HCMV could be able to reduce the activation of FcγRIII by immune IgG, and that vFcγR gp68 and gp34 might represent candidates for such inhibition.
To test the conjecture that vFcγR gp68 and gp34 could prevent host FcγR activation by HCMV IgG, we pursued the BW5147:FcγR-ζ reporter cell approach. To assess if and to what degree viral FcγRs can interfere with host FcγRs activation, we first compared responses of FcγRIII reporter cells co-cultured with IgG-opsonized HSV-1 wt infected vs. HSV-1 ΔgE-infected MRC-5 cells (Figure 2A). Inhibition of ADCC by PBMCs was reported to be a function of the prototypic HSV-1 FcγR gE [30], albeit the specific host FcγRs which are blocked by gE have not been elucidated yet. As before (Figure 1C), we observed a dose-dependent activation of the reporter cells upon co-cultivation with HSV-1-infected cells but not with mock infected cells. The ΔgE HSV-1 mutant led to an increased activation of FcγRIIIA upon opsonization of target cells with Cytotect, in accordance with published data [29], [30]. The same type of result was obtained when using BW5147:FcγRIIA-ζ reporter cells, indicating that gE also antagonizes activation of this host Fcγ receptor. Surprisingly, gE enhanced, rather than inhibited the IgG-dependent activation as deduced from the overall superior activation of the chimeric FcγRI-ζ by wt HSV-1 when compared to HSV-1 ΔgE (Figure 2A).
The results of HSV-1 gE encouraged us to subsequently test HCMV HB5Δgp68 and HB5ΔIRLΔgp34 mutants using the same experimental strategy. MRC-5 fibroblasts were left uninfected or infected with wt-HCMV strain HB5 versus HB5Δgp68 (Figure 2B), or with HB5ΔIRL (the parental virus of the following mutants) versus HB5ΔIRLΔgp34 and HB5ΔIRLΔgp68/Δgp34, resp., a double mutant lacking both gp68 and gp34 (Figure 2C). 72 h post HCMV infection, target cells were incubated with graded dilutions of Cytotect before BW:FcγRIIIA-ζ responder cells were added. HB5Δgp68 induced a clearly higher response over an extended range of IgG dilutions compared to wt-HCMV HB5 infected cells. Likewise, HB5ΔIRLΔgp34-opsonized cells induced clearly higher reporter cell responses compared with HB5ΔIRL opsonized targets, while HB5ΔIRLΔgp68/Δgp34 exhibited only marginal further increase of the response (Figure 2C). These results suggested that cells infected with virus mutants lacking viral Fc-binding proteins elicit exaggerated activation of FcγRIIIA, provided that the amount of opsonizable HCMV antigens is indeed comparable between the viruses analyzed. To verify this supposition, cells were labelled with F(ab)2 antibody fragments prepared from Cytotect and analysed by FACS. As shown in Figure S2, cells infected with HCMV mutants lacking gp34 and/or gp68 did not show higher levels of opsonizing antigens on the plasma membrane. To compare the relative impact of gp68 versus gp34 on FcγRIII activation with a higher degree of accuracy, i.e. in the context of an identical HCMV genome possessing a preserved UL/b′ gene region, another set of targeted vFcγR gene deletions was constructed based on the AD169varL derived BACmid pAD169 which carries unlike pHB5 only a single copy of TRL genes including TRL11 [33]. As demonstrated in Figure 2D, targeted deletion of UL119-118/gp68 and TRL11/gp34 reproduced the increased activation of BW:FcγRIIIA-ζ responder cells, while combined deletion of both vFcγRs only marginally enhanced the response further. Next we determined if gp34 and gp68 could affect further activating host FcγRs and performed co-cultivation assays with MCR-5 cells infected with the same panels of HCMV mutants after opsonization with Cytotect and incubated with BW:FcγRIIA-ζ and BW:FcγRI-ζ reporter cells (Figure 2C–D). While deletion of both HCMV Fcγ-binding proteins resulted in significantly enhanced responses by both FcγRIIA and FcγRI, the isolated removal of gp68 resulted in a slightly more drastic phenotype with regard to BW:FcγRIIA-ζ activation (Figure 2D). Notably, combined removal of gp34 and gp68 led to a Δgp34-like phenotype, contrasting to the additive effect seen with FcγRIII at low IgG concentrations. In conclusion, the data suggested that both of the HCMV-encoded FcγRs might have developed the ability to interfere with the activation of FcγRIII, FcγRIIA and FcγRI, while HSV gE blocks FcγRIII and FcγRIIA but fails to inhibit FcγRI activation.
To exclude the possibility that second site mutations which occurred during the BACmid mutagenesis procedure are responsible for the observed loss of HCMV-mediated inhibition of host FcγR activation by immune IgG, an entirely independent panel of virus deletion mutants and the appropriate rescued versions were generated. The mutants were constructed using the HCMV TB40/E-derived BACmid [34] taking advantage of i) a single gene copy of RL11 coding for gp34, ii) a complete HCMV ULb′ gene region lacking in HCMV HB5 but present in HCMV clinical isolates and iii) a technically more feasible re-insertion strategy of the vFcγR coding genes. MRC-5 fibroblasts were left uninfected or infected with the HCMV TB40/E wt expressing gp68 and gp34, or with gp68 and gp34 single gene deletion mutants, resp., or independent single gene revertant mutants expressing gp68 or gp34. Using BW:FcγRIIIA-ζ responder cells and graded concentrations of HCMV immune IVIG, the gp34 and gp68 TB40/E deficient mutants elicited a stronger FcγR-ζ activation response than the TB40/E wt (Figure S3A), while the density of opsonizing cell surface antigens was not altered (Figure S3B). The finding that three independent virus mutants lacking Fc binding proteins show congruent phenotypes makes unintended second site mutations as cause for the effect highly unlikely. Nevertheless, revertant viruses were assessed. As expected, both of the revertant viruses exhibited a wt-like phenotype (Figure S3A). In comparison to HCMV HB5, HCMV TB40/E shows a more protracted replication kinetic. Consistently, we observed more efficient IgG-dependent activation of FcγRIIIA-ζ at 96 hpi compared with 72 hpi. Therefore, HCMV TB40/E-based assays were performed 96 h post infection. The HCMV TB40/E results confirmed that both HCMV-encoded FcγRs inhibit the activation of FcγRIIIA and that their reinsertion into the virus genome reestablishes the vFcγR inhibition phenotype.
To test if gp34 and gp68 suffice to impair IgG-dependent activation of FcγRs, two factors of our experimental approach were modified: (i) gp34 and gp68 were expressed outside the context of HCMV infection by recombinant vaccinia viruses, and (ii) instead of polyclonal HCMV IVIG, a well-defined humanized therapeutic monoclonal IgG1 antibody (trastuzumab) was used as an activator of host FcγRs upon binding to its antigen HER2. rVACV expressing HSV gE-infected HER2 antigen positive SKOV-3 tumor cells were opsonized with graded concentrations of trastuzumab recognizing HER2 and compared with wt-VACV as well as mock-infected cells. The opsonized target cells were co-cultured with the panel of FcγR reporter cells (Figure 3A). Opsonized VACV-infected cells exhibited a reduced capacity to trigger FcγRIIIA in comparison to mock cells, most likely due to the protein host shut-off function of VACV. Importantly, trastuzumab-mediated FcγRIIIA triggering was further impaired by rVACV gE, providing proof of principle that ectopically expressed gE suffices to interfere with IgG1-dependent FcγRIII activation. In contrast to FcγRIII, trastuzumab reproducibly failed to induce FcγRII responses (Figure 3A). When trastuzumab-opsonized cells were probed with FcγRI transfectants, the presence of gE did not attenuate but rather enhanced the response (Figure 3A), confirming the unexpected phenotype in the HSV-infected cell setting observed before (Figure 2A). Next, rVACVs were used to express gp34 and gp68 ectopically in HER2 positive SKOV-3 targets which were opsonized with different concentrations of trastuzumab before co-culture with the same panel of responder cells as already described (Figure 3B). Both gp34 as well as gp68 significantly reduced activation of FcγRIII and FcγRI, albeit in this setting gp34 seemed slightly more potent than gp68. In summary, deploying a gain-of-function approach and using a monoclonal human IgG1, the results verified that both HCMV FcγRs are sufficient to prevent the activation of FcγRI and FcγRIII.
Trastuzumab is not capable to activate FcγRIIA (see above, Figure 3A–B). Nevertheless, we wished to assess the effect of ectopically expressed vFcγRs on FcγRIIA activation. Therefore, in a further approach CD20 transfected 293T cells [35] were infected with rVACV expressing gE, gp68 or gp34 before opsonized with rituximab another well-defined humanized therapeutic monoclonal IgG1 antibody (Figure S4A and S4B). All vFcγRs inhibited FcγRIIA activation verifying that ectopic expression of the viral Fcγ binding proteins gE, gp34 and gp68 hinder the activation of the host FcγRIIA in a gain-of-function approach.
Humans respond to HCMV infection with the production of IgG1 which is the immunodominant subclass, followed by IgG3, while HCMV-immune IgG2 and IgG4 is detected only at very low levels if produced at all [36], [37]. In contrast to HSV-1 gE, HCMV gp68 and gp34 bind monomeric IgG of all human subclasses, i.e. IgG1, IgG2, IgG3, and IgG4 [22], whereas gE does not bind IgG3 [38], [39]. To assess whether HCMV gp68 and gp34 can inhibit FcγRIIIA/CD16 activation through immune complexes formed by different IgG isotypes, we took advantage of a panel of rituximab-derived isotypic IgG antibodies. CD20 transfected 293T target cells [35] were infected with VACV wt or rVACV expressing gp68, gp34 or MULT-1 as a negative control and opsonized with anti-hCD20 IgG isotypes including an IgA constant region fused to the variable region of rituximab as an antibody control. CD20 expression revealed very similar levels of antigen expression on the cell surface of VACV target cells (data not shown). While opsonized IgG1 and IgG3 isotypes efficiently activated FcγRIIIA, very little to no activation was observed with IgG2, IgG4 and IgA, confirming previous data [28]. Both gp34 as well as gp68 strongly reduced activation of FcγRIIIA by IgG1 and IgG3 (Figure 4). The data documented the inhibitory potency of both HCMV FcγRs against IgG1 and IgG3-formed immune complexes and confirmed the functional distinction of gp34 and gp68 against HSV gE.
CD16/FcγRIII is an essential IgG receptor for activation of NK cells mediating ADCC responses [40], [41] but also found on human γδ T cells induced by HCMV infection [42]. The data obtained with the FcγRIII-ζ reporter cells strongly suggested that gp34, gp68 as well as HSV-1 gE operate as inhibitors of FcγRIII/CD16+ NK cells since BW:FcγRIII-ζ responses showed an excellent match with CD107a mobilization of primary human NK cells upon CD16/FcγRIII cross-linking [28]. Therefore, we tested the activation of primary human NK cells by fibroblasts infected with HSV-1, HCMV and mutants devoid of viral Fcγ binding proteins, respectively, in the presence of virus-immune opsonizing IgG in a CD107a degranulation assay [43]. The sources of the opsonizing IgG were sera donated by HSV/HCMV-seropositive donors (Figure 5A and Figure 5B, resp.) or Cytotect (Figure 5C). rhIL-2 overnight preactivated NK cells from HSV/HCMV sero-negative donors were enriched by negative selection and analyzed after 4 hours of co-incubation with infected cells opsonized with graded concentrations of immune IgG. HCMV encodes numerous inhibitors of NK cell activation [44], [45]. To focus on IgG-dependent NK cell activation, NK activation was calculated and depicted as percentage of IgG-specific CD107a mobilization (i.e. percentage of CD107a-positive cells obtained with the immune antibody opsonizing target cells minus the percentage of CD107a-positive cells obtained with non-immune antibody treated target cells). A higher ratio of IgG-dependent CD107a positive cells in the case of HSV-1 ΔgE-infected cells compared with wt HSV-1 infected cells was observed (Figure 5A). Likewise, the HB5Δgp68, HB5ΔIRLΔgp34 and HB5ΔIRLΔgp68/Δgp34 HCMV mutants yielded clearly increased IgG-dependent CD107a mobilization as HCMV HB5 (Figure 5B). As observed with BW:FcγR-ζ responder cells, gp34 and gp68 inhibited FcγRIIIA NK activation independently, but no additive effects were noted upon deletion of both vFcγRs. To exclude donor-specific effects, NK cells from six different donors were analyzed in degranulation assays comparing HCMV HB5 wt with HB5ΔIRLΔgp68/Δgp34 -infected targets opsonized with Cytotect as a source of immune IgG and non-immune sera as a negative control. All donors showed a higher percentage of IgG-dependent CD107a positive cells in the case of HB5ΔIRLΔgp68/Δgp34 -infected cells (Figure 5C). Taken together, these data demonstrated that vFcγR gE, gp34 and gp68 on the surface of infected cells mediate inhibition of IgG-dependent NK cell degranulation.
HCMV gp68 was found to bind the Fc CH2-CH3 interface of monomeric IgG at nanomolar affinity [24]. To get insight into the intermolecular interactions underlying the inhibitory function of gp34 and gp68 when blocking antigen-antibody complexes, we tested the occurrence of a physical complex on the surface of cells consisting of the target antigen, bound IgG and each of the HCMV FcγRs. We took advantage of immune complexes (composed of trastuzumab and its antigen HER2) which were shown to be sensitive to the blockade through gp34 and gp68 when activating FcγRIII and FcγRI (Figure 3B). HER2-expressing SKOV-3 cells were infected with rVACV expressing Flag-tagged gp34, gp68 or a control protein, ΔIg1-m138, a non-functional MCMV m138/fcr-1 truncation mutant [46] and opsonized with trastuzumab (‘T’) or with an IgG1 isotype control antibody, palivizumab (‘P’) (see sketch in Figure 6A). VACV-infected cells were thoroughly washed to remove unbound antibodies and subsequently lysed. To exclude Fcγ-mediated binding of vFcγRs through anti-Flag antibodies, vFcγR proteins were immunoprecipitated using α-Flag F(ab)2-coupled agarose beads. The precipitated proteins were separated by SDS-PAGE and analyzed by immunoblotting using an HER2 specific antibody (Figure 6B). An anti-human IgG-specific antibody was used to detect the co-precipitated antibody. An immuno blot confirmed expression and immunoprecipitation of the vFcγRs. Retrieval of palivizumab by gp34 and gp68 was weaker than retrieval of trastuzumab. This difference could be explained by the fact that trastuzumab could be retained by the cells via vFcγRs and via HER2, while palivizumab could only be retained by vFcγRs. Subsequently, gp34 and gp68 retrieved trastuzumab antibodies bound to HER2 during lysis and precipitation. Nevertheless, co-precipitation of human HER2 molecules occurred only in the presence of specific antibody trastuzumab and the HCMV FcγRs but not in the negative control ΔIg1-m138 (Figure 6B, lanes 5, 7 and 3, respectively). Binding of human IgG antibodies, trastuzumab and the isotype control antibody palivizumab, was observed to both vFcγRs (Figure 6B, lanes 5, 6, 7 and 8). No binding of trastuzumab or palivizumab was detectable to the ΔIg1-m138- Flag protein (Figure 6B, lanes 3 and 4). Taken together, the ability of cell surface resident vFcγRs gp34 and gp68 to bind to IgG immune complexes was demonstrated. This finding is compatible with the model of “antibody bipolar bridging” described for the HSV-1 FcγR gE [47]–[49]. According to this concept, epitope-bound IgG on the surface of a virus-infected cell is simultaneously sequestered by the gE:gI complex via its Fc domain, thus preventing the activation of immune effector molecules via host FcγRs.
Soluble truncation versions of HSV-1 gE and HCMV gp68 were instrumental to unravel structural requirements and stoichiometry of herpesviral FcγRs forming complexes with Fcγ [24], [49], [50]. To test whether membrane insertion of gp34 and gp68 is required to interfere with the activation of host FcγRs, recombinant C-terminally truncated ectodomains of HCMV FcγRs were generated and purified from supernatants of transfected human 293 cells. To evaluate if soluble gp34 (sgp34) and soluble gp68 (sgp68) are sufficient to block triggering of host Fcγ receptors, HER2-positive cells were opsonized with trastuzumab and different amounts of recombinant soluble vFcγRs were concomitantly added to BW:FcγRIIIA-ζ cells (Figure 7A) or BW:FcγRI-ζ cells (Figure 7B). Soluble ICOS ligand (sICOSL) served as a negative control protein. Both sgp34 and sgp68 were able to inhibit activation of the reporter cells expressing FcγRIIIA, although clear differences in concentration dependency between soluble vFcγRs were observed. In contrast to full-length HCMV FcγRs, sgp34 was more potent against FcγRIIIA compared to sgp68, since trace amounts of sgp34 hardly detectable in western blot (Figure S5) were sufficient for significant inhibition. In the case of FcγRI, sgp68 was not significantly reducing its activation by trastuzumab, suggesting the specific requirement of the gp68 transmembrane domain for effective inhibition of FcγRI/CD64.
To extend the data to HCMV infection and to test BW:FcγRIIA-ζ cells, a polyclonal antibody preparation (IVIG, Cytotect) was used to opsonize MRC-5 fibroblasts infected with the HCMV HB5ΔIRLΔgp68/Δgp34 mutant lacking both gp68 and gp34. Using BW:FcγRIIIA-ζ reporter cells, both of the soluble HCMV vFcγRs prevented activation when compared with treatment of cells with the sICOSL control (Figure 7C). Moreover, this approach allowed to test activation of BW:FcγRIIA-ζ cells which did not respond to trastuzumab (see Figure 3). As depicted in Figure 7D, FcγRII responses were also sensitive to sgp34 and, to a lesser extent, sgp68. These results provide proof of principle that soluble HCMV FcγRs retain FcγR blocking abilities, and that sgp34 is particularly efficient.
In an attempt to extend the previously made observation of sgp34 and sgp68-mediated inhibition of IgG-triggered FcγRIII/CD16+ BW5147 responder cells to primary human NK cells, purified IVIG Cytotect was coated directly to a plate serving as a source of ‘immune-complexed’ IgG. After blocking with D-MEM 10% FCS (vol/vol), soluble proteins, IL-2 pre-activated primary NK cells and α-CD107a-PECy5 antibody were added. Soluble ICOS ligand (sICOSL) served as a negative control protein. In the absence of coated IgG, only 10% of NK cells responded with CD107a mobilization, while in the presence of coated IgG more than 70% of NK cells translocated CD107a to the cell surface (Figure 8A), confirming the IgG-dependency of the elicited NK cell response. Importantly, both sgp34 and sgp68 were able to inhibit IgG-dependent NK degranulation, although a clear difference in the concentration dependency between soluble vFcγRs was observed. Consistent with the results received with BW:FcγRIIIA-ζ reporter cells (Figure 7C), sgp34 was more potent against IgG-dependent NK activation compared to sgp68, since trace amounts of sgp34 (Figure 8B) were sufficient to significantly interfere with degranulation of NK cells. Using a gain of function approach the data confirmed the inhibitory capacity of gp34 and gp68 to attenuate IgG-mediated NK cell activation.
Here we identified various members of the human FcγR family, i.e. FcγRI/CD64, FcγRII/CD32A and FcγRIII/CD16A, to be targeted by the HCMV FcγRs gp34 and gp68 which act as antagonists of ligand induced FcγR responses. This ability enables HCMV to evade from IgG effector responses and should have direct proviral effects in scenarios of post-acute and recurrent infection when glycoprotein-specific IgG antibodies are synthesized [51]. Several independent experimental approaches support this conclusion: i) HCMV HB5-derived mutants with deletions of the gp34 and gp68 coding genes, TRL11/IRL11 and UL118-119, respectively, showed significantly increased activation of host FcγRs upon opsonization of infected cells with polyclonal HCMV immune IgG using different types of responder cells (i.e. BW5147 transfectants expressing FcγR-ζ chain chimeras and primary human NK cells); ii) this HCMV phenotype was reproduced with targeted RL11 and UL118-119 mutants of the AD169varL and TB40/E strain and iii) fully reversed by retransfer of the responsible genes into the RL11- and UL118-119-deficient TB40/E genomes; iv) a gain-of-function approach based on ectopic VACV-based expression of gp34 and gp68 which allowed functional analysis of well characterized therapeutic human monoclonal antibodies and different IgG isotypes thereof and v) functional testing of recombinant soluble ectodomains of both HCMV inhibitors using BW5147:FcγR reporter cells as well as primary human NK cells. Importantly, the inhibitory effect of gp34 and gp68 was demonstrated at physiological concentrations of polyclonal HCMV immune IgG, i.e. within an extended concentration range of human serum and ten to fifty fold lower.
The quite complex and overlapping expression patterns of host Fc-IgG receptors on a multitude of diverse human immune cell (sub-)populations [17] have obstructed a systematic functional analysis of individual host FcγRs which differ with respect to molecular and functional features including the composition of their ectodomain, intracellular signaling and IgG subclass preferences. Only a recently developed methodologically broadly tested and proven reporter cell assay [28] enabled a comprehensive and quantitative functional assessment of potential viral antagonists and their relative effectiveness against distinct host FcγRs. The new methodology was complemented and validated by immunological as well as biochemical assays, i.e. the use of primary human NK cells as natural responder cells and immunoprecipitation studies, the results of which accord very well with the findings made with the BW5147:FcγR-ζ test system. Last but not least, the well-known viral FcγR inhibitor, HSV gE, was included as an internal control. Previous publications reported inhibition of ADCC, virion neutralization and complement mediated virolysis by HSV gE [47], [52], [53]. Since the blockade of ADCC by gE was not yet attributed to a specific host FcγR [30], [47], [54], analysis of the relative impact of HSV gE on distinct host FcγRs represents a novel aspect of our study. On the basis of the test performance of HSV gE, both gp34 and gp68 demonstrated an at least equivalent if not superior efficacy to block FcγRIII and FcγRIIA mediated responses. Surprisingly and contrasting with both HCMV vFcγRs, gE enhanced rather than attenuated FcγRI activation. This observation warrants further studies how HSV-infected cells affect FcγRI bearing immune cells like monocytes, macrophages, DCs and neutrophils in the presence of HSV-immune IgG.
The inhibition mechanism of IgG-mediated effector functions by gE has been suggested to involve ‘antibody bipolar bridging’ [47]. Pioneering studies of the Bjorkman laboratory demonstrated that the architecture of the gE/gI-IgG complex allows antibody bipolar bridging [49], whereby the gE binding site for Fcγ does not directly overlap with the binding sites to the host FcγRs or the C1q component of complement, which both bind to the upper hinge region of IgG or near the CH2 domain [55], [56]. Therefore, the structure of the gE/gI-Fc complex does not directly explain how gE binding to the Fc region of IgG leads to evasion from FcγR- and complement-mediated immune responses. Our biochemical data reveal formation of ternary heterocomplexes composed of antigen, IgG and gp34/gp68, i.e. a molecular configuration compatible with the minimal requirements of the concept of ‘bipolar bridging’ [47], [49]. The observation that soluble gp34 and gp68 remain potent inhibitors of FcγR activation demonstrates that the functional inactivation of the host FcγR on the responder cell does not require fixation of the opsonized IgG to the plasma membrane as insinuated by the classical concept of bipolar bridging. Although there is no crystal structure available for any HCMV vFcγR, detailed biochemical evidence was generated of how HCMV FcγRs recognize Fcγ, particularly for gp68. The gp68 Fcγ binding site was mapped to the CH2-CH3 interface region of Fcγ [24] which is remote from the FcγRII/III contact site, that involves the hinge between the Fcγ and Fab domains including the upper portion of the CH2 domain [55], [57]. Specifically, gp68 binding to Fcγ is affected by mutations at the CH2-CH3 domain interface of IgG, mapping its binding site to determinants situated nearby but not identical with those that are recognized by gE [24]. We observed robust functional differences between HSV-1 gE and HCMV gp68 in their manipulation of FcγRI, further substantiating the mechanistic differences between these viral inhibitors regarding their interaction mode with Fcγ. While the binding of IgG by gE must induce conformational changes of the antibody that result in an enhancement of FcγRI activation, gp68 binding to IgG induces the opposite effect. Since both gE and gp68 had concordant inhibitory effects on FcγRIII, our data further imply that FcγRI and FcγRIII must bind IgG in a differential fashion. Importantly, the length and flexibility of the hinge region varies considerably among the IgG subclasses, and IgG3 differs from the other subclasses by its unique extended hinge region which is approx. four times as long as the IgG1 hinge, leading to the most hinge-mediated flexibility among human IgG subclasses [58]. Notably, we demonstrate efficient FcγRIII blockade by IgG3-shaped immune complexes through gp68 and gp34 (Figure 4) which is not possible by HSV gE [36], [37]. Thus, by analogy with HSV gE, HCMV gp34 and gp68 represent promising and unique tools to further probe into the diverse structural requirements of FcγRI/II/III activation by immune complexes constituted by all IgG subclasses.
The presence of independent but redundant immunoevasins jointly targeting one particular immune control mechanism is a typical feature of cytomegaloviruses highlighting the antiviral power of the targeted immune component [6], [59]–[62]. At first glance, the HCMV FcγR antagonists gp34 and gp68 exhibit a surprisingly similar effect on the whole range of activating host FcγRs, despite their simultaneous synthesis during the early and late phase of HCMV replication [22]. As a consequence, removal of both inhibitors from the surface of HCMV infected cells did not reveal additive or even synergistic effects compatible with the notion that the two factors do not act in an obvious cooperative manner. This finding cannot be attributed to differences in the density of plasma membrane resident HCMV antigens between the HCMV gene deletion mutants compared in our study (see Figure S2 and S3). However, both of the antagonists could themselves represent antigens that are recognized by the F(ab) part of immune IgG, which could either directly activate host FcγRs or block Fcγ-mediated bridging of opsonizing IgG (as a counter defense of humoral immunity against vFcγRs) and thus indirectly enhance host FcγR triggering. Moreover, both of the HCMV FcγRs may fulfill further proviral but Fcγ-independent functions which exert separate pressures to adapt. This is exemplified by the MCMV m138/fcr-1 molecule which down-regulates the NKG2D ligands MULT-1, H60, RAE-ε [46], [63] as well as the B7-1 molecule CD80 [64] beyond its Fcγ binding activity. In addition, besides gp34 and gp68 additional HCMV FcγRs become expressed on infected cells (Mercé-Maldonado and Hengel, in preparation), one of which is encoded by RL13 [65]. Thus gp34 and gp68 may be part of a much more complex network of co-expressed HCMV FcγRs jointly combating their host opponents, and removal of one player could confound their interplay and nested hierarchies. Next, drastic quantitative and qualitative differences in the potency of gp34 vs. gp68 became apparent when soluble molecules were compared. Thus it is tempting to speculate that shedding of vFcγRs may be part of the molecular blueprint of particular vFcγRs.
While there is extensive knowledge on antigens and processed epitopes which rule anti-HCMV T cell responses [66], viral antigens that are targets of ADCC dependent cellular immunity remain poorly defined. Our finding that late but not early antigens dominate the FcγRIII/CD16 activating IgG response (Figure 1B) appears a particular characteristic of HCMV when compared with HSV and could point to structural glycoproteins known to become exposed on the cell surface as the HCMV replication cycle progresses, e.g. gB [67], gH [68] and UL128 [69]. Guided by human antibodies with defined specificity, our BW5147-based FcγR-ζ assay system could be instrumental to identify the relevant HCMV antigens and epitopes. In many tissues and organ compartments, including blood, HCMV is spreading intracellularly (e.g. via infected endothelial cells and leukocytes) rather than as free virions [70]. Therefore, ADCC-inducing IgG is plausible to represent a primary effective component of humoral immunity, which becomes only secondary attenuated by gp34 and gp68. Both immunoevasins could thus contribute to the relatively poor therapeutic efficacy of HCMV-immune IgG observed in a variety of clinical settings [12]–[14]. Thus a better knowledge of the optimal HCMV IgG epitopes on the one hand, and an understanding of the action of viral FcγR antagonists on the other hand, could provide us with a basis for the targeted induction or even rational synthetic design of IgG molecules that allow an improved immunotherapy of HCMV diseases.
Human MRC-5 lung fibroblasts (ATCC CCL-171), African green monkey CV-1 (ATCC CCL-70), HEK293 (ATCC CRL-1573) and CD20 transfected 293T cells (a kind gift from Irvin S. Y. Chen, University of California) [35] cells were maintained in culture with D-MEM (Gibco), 10% (vol/vol) heat-inactivated FCS, Penicillin (100 U/ml), Streptomycin (100 µg/ml) and Glutamine (2 mM). Mouse BW5147 thymoma cells (obtained from ATCC, TIB-47), transfectants thereof [28] and SKOV-3 cells were maintained in RPMI 1640 medium with 10% (vol/vol) FBS, Penicillin, Streptomycin, Glutamine, and Sodium Pyruvate (1 mM). The following viruses were used: the bacterial artificial chromosome plasmid (BACmid)-derived human cytomegalovirus (HCMV) strain HB5 [71], the HB5-derived mutants lacking UL118-120 (Δgp68) [22] or lacking IRL/RL11/UL118 (Δgp68/Δgp34) [31], TB40/E BAC [34], herpes simplex virus 1 strain F (HSV-1), HSV-1 ΔgE and revertant thereof [72], VACV wt Western Reserve, recombinant VACVs expressing the vFcγRs [22], [73], a rVACV expressing MULT-1 [74] and a rVACV expressing a non-IgG binding truncation mutant of MCMV m138 (m138ΔIg1Flag) [46]. The HB5-derived deletion mutant TRL10-14/IRL (Δgp34) was constructed following a previously published procedure [75]. The BAC plasmid pHB5-ΔIRL-frt (kindly provided by Eva Borst, Hannover, Germany), lacking the nucleotides 179150–192329 of the AD169-derived BAC pHB5 [71] was used to generate BAC plasmid pΔIRL/ΔTRL11 (lacking the viral FcγR gp34; Δgp34). For its construction, the primer pair AZ-TRL11-tet1 5′ACAGACGACGAAGAGGACGAGGACGACAACGTCTGATAAGGAAGGCGAGAACGTGTTTTGTCCAGTGAATTCGAGCTCGGTAC-3′ and AZ-TRL11-tet2 5′TGTATACGCCGTATGCCTGTACGTGAGATGGTGAGGTCTTCGGCAGGCGACACGCATCTTGACCATGATTACGCCAAGCTCC-3′ was used to amplify the tetracycline resistant gene (TetR) from pCP16 for insertion into pΔIRL-frt. Recombinant TB40 HCMV were generated according to a previously published procedure [75] using BAC plasmid TB40/BAC4 [34]. For construction of the TB40 Δgp34 mutant, a PCR fragment was generated from the contiguous primers PL-delRL11-1 (5′-TCCCCGTTGATCGAACCGACGGGCACAGACGACGAAGAGGACGAGGACGACGACGTCTGACCAGTGAATTCGAGCTCGGTAC-3′) and PL-delRL11-2 (5′-CATGCATGTTATTTGCGTGTACGATGACTTGTTTCGCCGTCGATGTTGTGTACGCATCTTTTACTCCAAATCCCCGTCCACCCACCATGATTACGCCAAGCTCC-3′) using the plasmid pSLFRTKn [22] as template DNA. For construction of the TB40 Δgp68 mutant a PCR fragment was generated from the contiguous primers PL-delUL119-1 (5′-GGTCTCCTGCGGCCTGAGTCCCGAGATAAGCAGCTCTTGAGCAGTAGCGTTGTAGGAGAGCCAGTGAATTCGAGCTCGGTAC-3′) and PL-delUL119-2 (5′-AGGTGACGCGACCTCCTGCCACATATAGCTCGTCCACACGCCGTCTCGTCACACGGCAACGACCATGATTACGCCAAGCTCC-3′) using the plasmid pSLFRTKn [22] as template DNA. The PCR fragments containing a kanamycin resistance gene were inserted into TB40/BAC4 by homologous recombination in E. coli. The Knr was excised from both BACs by flp-mediated recombination [75] generating the HCMV BACs TB40-Δgp34 and TB40-Δgp68. The revertant viruses were constructed from PCR-fragments containing the respective gene and flanking homologies subcloned into the BamHI site of the shuttle vector pST76KSR. The following primer pairs were used: AZ-RL11rev-1: GtGGATCCGAGTGTTGAAGGGTAACGTGAGGGA and AZ-RL11rev-2 GCTCTAGAGCATGCAGATCTGTCTTGTAGCACGATGTGGTGGT for the gp34 revertant and AZ-UL118rev-1: GCTCTAGAGCATGCAGATCTACCACTGCTTGAAGTAGGGCACC and AZ-UL118rev-2: GtGGATCCGGTGGTATGAGCCTGAAGTGAGCAT for the gp68 revertant. The viral FcR genes from plasmids pST76KSR-RL11rev and pST76-KSR-UL119rev were inserted into TB40-Δgp34 and TB40-Δgp68, respectively, by two-step homologous recombination [76] resulting in the BACs TB40-gp34Rev and TB40-gp68Rev. Correct mutagenesis of all recombinant HCMV-BACs was confirmed by restriction analysis and sequencing of the respective genome region. Recombinant viruses TB40-Δgp34, TB40-Δgp68, TB40-gp34Rev and TB40-gp68Rev were reconstituted by transfection of MRC-5 using Superfect reagent (Qiagen, Germany) as described by Borst et al. [71]. The AD169varL deletion mutants were generated according to a previously published procedure [75] using the BACmid-cloned AD169varL genome pAD169 [33] as parental BACmid. Briefly, a PCR fragment was generated using primers listed in Table S2. The PCR fragment containing a kanamycin resistance gene was inserted into the parental BACmid by homologous recombination in E. coli. For the construction of mutants harboring deletions of two non-adjacent genes, the kanamycin cassette was removed by flp-mediated recombination before introducing the second deletion. Recombinant mutant viruses were reconstituted from BACmid DNA by Superfect (Qiagen, Germany) transfection into HCMV-permissive MRC-5 fibroblasts.
Infection of cells with HCMV and HSV was enhanced by centrifugation at 800 g for 30 min. If not stated otherwise, the cells were infected with 2–3 PFU/cell.
A clinically used IVIG preparation [11] Cytotect [32], [77] (batch no. A158024 and B797053, Biotest Pharma GmbH, Germany) containing ELISA reactive IgG specific for HCMV and HSV was used. For the FACS analysis of HCMV and HSV surface antigens, F(ab)2 fragments were generated from Cytotect with the Pierce F(ab)2 Micro Preparation Kit (Thermo Fisher Scientific Inc., Rockland, IL, USA) according to the manufacturer's instructions and controlled by Western Blot (data not shown). For the experiments with HCMV and HSV, a pool of two ELISA seronegative donors were used as a negative control. Trastuzumab was purchased from Genentech, Inc., USA and palivizumab from MedImmune, USA. The humanized anti-CD20 IgG1, IgG2, IgG3, IgG4 isotypes and IgA were purchased from InvivoGen, Toulouse, France. For the CD107a NK degranulation assay, an HCMV- and HSV-seropositive donor and a negative serum donor as sources of immune and non-immune IgG, respectively, were used. For proofing that the assay was antibody-antigen specific (Figure S1), a pool of 6 HCMV- and HSV-seropositive donors and a pool of 2 negative serum donors as sources of immune and non-immune IgG, respectively, were used. For serum preparation, blood was drawn from healthy volunteers after written informed consent.
The experiments were approved by the Ethics Committee of the University Hospital Düsseldorf (no. 3410) in accordance with the Declaration of Helsinki. For serum and NK cells preparation, blood was drawn from healthy volunteers after written informed consent.
This assay was described elsewhere [28]. Briefly, in a standard assay, target cells were incubated with dilutions of human sera, IVIG, the anti-hCD20 IgG isotype collection or trastuzumab in D-MEM with 10% (vol/vol) FCS for 30 min at 37°C. Cells were washed before co-cultivation with BW:FcγR-ζ transfectants (ratio E∶T 20∶1) for 16 h at 37°C in a 5% CO2 atmosphere. Then mIL-2 secreted was measured by ELISA. When applied to VACV infected cells, a previous step of UV-inactivation at 4000 Jules/m2 and 2 steps of washing with PBS were performed before opsonization with Abs. When applied to the inhibition through soluble vFcγR, soluble proteins were added concomitantly with BW:FcγR-ζ transfectants. For herpesviral late antigens IgG-dependent activation of BW:FcγRIIIA cells, late phase gene expression was blocked by the use of phosphonoacetic acid (PAA) (250 µg/ml), which blocks viral genome replication and late gene expression. Afterwards, a co-cultivation assay was performed as described above.
The N-terminus of gp34 was amplified by PCR using the following primers 5′-GCTTAGGGATCCATGCAGACCTACAGCACCCC-3′) [22] and 5′-TCTCACTAGTGGACCACTGGCGTTTTAAATC-3′. Cloning of the N-terminus of gp68 was previously described [24]. The N-terminus of ICOSL (ICOS ligand) was amplified by PCR using the following primers 5′-GAGGTAAGATCTCGCACCATGCGGCTGGGC-3′ and 5′-CTCTCACTAGTCGTGGCCGCGTTTTTC-3′. Sequencing of the coding sequences showed an amino acid exchange in ICOSL from V128 to I128 but with no detectable functional difference. Each PCR product was cloned in pGene/V5-His B vector (Invitrogen, USA) in frame with the V5-His epitope tag using the restriction sites BglII and SpeI (italics in primer sequences) and then subcloned in pIRES-EGFP vector (Clontech, USA). For enhanced expression of gp34V5-His and gp68V5-His, γ-Globin cloned from pSG5 vector (Stratagene, USA) was inserted into pIRES-EGFP between the CMV-IE promoter and the coding sequences. The plasmids were transfected in HEK293 cells using Superfect (Qiagen, Germany) and transfected cells were selected with 1.25 mg/ml of Geneticin (Sigma-Aldrich, Germany). After 4–5 days, supernatants were collected, volume reduced, diluted with PBS (1∶3), adjusted to a 10 mM Imidazole concentration and passed over a His-Trap FF crude column (GE Healthcare, USA). Proteins were eluted in Imidazole/Phosphate buffer (250 mM Imidazole, 20 mM sodium phosphate, 500 mM NaCl) and then dialyzed to PBS. Comparable protein amounts were adjusted based in Western blot analysis using α-V5 antibody (Invitrogen, USA).
MRC-5 cells were infected with 2 PFU/cell of wt HCMV and HCMV vFcγR mutants during 72 h and with HSV-1 wt, ΔgE and ΔgE-revertant during 24 h. Cells were resuspended in PBS containing 2 mM EDTA, washed twice in PBS supplemented with 3% (vol/vol) FCS and mock stained or stained with human Fcγ fragment-FITC (Rockland Immunochemicals, USA). 1×104 living cells were obtained in FACSCanto II using the FACS Diva software and analyzed with FLowJo (Tree Star Inc, USA).
MRC-5 cells were infected with 1 PFU/cell of wt HCMV and HCMV ΔvFcγR mutants for 72 h and with 10 PFU/cell of HSV-1 wt and ΔgE for 24 h. Cells were resuspended in PBS containing 2 mM EDTA, washed twice in PBS supplemented with 3% (vol/vol) FCS. HCMV infected cells were stained with the F(ab)2 preparation of Cytotect, goat anti-human-F(ab)2-Biotin, and Streptavidin-PE (AdB Serotec, UK) or Fcγ fragment-FITC (Rockland Immunochemicals, USA). The comparability of infection of the different HCMV ΔvFcγR mutants was controlled by intracellular staining of CMV nuclear antigens with CCH2 and DDG9 antibodies (Dako, Denmark) and goat anti-mouse-APC (BD Pharmingen, USA) after fixation with 1,5% PFA and permeabilization with PBS supplemented with 3% (vol/vol) FCS and 0,05% (vol/vol) Saponin. HSV infected cells were stained with the F(ab)2 preparation of Cytotect, goat anti-human-F(ab)2-Biotin (AdB Serotec, UK) or Fcγ fragment-Biotin (Rockland Immunochemicals, USA) and Streptavidin-APC (Jackson Immunoresearch, USA). After DAPI staining, 1–2×104 living cells were obtained in a FACSCanto II using the FACS Diva software and analyzed with FlowJo (Tree Star Inc, USA).
PBMCs were prepared from EDTA-blood of healthy donors using Lymphoprep (Axis-Shield, Norway) differential centrifugation. PBMCs were incubated during 3 h at 37°C to allow adherence of unwanted cells. Suspension cells were collected and resuspended in media containing 100 IU/ml of human rIL-2 (PromoKine, Germany) and incubated overnight at 37°C. Cells were resuspended and further processed to obtained polyclonal NK cells using a MACS negative selection NK cell isolation kit (Miltenyi Biotec, Germany). NK cell purity was tested in FACS and was usually above 96% (data not shown). For measuring degranulation by co-cultivation of immune IgG and a viral target, HCMV or HSV infected fibroblasts were opsonized with a serum of a healthy donor positive for HCMV and HSV or with IVIG. As a control, a healthy seronegative donor was also analyzed. Opsonization was done at 37°C for 30 min at 5% CO2. Two steps of washing with D-MEM 10% (vol/vol) FCS followed to remove unbound IgG. 1×105 polyclonal NK cells (E∶T ratio of 10∶1) were added in each well and the CD107a assay was performed as elsewhere described [43]. Briefly, polyclonal human NK cells were incubated 4 h at 37°C in the presence of 6 µg/ml Golgi Stop (Monensin, BD Pharmingen, Belgium), 10 µg/ml Golgi Plug (BrefeldinA, BD Pharmingen, Belgium), and CD107-PeCy5 mAb (BD Pharmigen, Belgium). NK cells were collected, washed twice in ice cold PBS containing 2 mM EDTA and stained for extracellular markers (CD56, CD3). 1×104 cells were counted and analyzed.
SKOV-3 cells were infected with 2 PFU/cell VACV for 14 h. Infected cells were incubated with 1 µg/ml trastuzumab or palivizumab for 30 min at 4°C, and non-bound antibody was removed by washing cells with PBS. Lysis buffer (200 mM NaCl, 10 mM MgCl2, 10 mM KCl, 20 mM HEPES, 0.5% (vol/vol) NP-40, 0.1M EDTA, 10% (vol/vol) glycerol, 0.1 mM sodium orthovanadate, 0.1 mM PMSF, 0.5 µM Pepstatin, 1 mM dithiothreitol, pH 7.4) was given to cells and, after removal of cell nuclei by centrifugation, lysates were incubated with agarose immobilized anti-FLAG antibody (Bethyl Laboratories, Inc. USA) during 2 h at 4°C. Previously, a sample of each lysate was taken for subsequent western blot expression analysis. Lysis buffer was used to wash the agarose pellet and proteins were eluted with Laemmli sample buffer. Proteins were separated by sodium-dodecyl-sulfate (SDS)-8% polyacrylamide gel electrophoresis (PAGE) and transferred to nitrocellulose filters. Western Blot was performed with anti-ErbB2-specific rabbit mAb V2W (Abcam Inc, USA), anti-Flag-specific mouse mAb M2 (Sigma-Aldrich, USA), anti-human-peroxidase (Sigma-Aldrich, USA), anti-rabbit-peroxidase (Sigma-Aldrich, USA) and anti-mouse-peroxidase (Dianova, Germany). Proteins were visualized using ECL chemiluminescence system (GE-Healthcare, Germany).
HCMV UL119-118: P16739
HCMV RL11: Q6SX56
HSV gE: Q703F0
Human FcγRIIIA/CD16: P08637
Human FcγRIIA/CD32A: P12318
Human FcγRI/CD64: P12314
Mouse TCR zeta chain: P24161
Human ICOSL: O75144
Human cytomegalovirus (strain AD169): 10360
Herpes simplex virus (type 1/strain F): 10304
Vaccinia virus Western Reserve: 696871
|
10.1371/journal.ppat.1004134 | Suppressor of Cytokine Signaling 4 (SOCS4) Protects against Severe Cytokine Storm and Enhances Viral Clearance during Influenza Infection | Suppressor of cytokine signaling (SOCS) proteins are key regulators of innate and adaptive immunity. There is no described biological role for SOCS4, despite broad expression in the hematopoietic system. We demonstrate that mice lacking functional SOCS4 protein rapidly succumb to infection with a pathogenic H1N1 influenza virus (PR8) and are hypersusceptible to infection with the less virulent H3N2 (X31) strain. In SOCS4-deficient animals, this led to substantially greater weight loss, dysregulated pro-inflammatory cytokine and chemokine production in the lungs and delayed viral clearance. This was associated with impaired trafficking of influenza-specific CD8 T cells to the site of infection and linked to defects in T cell receptor activation. These results demonstrate that SOCS4 is a critical regulator of anti-viral immunity.
| The suppressor of cytokine signaling proteins are key regulators of immunity. As yet there is no described biological role for SOCS4, despite its broad expression in cells of the immune system. Given the important role of other SOCS proteins in controlling the immune response, we have generated SOCS4-mutant mice and used a mouse influenza infection model to investigate the biological function of SOCS4. We demonstrate that mice lacking SOCS4 rapidly succumb to infection with a pathogenic H1N1 influenza virus and are hypersusceptible to infection with the less virulent H3N2 strain. This is the first demonstration of a functional phenotype in SOCS4-deficient mice. Our study reveals that in SOCS4-deficient animals, there is a dysregulated pro-inflammatory cytokine and chemokine production in the lungs and delayed viral clearance. This is associated with impaired trafficking of virus-specific CD8 T cells to the site of infection and linked to defects in T cell receptor activation. These results demonstrate that SOCS4 is a critical regulator of anti-viral immunity. Understanding the regulation of the inflammatory response to influenza is particularly relevant given the current climate concerning pandemic influenza outbreaks.
| Influenza is a highly infectious, acute respiratory disease that causes profound morbidity and mortality. Annual seasonal influenza epidemics result in ∼500,000 deaths worldwide and substantial losses to global economies [1]. The development of a “cytokine storm” coupled with damage to pulmonary epithelium has been consistently observed in severe cases of influenza infection in humans. The mechanisms underlying this pathology, and an understanding of why some individuals respond excessively to virus, to an extent that results in hospitalisation or death, remains relatively unexplored.
The initial innate immune response to viral infection is characterized by an influx of neutrophils, monocytes and macrophages into the lung parenchyma and alveolar spaces, with the elevated levels of inflammatory cytokines/chemokines correlating strongly with pathogenesis and viral load [2]. However, exaggerated cytokine and chemokine responses have been observed in the lungs of critically ill patients in the absence of high viral load [3], suggesting that inflammation-driven pathology can occur independently of viral load.
The adaptive response subsequently results in the generation of strain-specific B cells and cross-strain protective CD4 and CD8 T cells. Influenza-specific CD8 T cells are largely responsible for host immunity to primary influenza infection and operate to promote the efficient elimination of virus, and host recovery, via the production of pro-inflammatory cytokines and direct killing of virus-infected cells [4]. Acquisition of these effector functions occurs in the draining lymph nodes where upon T cell receptor (TCR) recognition of the influenza-specific peptide:MHC complex, CD8 T cells become activated and then migrate to the infected lungs.
Cytokine binding to the cognate receptor complexes triggers an intracellular signaling cascade, most often coupled to the JAK-STAT pathway, which orchestrates an intricate series of transcriptional changes leading to the appropriate cellular response. The suppressors of cytokine signaling (SOCS) proteins are key negative regulators of the JAK/STAT pathway and are thus involved in the fine-tuning of the cytokine networks responsible for an adequate and efficient innate and adaptive immune response [5]. The family is composed of eight members, SOCS1 to 7 and cytokine-inducible Src-homology 2 protein (CIS) [6]. All proteins share a central SH2 domain and carboxyl-terminal SOCS box, but differ in their amino termini. SOCS4 to 7 are particularly distinguished by a long N-terminal region, which bears little homology to other SOCS proteins [7]. The SOCS box interacts with elongins B and C, and together with Rbx2 and Cullin-5 forms an E3 ubiquitin ligase [8]. The SOCS proteins therefore act as adaptors to target substrates bound to their SH2 or N-terminal regions for ubiquitination and proteasomal degradation [9]. In addition, SOCS1 and 3 can bind directly to JAK via their kinase inhibitory region (KIR) and SH2 domains, inhibiting JAK phosphorylation of substrates and downstream signaling [10], [11]. CIS and SOCS2 are thought to bind to phosphotyrosine residues within the receptor cytoplasmic domains to block recruitment of other signalling intermediates [12], [13], [14].
The generation of knockout mice has proven a powerful tool in defining the physiological role of the SOCS proteins. SOCS1 for instance, was revealed as a critical regulator of IFNγ signaling and γc-cytokine-dependent T cell homeostasis [15], SOCS2 as a regulator of growth hormone signaling [16] and conditional deletion of the Socs3 gene has identified a role for SOCS3 in regulating IL-6 and G-CSF signaling [17], [18]. Although a wealth of information is available on the role of CIS and SOCS1-3, there is much less data regarding the targets and pathways regulated by the remaining family members, including SOCS4.
In vitro studies have suggested that SOCS4 is involved in regulating epidermal growth factor (EGF) signaling [19], and indeed the SOCS4-SH2 domain binds with high affinity to a phosphopeptide corresponding to an EGF receptor (EGFR) autophosphorylation site (Tyr1092; 0.5 µM) [20]. This later study provides some information as to the binding preferences of the SOCS4-SH2 domain [20], however the relevance of the interaction with the EGFR remains to be elucidated. Other studies have suggested that SOCS4 may be regulated by parasitic infection [21], is linked to better outcomes in cancer patients [22], or may regulate pre-granulosa cells during folliculogenesis [23]. Thus far, none of these studies present a compelling case for a physiological role for SOCS4.
In the current study, we have generated the first loss of function allele of murine Socs4, a point mutation identified in a library of ENU-mutagenised mice [24]. At the protein level, this mutation is predicted to cause the substitution of a stop codon for arginine 108, resulting in deletion of the remaining 90 residues of the N-terminal domain, as well as the entire SH2 domain and SOCS box. Given that SOCS proteins play a critical role in regulating immune responses, and expression of SOCS4 in lymphocytes, we investigated the role of SOCS4 in a defined viral infection model in which T cells regulate pathogen clearance. We found that Socs4R108X/R108X mice were highly susceptible to infection with influenza A, showing increased morbidity and a delay in viral clearance comparable to that observed for mice lacking CD8 T cells [25] or IL-18 [26]. The increased lethality appeared to result from an elevation in key pro-inflammatory cytokines and chemokines, such as IL-6, IFNγ and MCP-1; whilst in the latter phase of the infection, Socs4R108X/R108X mice displayed impaired trafficking of virus-specific CD8 T cells to the lungs. The defect in trafficking appeared to be qualitatively linked to the activation status of these cytotoxic T lymphocytes and reveals a novel role for SOCS4 as a positive regulator of TCR signaling.
Socs4R108X/R108X mice were viable, fertile and showed no overt phenotype under steady-state conditions, including normal thymocyte development and composition of peripheral immune cells (Fig. S1). To investigate the role of SOCS4 in the response to viral challenge, homozygous Socs4R108X/R108X mice and age-matched littermates or Balb/c controls were inoculated intranasally (i.n.) with 20 pfu of the virulent H1N1 influenza strain A/Puerto Rico/8/34 (PR8) and monitored for weight loss. In accordance with ethical guidelines, mice were considered moribund upon losing greater than 20% of their initial body weight, and removed from the study. At this relatively low challenge dose, Socs4R108X/R108X mice exhibited significantly enhanced disease progression and mortality compared to wild-type controls (Fig. 1A and B) and this correlated with an increased viral load in the lungs (Fig. 1C).
To dissect the defective response in more detail, Socs4R108X/R108X mice and wild-type controls were subsequently infected i.n. with 103 pfu of the less virulent H3N2 A/Hong Kong x31 (X31) influenza virus. Similarly to infection with PR8, Socs4R108X/R108X mice showed enhanced susceptibility to X31 infection with significantly greater weight loss (P = 0.005; Fig. 1D) and higher lung viral loads (Fig. 1E) when compared to controls. Initially (day 1 and 2 post-infection) there were no differences in viral titers between the groups, indicating that viral uptake and replication are not affected by SOCS4 deficiency. However, Socs4R108X/R108X mice displayed higher viral titers (0.5 to 1.8 log higher) on day 3 and day 5 post-infection and this difference was greatly exacerbated by day 6 (2 log difference), when the majority of control mice had cleared the infection. Virus was undetectable in the lungs of wild-type controls on day 7, however some of the Socs4R108X/R108X mice still retained low viral loads (Fig. 1E). The augmented susceptibility to influenza infection was independent of genetic background, with similar results observed in congenic Socs4R108X/R108X mice on the C3H/He and C57BL/6 backgrounds (data not shown).
To investigate whether the increased susceptibility to influenza derives from a defect in the hematopoietic compartment, we generated chimeric mice by transplanting bone marrow cells from Thy1.2 Socs4R108X/R108X or Thy1.2 wild-type mice into lethally irradiated Thy1.1 wild-type recipients. The chimeric mice were subsequently infected i.n. with 103 pfu of X31 virus and disease progression monitored for 6 days. Although not significantly different, weight loss and particularly viral load were both increased in Socs4R108X/R108X mice compared to controls (Fig. 1F and G), indicating that a defect in the hematopoietic compartment most likely contributes to these aspects of the phenotype.
The poorer outcomes associated with virulent influenza strains are thought to be due to excessive production of proinflammatory cytokines and chemokines. This dysregulated immune response and the resulting lung inflammation and damage is one mechanism by which pandemic infections cause high mortality [27], [28]. Since Socs4R108X/R108X mice showed greater susceptibility to infection, we compared cytokine and chemokine profiles in lung homogenates from Socs4R108X/R108X and wild-type mice infected with X31 virus. In general, we did not observe significant differences in cytokine and chemokine levels in the latter phase of the infection (days 5–7; data not shown). However, at day 3 post-infection, the Socs4R108X/R108X mice showed significantly higher levels of cytokines and chemokines, such as IL-1β, IL-4, IL-5, IL-6, IL-12p40, IL-13, IFN-γ, and KC (CXCL1), MIP-2 (CXCL2) and MCP-1 (CCL2), respectively (Fig. 2A). There were elevated but not significantly different, levels of TGF-β and IL-10, and no difference was detected in levels of several other cytokines and chemokines (data not shown). The elevated levels of pro-inflammatory cytokines, in particular IL-6, IFNγ and IL-1β, are likely to account for the increased morbidity observed in the Socs4 mutant mice. No difference was observed between Socs4R108X/R108X and control mice in cytokine and chemokine production in either spleen or lungs on d3 following systemic administration of polyinosinic-polycytidylic acid (poly I:C), a non-dynamic, virus-like stimulus (Fig. S2). This indicates that the increased cytokine/chemokine production (Fig. 2A) is a specific response to infection with influenza virus, and may reflect the modestly increased viral titres on d3 (Fig. 1E). It further suggests that that the Toll-like receptor (TLR)3 pathway is not perturbed and therefore not directly regulated by SOCS4. This is consistent with our data suggesting a hematopoietic, rather than an innate epithelial defect.
In order to determine whether the elevated cytokine levels in the lungs were caused by increased production per infiltrating cell or increased numbers of a particular subset of cells, we profiled infected lungs on day 3 post-infection. While total cell numbers in Socs4R108X/R108X-infected lungs were slightly elevated compared to wild-type controls, this difference did not reach statistical significance (Fig. 2B, top panel). Similarly, we observed slightly elevated levels of B cells and NK cells in Socs4R108X/R108X-infected lungs and no difference in neutrophil or macrophage numbers, including alveolar macrophages (CD11c+ Ly6Cint F4/80+) and inflammatory monocytes (CD11c- Ly6Glow Ly6C+ F4/80+) (Fig. 2B, middle panels). Surprisingly, modest but significantly higher levels of total T cells (TCRβ+) including higher levels of CD4 (CD44+) and CD8 (CD44+) T cells were detected in the lungs on day 3 post-infection (Fig. 2B, bottom panels). No difference was observed on day 2 post-infection, whereas on day 6 the cellularity of lungs of wild-type controls was higher than in SOCS4 mutant mice (Fig. S3). We also confirmed Socs4 expression in the infiltrating immune cells recovered by bronchioalveolar lavage (BAL). Socs4 mRNA expression increased over time, peaking at day 3 post-infection (due either to the changing cellular composition or to up-regulation within the infiltrating cells). In contrast, Socs4 mRNA was ∼15-fold lower in equivalent amounts of total lung RNA (Fig. 2C).
Together, these data suggest that activated T cells (not necessarily virus-specific, but CD44hi) accumulate in the lungs during the initial phase of infection in response to the elevated cytokine and chemokine levels. This may be a response to elevated chemoattractant levels or may potentially result from cytokine-driven proliferation (a bystander reaction rather than antigen-driven expansion). In summary, there is an increased net production of pro-inflammatory cytokines and chemokines, and while the initiating defect remains unclear, SOCS4 appears to have a classical role as a negative regulator of cytokine production and/or response in the innate immune reaction to influenza infection.
Following intranasal inoculation with virus, CD8 T cells are primed and activated in the mediastinal lymph node (MLN) [29] and the anti-viral competence of these cells then depends on their ability to migrate to the site of infection [30]. We therefore used MHC class I tetramer staining to track and quantify influenza-specific (KdNP147) CD8 T cells in MLN, BAL and spleen, at days 5, 6 and 7 post-infection. Overall, the expansion of KdNP147-CD8 T cells did not differ between Socs4R108X/R108X mice and wild-type controls (Fig. 3A). However, at days 5 and 6 post-infection Socs4R108X/R108X mice had significantly less virus-specific cells at the site of infection (BAL) when compared to controls and instead, these cells appeared to localise to the spleen. No difference was observed on day 7 post-infection (Fig. 3B), at which time-point, viral titers in the majority of Socs4R108X/R108X mice were below detection levels (Fig. 1E). No differences were observed in MLN during the course of experiment (data not shown).
Although, virus-specific cells did not traffic efficiently to the site of infection, the clearance of virus on day 7 post-infection indicates that the loss of SOCS4 does not affect the ability of CD8 T cells to kill the virus. To confirm this, CTLs were analysed for markers of cytotoxic degranulation and function. Mice were primed with X31 and cells were harvested from lungs, draining lymph nodes and spleen (day 10 post-infection). KdNP147-CD8 T cells were analysed for granzyme B (GzmB) production as well as IFN-γ, TNF and IL-2. Alternatively, cells were stimulated ex vivo with KdNP147 peptide and stained with antibodies against lysosome-associated membrane protein 1 (LAMP-1, CD107a). No differences in GzmB production, cytokine production or CD107a mobilisation were observed between Socs4R108X/R108X mice and wild-type controls (Fig. 3C; data not shown).
The defect in trafficking of KdNP147-CD8 cells to the site of infection could result from reduced activation of the T cells following MHC-antigen presentation or alternatively, from altered homing signals. To investigate this further, we examined a panel of T cell activation markers (CD62L, CD69, CD44, CD25) and homing receptors (CCR5, CXCR3, CCR7) at days 5, 6 and 7 post-infection. The majority of markers examined did not differ between wild-type and Socs4R108X/R108X CD8 T cells (data not shown). However, at day 5 post-infection Socs4R108X/R108X CD8 cells in the draining lymph nodes (MLN) showed comparatively higher levels of CD62L expression (Fig. 4), reflecting reduced activation. By day 6 there were reduced numbers of CD62L positive Socs4R108X/R108X CD8 cells in the lymph nodes (MLN), with no differences in the percentage of CD62Lhi versus CD62Llo expressing cells (Fig. 4B and C), and no differences observed at day 7 post-infection (Fig. 4B and C).
We also reconstituted Socs4R108X/R108X mice with wild-type Thy1.1 donor bone marrow. When analysed 8 weeks post-reconstitution for expression of congenic markers, we observed a much lower degree of chimerism (50–84% in CD4 and 43–78% in CD8 compartments) (Fig. S4A), compared to reconstitution of wild-type mice with Socs4R108X/R108X or wild-type donor cells. This result suggests that SOCS4 may regulate stromal signals, and although unlikely to contribute to the defective anti-viral response described here, warrants further independent investigation. Nevertheless, this competitive reconstitution enabled us to investigate the contribution of both donor and recipient cells to anti-influenza responses. Consistent with the level of chimerism, a higher proportion of wild-type donor cells were detected in the spleens and MLNs of all Socs4R108X/R108X recipient mice at day 6 post-infection. In contrast, the ratios of donor and recipient CD4 and CD8 T cells in the lungs (BAL) were approximately equal (Fig. S4B). Interestingly, the Socs4R108X/R108X Thy1.2 T cells in the lungs appeared to be significantly less activated than wild-type Thy1.1 cells, as evidenced by lower CD69 levels (Fig. S4C), whilst all T cells showed similar levels of CD44 on their surface (Fig. S4D).
Together, these data point to influenza-specific T cells receiving inadequate and/or incorrect signals in the lymph nodes, which in the Socs4R108X/R108X mice, results in their inability to migrate to the site of infection. In turn, this results in a reduced ability to clear the virus. To investigate this further, CD4 and CD8 T cells were isolated from Socs4R108X/R108X and wild-type mice, and signaling responses analysed following CD3 engagement.
We initially examined SOCS4 expression in purified splenic CD4 and CD8 cells following activation with anti-CD3 antibodies. SOCS4 expression was compatible with a role in TCR signaling, with Socs4 mRNA induced within 24 h of TCR stimulation with anti-CD3 antibodies, and peaking at 72 h. In addition, the magnitude of Socs4 expression was 4-fold higher in CD8 versus CD4 cells (Fig. 5A). To investigate T cell responses in the absence of functional SOCS4, purified CD8 T cells from Socs4R108X/R108X or wild-type mice were stimulated with anti-CD3 antibodies and analysed by flow cytometry for the expression of the T cell activation markers CD69, CD44, CD25 and CD62L. In comparison to wild-type cells, surface expression of CD62L remained high, indicating that the Socs4 mutant cells responded poorly to TCR activation (Fig. 5B). CD69, CD44 and CD25 expression levels were comparable in wild-type and Socs4R108X/R108X cells (data not shown). To further investigate the consequences of reduced TCR responses, cells were labelled with Cell Trace Violet (CTV) dye, and proliferation measured in response to stimulation with anti-CD3 antibodies. As shown in Fig. 5C and D, anti-CD3-induced proliferation was impaired in Socs4R108X/R108X CD8, but not in CD4, T cells. The defective proliferation seemed to result from a decrease in the number of cells that were mobilizing per division (Fig.5C, left panel) and resulted in a corresponding decrease in the total number of cells by day 4 of culture (Fig. 5D, middle panel). These differences resulted from a proliferative defect as opposed to differences in the rate of cell death, since the percentage of propidium iodide positive cells was equivalent for both Socs4R108X/R108X and wild-type cells (data not shown).
These results indicate a defect that is intrinsic to the Socs4R108X/R108X T cells and further suggest that signaling through the TCR may be qualitatively different, thus accounting for the reduced activation observed in the lymph nodes during infection and in vitro following TCR engagement.
Loss of functional SOCS4 protein led to a dramatic phenotype following influenza A infection. Socs4R108X/R108X mice were highly susceptible to primary infection with the virulent PR8 H1N1 strain exhibiting increased mortality associated with weight loss and delayed viral clearance. Similar results were observed following infection with the less virulent X31 H3N2 strain, however the anorexia was somewhat ameliorated, enabling us to dissect the underlying pathology in greater detail. The transient weight loss induced by influenza infection is known to reflect viral pathogenicity and as in this study, correlates with increased cytokine/chemokine levels in the lungs [2]. This is the first description of SOCS4-deficient mice and suggests that SOCS4 will play an important role in immune regulation during infection. Deletion of the C-terminal 328 amino acids removes the main functional domains, the SH2 domain and SOCS box, and the conserved SOCS4/5 N-terminal motif [7] is no longer intact. There is no described biological function for the remaining 108 amino acids of SOCS4. The long N-terminal regions of SOCS proteins are predicted to be largely disordered [7], and although such regions within a full-length molecule can play a role in multi-protein complex formation, it is unlikely that a short, disordered fragment will be functional. We have however, been able to express the truncated 108-residue fragment in 293T cells (data not shown). It is possible that, if expressed in vivo, the 108-residue fragment could retain binding to its endogenous target and compete off other signaling intermediates, acting as a dominant negative. It might be speculated that residues 1–108 are involved in binding to a receptor subunit in a similar fashion to SOCS5 binding to the IL-4Rα [31]but the identity of such a receptor complex is currently unknown. Regardless, given that expression of this putative fragment is under the control of the SOCS4 promoter, the observed phenotype indicates that normal SOCS4 function has been disrupted at the endogenous level and reflects a biological role for SOCS4 in regulating anti-viral immunity to influenza A.
So far very little is known regarding the role of different SOCS proteins in influenza infection, outside of the ability of SOCS1 and SOCS3 to regulate specific cytokine receptor complexes. SOCS1 and SOCS3 have been implicated as negative regulators of innate immune responses (type I interferons) via a RIG-I dependent pathway [32], [33], [34]. SOCS3, but not SOCS1, has also been shown to inhibit type I interferon signaling via an NF-κB-dependent pathway [35]. SOCS1 and SOCS3 expression has been associated with symptomatic influenza infection, whereas SOCS2 and SOCS5 have been linked to asymptomatic disease [36].
The pathogenicity of virulent influenza infections is not well understood, but it is accepted that pathogenic strains (such as PR8) cause dysregulation of innate immunity resulting in aberrant cytokine and chemokine production (also known as a cytokine storm), which results in the lung tissue damage. Following infection with a viral strain of relatively moderate virulence, the absence of a functional SOCS4 protein led to dysregulated cytokine production akin to a cytokine storm. This uncontrolled inflammatory response is most likely responsible for the excessive weight loss in Socs4R108X/R108X mice, but it is unclear whether this increased cytokine production also contributes to the delayed viral clearance. Given that the delay in clearance is most pronounced during the adaptive phase of the immune response we postulate that the two phenomena (weight loss and delayed viral clearance) are independent of each other and result from multiple defects caused by the absence of a functional SOCS4.
Airway epithelial cells are the primary targets for influenza infection and are capable of producing cytokines as early as 3–6 hours post-infection [37]. Similarly, a variety of immunomodulators are produced by infected monocytes and macrophages [38]. IL-6 and IL-1β were among the proinflammatory cytokines elevated in the lungs of Socs4R108X/R108X mice (Fig. 2A). IL-6 is known to promote pulmonary inflammation and is a potential biomarker for patients at risk following infection with H1N1 [39]. In addition to its anti-viral activity [40], IL-1 has also been shown to contribute to acute inflammatory lung pathology following influenza infection [41]. The effects of IL-1 and IL-6 are synergistic [42] with both hyper-induced in H5N1 infected human macrophages leading to acute respiratory distress [43]. Although IL-6 expression is dysregulated in Socs4R108X/R108X mice, it does not follow that SOCS4 regulates IL-6 directly. We have tested this theory (data not shown) and are confident that SOCS4 does not directly inhibit gp130 signaling. The elevated levels of MCP-1 and IL-12p40 are also of interest. IL-12p40 is produced by epithelial cells [44] and forms a homodimer (IL-12p80) with both monomeric and dimeric forms implicated in instigating the inflammatory response in lungs [45]. High MCP-1 levels have been associated with both the profound inflammatory responses observed in Ifitm3-/- mice infected with influenza [46], and with the increased severity of disease observed in humans with a variant Ifitm3 gene [47].
Surprisingly, we did not detect major differences in the inflammatory cell infiltrates in the lungs of Socs4R108X/R108X and control mice. Numbers of neutrophils, NK cells, macrophages and inflammatory monocytes were comparable between the two groups. We do not understand the mechanism by which the lack of functional SOCS4 protein results in enhanced cytokine/chemokine production, but given the negligible expression of Socs4 in lung tissue and the relatively equal number of infiltrating cells, we propose that multiple immune cell types are producing greater amounts of cytokine per cell. This is also consistent with retention of the weight loss phenotype in wild-type mice reconstituted with Socs4R108X/R108X bone marrow.
However, our analysis of infected lungs did reveal significantly elevated numbers of T cells on day 3 post-infection, with both CD4 and CD8 showing an activated CD44hi phenotype. This increased number of T cells in the lungs of Socs4R108X/R108X mice is suggestive of proliferation due to bystander activation driven by cytokines rather than antigen-driven expansion of specific T cells or stimulation of cell division by cross-reactive antigens [48], [49], [50]. Interestingly, IL-6 appears to be one of the major factors driving the spontaneous proliferation of naïve T cells (both CD8 and CD4) [51], [52], [53]. Geginat et al. [52] further demonstrated that IL-4 in combination with IL-6 (and other cytokines such as IL-10, IL-12 and TNF) selectively induced the proliferation of naïve CD4 T cells. In our Socs4R108X/R108X mice we observed significantly increased levels of IL-4 and IL-6 as well as elevated levels of IL-10, which combined with other cytokines in a complex lung milieu might lead to the activation and proliferation of T cells already present in the lungs [54].
In addition to the early excessive inflammation, we observed a clear difference in the distribution of virus-specific CD8 T cells in Socs4R108X/R108X mice, with tetramer positive cells appearing to accumulate in the spleen of Socs4R108X/R108X mice instead of migrating to the site of infection, and this discrepancy was particularly apparent by day 5 post-infection (Fig. 3B). Activation of CD8 T cells occurs in response to TCR engagement by antigen in the draining lymph nodes. CD62L expression on naïve T cells is downregulated, with concomitant upregulation of the surface markers CD44, CD25 and CD69 [55]. The most highly activated T cells are then thought to traffic to the lungs, whereas less activated T cells migrate to the spleen and peripheral LNs [29]. Thus, cells migrating to the spleen display a CD69 low/negative phenotype, whilst CD8 T cells in the lungs show a CD69 high phenotype [29]. Consistent with this concept, CD62L down-regulation was less pronounced on Socs4R108X/R108X CD8 cells in the draining lymph nodes at day 5, suggesting that despite the high levels of virus, these cells were not being activated to the same extent as wild-type cells. On day 6 post-infection, Socs4R108X/R108X (Thy1.2) CD8 T cells migrating to the lungs had significantly lower levels of CD69 expression than wild-type (Thy1.1) CD8 cells. Although both subsets appear to be recruited to the lungs in similar ratios, their activation status differed significantly in an identical environment. Given that this result is in the context of a chimeric bone marrow reconstitution where up to 60% of the hematopoietic cells were wild-type, it strongly indicates an intrinsic CD8 T cell defect. CD69 expression declines in the absence of antigenic stimulation [56], however, again the high viral load in the lungs indicates that adequate levels of antigen were available for presentation in the lymph nodes, so that even in the presence of antigenic stimulation Socs4R108X/R108X T cells showed a much weaker activation compared to the wild-type cells.
We conclude that this defect in activation and trafficking of Socs4R108X/R108X CD8 cells to the lungs is likely to be linked to the reduced activation of the virus-specific cells, which in turn suggests impaired TCR activation and signaling. The significantly lower numbers of influenza-specific effector cells available in the Socs4R108X/R108X lungs no doubt result in the much higher viral loads observed (Fig. 1).
A defect in TCR signaling is supported by our in vitro data showing differences in the activation marker CD62L and the reduced capacity of Socs4R108X/R108X CD8 T cells to proliferate in response to anti-CD3-activation. Similar proliferative deficiencies have been observed previously in various settings where TCR signaling was affected [57], [58]. The in vitro proliferative defect (Fig. 5C) is in contrast to the normal expansion of CD8 T cells in response to influenza A infection, as indicated by total numbers of virus-specific CD8 T cells (Fig. 3A). However, such a discrepancy might be due to differences in CD3 versus peptide/MHC stimulation and/or several factors present in vivo that are not accounted for in the in vitro setting, including cytokine milieu or co-receptor stimulation [57]. In fact, in vitro co-stimulation alone can overcome a proliferative defect in cells stimulated only with anti-CD3 antibodies [59]. Therefore, in vivo viral challenge might be sufficient to induce a normal level of T cell expansion and overcome a proliferative impairment detected in vitro. CD62L is an L-selectin that plays a major role in directing lymphocytes to the site of infection and inflammation. Downregulation of CD62L reflects the activation status of T cells and is linked to gene transcription, mRNA stability and shedding from the surface due to increased activity of membrane proteases[60]. We suggest that loss of SOCS4 may affect TCR signalling resulting in maintained CD62L expression. We have also identified differences in CD69 expression between Socs4R108X/R108X and control CD8 T cells supporting this conclusion and indicating that other activation markers can also be affected, depending on the experimental context. We currently postulate that SOCS4 regulates TCR signalling rather than CD62L shedding, although at this stage we cannot rule out the involvement of SOCS4 in ubiquitylation of membrane proteases involved in CD62L shedding.
Our data describe for the first time a phenotype for Socs4R108X/R108X mice in the context of a viral challenge. While SOCS4 plays a negative regulatory role in inflammatory responses to influenza, it appears to be a positive regulator of TCR signaling. The former is consistent with the traditional role of SOCS proteins and a plausible mechanism of action is through inhibition of the JAK/STAT pathways, although the exact target/s are currently unknown. The positive involvement in TCR signaling is a novel function for a SOCS protein.
All mice were bred at the Walter and Eliza Hall Institute's animal facility. Animal experiments followed the NHMRC Code of Practice for the Care and Use of Animals for Scientific Purposes guidelines and were approved by the Walter and Eliza Hall Institute's Animal Ethics Committee (AEC 2008.032 and 2011.031).
SOCS4 mutant mice (Socs4R108X/R108X) were generated through ENU mutagenesis by Ingenium Pharmaceuticals AG (Martinsried, Germany) on the C3HeB/FeJ background. The Socs4R108X mutation was subsequently backcrossed onto both the Balb/c and C57BL/6 backgrounds for 10 generations.
Mice were lightly anaesthetized by inhalation of methoxyflurane, and infected intranasally (i.n.) with 20 plaque forming units (pfu) of PR8 (A/Puerto Rico/8/1934 (H1N1)) or 103-104 pfu of X31 (A/X31(H3N2)) influenza virus in 30 µL PBS. Virus stocks were grown in the allantoic cavity of 10 day old embryonated chicken eggs, from which the viral titre was determined by plaque assay on monolayers of Madin derby canine kidney (MDCK) cells. The weight of mice was monitored daily from day 3 post-infection. Mice were euthanised at various time points following infection and tissues collected for analysis.
Lungs taken from mice after primary viral infection were homogenised and the virus-containing supernatant above the cell debris was harvested and stored at −80°C. Titres of infectious virus in the lung supernatants were determined by plaque assay on monolayers of MDCK cells.
Spleen, mediastinal lymph node (MLN), and bronchoalveolar lavage (BAL) samples were recovered from mice at different stages during the acute phase of the primary infection. BAL samples were incubated on plastic petri-dishes for 1 h at 37°C to remove macrophages for tetramer experiments or used for phenotypic staining without adherent cell depletion. BAL fluid was collected and stored at −80°C for cytokine analysis. The spleens and MLN were disrupted, processed to single-cell suspensions and enriched for CD8 T cells by negative depletion using goat anti-mouse IgG and IgM antibodies to non-CD8 cell lineages (Jackson ImmunoResearch Labs, West Grove, PA, USA). Lungs collected from infected mice were minced, incubated in 2 mg/mL collagenase for 30 min at 37°C, and processed to a single-cell suspension.
Cells from BAL, MLN and spleen were stained with a tetramer conjugated to Strepavidin-PE (Molecular Probes, Eugene, OR, USA) at an optimal staining dilution (1∶100 KdNP147) for 1 h at room temperature (RT). All batches of tetramer used in this study were titrated and the optimal dilution was based on both the percentage of epitope-specific CD8 T cells and the mean fluorescence intensity (MFI) of tetramer staining. Cells were washed twice in FACS buffer, and stained with antibodies to CD4, B220 and F4/80 (all FITC conjugated, dump gate), CD8-APC Cy7, CD3ε-PerCP Cy5.5, CD62L-APC, CD44-PE Cy7 (BD Biosciences or BioLegend) for 30 min on ice, washed twice and analysed by flow cytometry on the FACS Canto (BD Biosciences) and analysed by FlowJo software (Tree Star). For phenotypic staining, different combinations of antibodies were used as indicated in the text.
Cells stained for tetramer and CD8α were fixed and permeabilized using a BD Cytofix/Cytoperm kit (BD Pharmingen), then stained for intracellular GzmB using anti-human GzmB-APC (clone GB12; Caltag Laboratories) as previously described [61]. Acquisition of cell surface CD107a was used to measure activation-induced degranulation by antigen-specific CD8+ T cells [62]. Briefly, cells were incubated for 1 h at 37°C with 1 µM NP147–155 peptide, 10 U/ml hrIL-2 and FITC-conjugated anti-CD107a (clone 1D4B; BD Pharmingen). Monensin (1 µl/ml; BD Pharmingen) was then added and cells were incubated for a further 4 h. Cells were stained for surface CD8α expression, then fixed and permeabilized before intracellular staining with anti-IFN-γ FITC (clone XMG1.2; BioLegend) and anti-GzmB APC. Negative controls incubated in the absence of peptide were used to control for spontaneous production of cytokine or expression of CD107a.
To establish chimeras, recipient mice (Balb/c Thy1.1) were irradiated with two doses of 550R 3 h apart. At 3–5 h after the final irradiation, recipients were reconstituted with 3×106 T cell-depleted bone marrow (BM) cells isolated from femurs and tibias of donor mice (Thy1.2 Socs4R108X/R108X or Balb/c). Briefly, bone marrow was flushed from the femurs of 5–7 week old mice, washed once and incubated in complement-fixing antibodies anti-CD4 (RL172), anti-CD8 (3.186) and anti-Thy-1 (J1j) for 30 min on ice. After washing the cells once in HEPES Earle's medium containing 2.5% fetal calf serum (FCS) (HEM2.5), antibody-binding cells were incubated in rabbit complement at 37°C for 20 min. The mice were allowed to reconstitute for at least 8 weeks prior to use, blood samples were collected (by retro-orbital bleeding), and reconstitution of T cell compartment assessed by FACS analysis.
Cytokine levels were assessed by capture ELISA or BioPlex Pro Assay (BioRad, CA, USA). Lung homogenates were stored at −80°C prior to analysis. BioPlex detection was performed according to the manufacturer's instructions. For ELISAs, U-bottom 96-well plates (Costar, NY, USA) were coated overnight at 4°C with capture antibodies against IL-4, IL-6, IL-10, IFN-γ, GM-CSF, KC, MIP-1α or MIP-2 (BD Pharmingen, CA or RD Systems, MN, USA) and were then incubated with serial dilutions of samples for 2 h at RT, washed and incubated with the appropriate biotinylated detection antibody for 2 h at RT, followed by incubation with streptavidin-horseradish peroxidase. Plates were developed with 3,3′,5,5′-tetramethylbenzidine (TMB) in 0.1 M sodium acetate pH 6, and colour development terminated by the addition of 2N H2SO4. Absorbance values were read at 450 nm. TGF-β ELISA was performed according to the manufacturer's instructions (RD Systems, MN, USA).
Single-cell suspensions were generated from spleens, and CD8 or CD4 T cells purified by negative selection using magnetic beads (Dynabeads, Invitrogen or BioMag, Qiagen). Enrichment of cells was verified by flow cytometry. Purified T cells were labelled with Cell Trace Violet (CTV) (Molecular Probes, OR, USA) according to the manufacturer's instructions. In vitro stimulation assays were performed by plating cells at 104 cells per well in RPMI 1640 medium containing 10% (v/v) heat-inactivated FCS (Sigma-Aldrich, MO, USA), 5×10−5 M 2-ME (Sigma-Aldrich), 100 µg/mL streptomycin and 100 U/mL penicillin (Invitrogen Life Technologies) into anti-CD3-coated (10 µg/mL, clone KT3-1-1) 96-well plates. Recombinant mouse IL-2 (20 ng/mL) was also added to wells. Cells were harvested at different timepoints, propidium iodide (2 µg/mL) and 5000 of Sphero Nile Red Fluorescent Particles (BD Biosciences) added per well and cell division analysed by flow cytometry. Each sample was analysed in duplicate.
Q-PCR analysis of Socs4 mRNA expression was performed essentially as described [63].
Statistical analyses were performed using the unpaired t-test provided within GraphPad Prism 5 software and Compare Groups of Growth Curves software package available on the Walter and Eliza Hall Institute Bioinformatics Division's website http://bioinf.wehi.edu.au/.
|
10.1371/journal.pntd.0007078 | Synthetic peptides as a novel approach for detecting antibodies against sand fly saliva | Hosts repeatedly bitten by sand flies develop antibodies against sand fly saliva and screening of these immunoglobulins can be employed to estimate the risk of Leishmania transmission, to indicate the feeding preferences of sand flies, or to evaluate the effectiveness of vector control campaigns. Previously, antibodies to sand fly saliva were detected using whole salivary gland homogenate (SGH) or recombinant proteins, both of which also have their disadvantages. This is the first study on sand flies where short peptides designed based on salivary antigens were successfully utilized for antibody screening.
Specific IgG was studied in hosts naturally exposed to Phlebotomus orientalis, the main vector of Leishmania donovani in East Africa. Four peptides were designed by the commercial program EpiQuest-B, based on the sequences of the two most promising salivary antigens, yellow-related protein and ParSP25-like protein. Short amino acid peptides were synthesised and modified for ELISA experiments. Specific anti-P. orientalis IgG was detected in sera of dogs, goats, and sheep from Ethiopia. The peptide OR24 P2 was shown to be suitable for antibody screening; it correlated positively with SGH and its specificity and sensitivity were comparable or even better than that of previously published recombinant proteins.
OR24 P2, the peptide based on salivary antigen of P. orientalis, was shown to be a valuable tool for antibody screening of domestic animals naturally exposed to P. orientalis. We suggest the application of this promising methodology using species-specific short peptides to other sand fly-host combinations.
| Previously, two types of antigens were used for detection of antibodies to sand flies: 1) salivary gland homogenate (SGH), which requires maintaining a sand fly colony and laborious work to obtain a sufficient amount of antigen or 2) recombinant proteins with the need to use cell expression and a complicated purification procedure. In contrast, synthetic peptides have never been studied for sand flies despite it being easier to produce them in sufficient quantities and purity. In this study, we screened specific antibodies to sand flies in domestic animals using synthetic peptides based on the two most antigenic salivary proteins of Phlebotomus orientalis. This sand fly is the main East African vector of Leishmania donovani, causative agent of visceral leishmaniasis, and we detected specific anti-P. orientalis IgG in naturally exposed dogs, goats, and sheep from Ethiopia. We showed that, in dogs and goats, the peptide named OR24 P2 is more suitable for antibody detection then the recombinant proteins. Therefore, we recommend this peptide to replace SGH in larger epidemiological studies for evaluation of the effectiveness of vector control programmes or to estimate the risk of Leishmania transmission.
| The specific IgG antibody response against salivary proteins is induced in repeatedly exposed hosts after being bitten by the female sand fly (reviewed by Ribeiro and Francischetti [1] and Lestinova et al. [2]). In sand flies this antibody response is species-specific [3,4] and correlates with the biting intensity [5–8]. IgG values decrease after the hosts are protected against sand flies [9], therefore the detection of antibodies can be used for testing the efficacy of vector control campaigns [10,11].
Antibody detection with the whole salivary gland homogenate (SGH) as antigen is impractical in large epidemiological studies due to the possibility of crossreactivity with other insects [9], variability of saliva composition during sand fly aging [12,13], and the workload required to obtain sufficient quantity of the antigen. In the past decade, sand fly SGH was replaced by several antigenic recombinant proteins, expressed in bacterial or mammalian cells, and with various degrees of success (reviewed by Lestinova et al. [2]). In humans, successful detection of anti-sand fly IgG with recombinant proteins was described by Teixeira et al. [14] and by Souza et al. [15] for Lutzomyia longipalpis and by Marzouki et al. [16,17] and Mondragon-Shem et al. [18] for Phlebotomus papatasi. In domestic animals, using recombinant antigens, antibodies against sand fly saliva were detected in sera of dogs bitten by L. longipalpis or P. perniciosus [14,19,20] and in sera of dogs, sheep, and goats exposed to P. orientalis [21]. In wild animals these studies were performed with rabbits and hares bitten by P. perniciosus [22] and with foxes exposed to L. longipalpis [14].
However, production of recombinant proteins requires cell expression and a complicated purification procedure. Therefore, we focused on linear B-cell epitopes (synthetic peptides, representing short amino acid sections of the antigenic proteins), which can be produced in large amounts with high purity. This approach was previously applied to mosquitoes as well as to tsetse flies. In Anopheles gambiae the peptide designed based on the salivary protein gSG6 was validated in many field studies [23–26] and promising results were also achieved with peptide based on the salivary protein of Aedes aegypti and human serum samples [27]. In tsetse flies, peptides originating from saliva of Glossina palpalis gambiensis and G. morsitans specifically bound anti-tsetse fly antibodies in human and cattle sera, respectively [28–30].
In this study we applied, for the first time, this novel approach to sand flies and used short peptides to detect specific IgG response in domestic animals (dogs, goats, and sheep) naturally exposed to P. orientalis. Our main aim was to compare the peptides with previously described recombinant proteins [21] and to assess whether this methodology is also applicable to large scale surveillance.
BALB/c mice were maintained and handled in the animal facility of Charles University in accordance with institutional guidelines and the 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 (Permit Number: MSMT-10270/2015-6) and were performed under the Certificate of Competency (Registration Number: CZ 02457) in accordance with the Examination Order approved by Central Commission for Animal Welfare of the Czech Republic. Sera of domestic animals were collected within the study by Rohousova et al. [31]. Their collection was approved by the Ethiopian National Research Ethics Review Committee (NRERC) under approval no. 3.10/3398/04.
Sera of domestic animals naturally exposed to P. orientalis in Ethiopia were obtained during the previous study by Rohousova et al. [31] and include 40 sheep, 94 goats, and 30 dogs. Sera from 10 sheep, 15 goats, and 10 dogs from non-exposed animals originating from the Czech Republic served as negative controls. More details of all the samples are provided by Rohousova et al. [31]. Twenty laboratory Balb/c mice were divided into four groups of five animals. Three groups were exposed at least ten-times to about 150 insectary-bred sand fly females (at two-week intervals) of either P. orientalis, P. papatasi, or Sergentomyia schwetzi; the fourth group was used as the non-exposed control.
The Phlebotomus orientalis colony originating from Ethiopia (for more details see Seblova et al. [32]) was reared under standard conditions as described by Volf and Volfova [33]. Salivary glands were dissected from 4–6 day old female sand flies in 20mM Tris buffer with 150mM NaCl and stored at -20°C. Before use, salivary glands were disrupted by freeze-thawing three times in liquid nitrogen.
Peptides were designed from amino acid sequences based on the two most suitable recombinant proteins of P. orientalis (rPorSP24 and rPorSP65) as previously described [21]. Two peptides from each protein were selected in the software EpiQuest-B (Aptum Biologics Ltd., www.epiquest.co.uk). In EpiQuest-B, immunodominant parts of protein sequences were distinguished and their antigenicity indices were calculated based on three algorithms–the peptide immunogenicity, the probability of antibody-accessibility (exposure on the protein surface), and the uniqueness of protein sequence. The probability of peptide-antibody binding increases with the antigenicity index.
These four generated sequences (Table 1) were sent to a commercial laboratory (Genosphere Biotechnologies, France), where they were synthesised and conjugated with two molecules of polyethylene glycol, which acts as a spacer on ELISA plates and facilitates improved accessibility of antibodies. After the spacer, one molecule of biotin was added, which enabled avidin-biotin peptide binding to ELISA plates coated with diluted avidin. Peptides were diluted in sterile PBS at a concentration 1 mg/ml and stored in -80°C.
ELISA Clear Flat-Bottom Plates (3855: Thermo Fisher Scientific, USA) were coated with avidin (A9275: Sigma-Aldrich, UK) at a concentration 5 0μg/well, diluted in 20mM carbonate-bicarbonate buffer (pH 9.5) and incubated overnight at 4°C. Plates were washed three times with PBS-Tw (0.05% Tween 20), blocked with 6% blocking medium diluted in PBS (see S1 Table) for 2 hours at 37°C and then washed twice. Peptides diluted in 2% blocking medium in PBS-Tw were added to the wells at a concentration 5 μg/well and the plates incubated for one hour at 37°C. After washing three times, sera diluted in 2% blocking medium in PBS-Tw were incubated on the plates for one hour at 37°C. Plates were washed five times and secondary antibodies diluted in PBS-Tw were added and incubated for one hour at 37°C. Finally, plates were washed six times, the reaction developed with phosphate-citrate buffer (pH 5.5) in the dark for six minutes at room temperature and stopped with 10% sulfuric acid. The optical density was measured at 492 nm using the Infinite M200 microplate reader (Tecan, Switzerland). At each step, 100 μl of each solution per well was used and all serum samples were tested in duplicate.
When the salivary gland homogenate (SGH) was used as antigen, ELISA plates were coated with 0.2 gland/well [21]. The step with peptide incubation was replaced by incubation with 2% blocking medium in PBS-Tw and the rest of the protocol remained the same. Blocking media, sera and conjugate dilutions for individual host species are indicated in S1 Table.
The non-parametric Spearman test was used to assess correlations between total anti-SGH and anti-peptide IgG levels using GraphPad Prism version 6 (GraphPad Software, Inc., San Diego, CA, USA). For evaluation of the possible crossreactivity with other sand fly species the non- parametric Wilcoxon Rank-Sum test in GraphPad Prism version 6 was used. Statistical significance was considered when the p-value was < 0.05. Cut-off values were calculated from the mean optical density of control sera plus 3 standard deviations. The optical density values of anti-SGH antibodies were used as the gold standard to validate peptides in ELISA tests using positive (PPV) and negative predictive values (NPV), sensitivity, and specificity.
For designing the peptide sequences, two of the most antigenic proteins previously tested in recombinant form were used: rPorSP24 (yellow-related protein) and rPorSP65 (ParSP25-like protein). The antigenicity was calculated for both protein sequences in EpiQuest-B and two peptides with the higher antigenicity indices (Fig 1) were chosen from each protein: OR24 P1, OR24 P2, OR65 P1 and OR65 P2.
First, the synthetic peptides were tested by ELISA for possible crossreactivity with antibodies against salivary antigens of sympatric sand fly species (P. papatasi and Sergentomyia schwetzi) using sera of experimentally bitten Balb/c mice. Five mice were exposed to single sand fly species–either P. orientalis, P. papatasi, or S. schwetzi, and five mice served as non-exposed controls. Significant differences in OD values were detected with sera of mice bitten by P. orientalis compared to all the other three groups, as shown in Fig 2. No differences were observed in non-exposed controls and mice exposed to P. papatasi or S. schwetzi (Fig 2).
The four aforementioned peptides were used as antigens in ELISA experiments to detect the specific anti-P. orientalis SGH antibodies from three animal species–dogs, goats, and sheep. Their antigenicities were compared with the whole SGH, and the statistical values calculated were cut-off, positivity, correlation coefficient, PPV, NPV, specificity, and sensitivity (Table 2).
OR24 P1 showed the closest cut-off value to SGH with canine sera, as well as high correlation coefficient (> 0.75), PPV, specificity (> 0.65) and very high NPV and sensitivity (> 0.9). With the goat sera, the correlation (0.6) was lower as were other statistical values (all between 0.6–0.7). With sheep sera, all statistical values were very high (> 0.9) except for the low sensitivity (0.25).
Peptide OR24 P2 had a high correlation coefficient with the SGH for dog and goat sera. It reached comparable PPV, NPV, specificity, and sensitivity as OR24 P1 for dogs and the highest PPV, NPV, specificity and sensitivity for goats (all above 0.75). Correlation and PPV (< 0.45) were low for sheep sera.
OR65 P1 showed similar statistical values as the previous two peptides for dogs but the specificity was slightly lower (0.5). With goat sera, there was a very high cut-off value and a low correlation coefficient (0.55). The second highest correlation (0.7) was achieved with this peptide and with sheep sera but the PPV and sensitivity (< 0.35) were low.
Despite high statistical values for OR65 P2 and canine sera (0.8), high cut-off and the lowest correlation coefficient (0.65) were observed with this host species. In contrast, OR65 P2 was the second best antigen for goats, with the lowest cut-off value, high correlation, PPV, NPV, specificity, and sensitivity (all above 0.65). Although high PPV, NPV, and specificity (> 0.9) were detected with sheep sera, this peptide achieved low correlation and NPV (< 0.6).
The correlation analysis for the most promising peptide (OR24 P2) with dogs and goats is shown in Fig 3.
Previously, anti-sand fly IgG was detected by using SGH or recombinant proteins prepared based on the most antigenic proteins (reviewed by Lestinova et al. [2]). In this study, we focused on the novel approach of detecting IgG with short amino acid chain synthetic peptides. These peptides can be synthesised in large amounts with very high purity and without the need for cell expression of recombinants. We designed four peptides, two from each of the most promising P. orientalis salivary proteins: yellow-related protein (PorMSP24, ACCN: AGT96461) and ParSP25-like protein (PorMSP65, ACCN: AGT96466), and tested them with sera of domestic animals naturally exposed to Phlebotomus orientalis.
Phlebotomus orientalis is the main East African vector of Leishmania donovani–the causative agent of visceral leishmaniasis [35]. Previous studies revealed that salivary antigens of P. orientalis belong to several protein families, specifically yellow-related proteins, odorant-binding proteins, apyrases, antigen 5-related proteins and ParSP25-like proteins [34]. Recombinant yellow-related proteins and ParSP25-like proteins were used to replace P. orientalis SGH for antibody screening of domestic animals in Ethiopia [21].
Application of recombinant yellow-related proteins of Lutzomyia longipalpis was also described for canine, fox, and human sera [14,15], and of P. perniciosus for hare, rabbit, and canine sera [19,20,22]. However, so far, the recombinant ParSP25-like protein has not been used for other sand fly species except P. orientalis [21].
Peptides based on salivary antigens have not previously been used for studies on sand flies but they have been applied to detection of specific antibodies in hosts bitten by mosquitoes or tsetse flies. For mosquitoes, the peptides were first used for Anopheles gambiae, to study specific IgG responses among humans living in different foci of Plasmodium falciparum transmission [23,25,36], to correlate IgG levels with the risk of malaria transmission [26], and to monitor the effect of vector control campaigns [24]. Ndille et al. [27] used salivary peptide of Aedes aegypti to describe a positive correlation between specific IgG responses in humans with rainfall and mosquito seasonality. In tsetse flies, differences in anti-salivary peptide IgG titers were observed between two human populations with diverse abundance of Glossina palpalis gambiensis [28], and before and after vector control [29]. Somda et al. [30] suggested that the peptide based on salivary protein of G. morsitans was not suitable for IgG screening of domestic animals in areas with high tsetse fly abundance, because it was only recognized by sera of cattle with low exposure. Our study is the first, for sand flies, in which antigenicity and IgG detection are compared for peptides, whole SGH and recombinant proteins.
The peptides designed for P. orientalis are species-specific; no crossreactivity was observed with sera of mice exposed to the sympatric sand fly species P. papatasi and Sergentomyia schwetzi or with sera of non-exposed controls; there was similar high species specificity for the SGH and recombinant proteins of P. orientalis [21].
Previous work on dogs with recombinant proteins showed that the best protein-SGH performing recombinant (ParSP25-like protein) had very low specificity. This implied high probability of false positivity among non-exposed animals. Higher specificity was achieved with recombinant yellow-related protein [21]. Comparable correlation with recombinant yellow-related protein was detected with three peptides–OR24 P1, OR24 P2, and OR65 P1. The first two of these peptides also showed much higher specificity (0.7) than both of the recombinants.
With goats, low correlation was observed with both recombinant proteins [21]. In contrast, peptide OR24 P2 based on the sequence of yellow-related protein reached high correlation with SGH (0.8) as well as values > 0.75 for PPV, NPV, specificity, and sensitivity.
Results with sheep sera were difficult to interpret due to the very low positivity with SGH (10%). Similar positivity was found with all four peptides: even a relatively small change in the number of false positives or false negatives would significantly change calculation of PPV, NPV, specificity, or sensitivity. Although the correlation with two peptides (OR24 P1 and OR65 P1) was above 0.7, we therefore do not recommend these peptides for screening sheep sera. However, promising results for sheep sera have previously been achieved with both recombinant proteins [21].
In summary, we tested four short amino acid sequence peptides, designed based on two most antigenic P. orientalis salivary proteins, for detection of antibodies to sand fly saliva, in three species of domestic animals from Ethiopia. One of the peptides, OR24 P2, showed promising results with sera of dogs and goats. We therefore suggest that this peptide may replace SGH or recombinant proteins in surveillance for anti-P. orientalis IgG. As it was shown, synthetic peptides might work only for some host species. For future detection of human antibodies to sand fly saliva, we recommend comparison of the efficacy of recombinant proteins and synthetic peptides.
|
10.1371/journal.pntd.0003821 | Toward Measuring Schistosoma Response to Praziquantel Treatment with Appropriate Descriptors of Egg Excretion | The control of schistosomiasis emphasizes preventive chemotherapy with praziquantel, which aims at decreasing infection intensity and thus morbidity in individuals, as well as transmission in communities. Standardizing methods to assess treatment efficacy is important to compare trial outcomes across settings, and to monitor program effectiveness consistently. We compared customary methods and looked at possible complementary approaches in order to derive suggestions for standardizing outcome measures.
We analyzed data from 24 studies conducted at African, Asian, and Latin American sites, enrolling overall 4,740 individuals infected with Schistosoma mansoni, S. haematobium, or S. japonicum, and treated with praziquantel at doses of 40–80 mg/kg. We found that group-based arithmetic and geometric means can be used interchangeably to express egg reduction rates (ERR) only if treatment efficacy is high (>95%). For lower levels of efficacy, ERR estimates are higher with geometric than arithmetic means. Using the distribution of individual responses in egg excretion, 6.3%, 1.7% and 4.3% of the subjects treated for S. haematobium, S. japonicum and S. mansoni infection, respectively, had no reduction in their egg counts (ERR = 0). The 5th, 10th, and 25th centiles of the subjects treated for S. haematobium had individual ERRs of 0%, 49.3%, and 96.5%; the corresponding values for S. japonicum were 75%, 99%, and 99%; and for S. mansoni 18.2%, 65.3%, and 99.8%. Using a single rather than quadruplicate Kato-Katz thick smear excluded 19% of S. mansoni-infected individuals. Whilst the effect on estimating ERR was negligible by individual studies, ERR estimates by arithmetic means were 8% lower with a single measurement.
Arithmetic mean calculations of Schistosoma ERR are more sensitive and therefore more appropriate to monitor drug performance than geometric means. However, neither are satisfactory to identify poor responders. Group-based response estimated by arithmetic mean and the distribution of individual ERRs are correlated, but the latter appears to be more apt to detect the presence and to quantitate the magnitude of suboptimal responses to praziquantel.
| To identify whether a person is infected with parasitic worms, stool or urine samples are examined for worm eggs. The drug praziquantel is used against the parasitic disease schistosomiasis. However, there is no definitive agreement as to how the efficacy of praziquantel is best expressed. We put together a database from various studies of the efficacy of praziquantel against schistosomiasis. Efficacy was measured using customary methods: cure rate (CR: percentage of people with eggs in their stool/urine before treatment who became egg-negative after treatment); and egg reduction rate (ERR; percentage reduction in the number of eggs in the stool/urine after treatment, where the mean number of eggs from all people treated is calculated using either geometric or arithmetic means). We found that arithmetic and geometric means can be used interchangeably only if treatment efficacy is very high; arithmetic means are more sensitive to capture drops in efficacy expressed by ERR. A valid complement for drug efficacy monitoring is to study the distribution of individual responses in egg excretion that allows identifying in a single measure both those who had an adequate response to treatment and those who respond less well; e.g., the 5% of the patients with the lowest ERRs.
| Schistosomiasis is a parasitic disease caused by blood flukes of the genus Schistosoma. The three main species infecting humans are S. haematobium (causing urogenital schistosomiasis), S. japonicum, and S. mansoni (the latter two responsible for intestinal schistosomiasis) [1]. The backbone of the global strategy for controlling the morbidity caused by schistosomiasis is the periodic administration of single-dose oral praziquantel (usually given at 40 mg/kg body weight). This strategy is termed “preventive chemotherapy”, whereby praziquantel is administered without prior diagnosis [2,3] to entire communities or target groups, most importantly school-aged children [4–6], depending on the level of endemicity [7,8].
How efficacy of antischistosomal drugs (and anthelmintic treatments in general) should be measured has been, and still is, a matter of debate in the research and disease control communities. One limitations of the current treatment outcome measure–i.e., parasite egg excretion–is that it is a proxy for drug effects on adult worms, which could also be confounded by various factors, including facultative temporary cessation of excretion by the adult worm [3]. Other, more direct but not yet widely used methods of worm vitality are the detection of specific antigens, like the circulating cathodic antigen (CCA) and the circulating anodic antigen (CAA) [9,10]. Clinical trials have traditionally used cure rates (CRs) as the main drug efficacy endpoint, and expressed results as the proportion of infected individuals who convert to a negative stool or urine sample post-treatment [11–13]. However, the World Health Organization (WHO) recently issued guidelines that recommend the egg reduction rate (ERR) as the primary outcome measure, especially when assessing programmatic treatment effectiveness [3]. This entails a quantitative diagnostic test based on the microscopic detection and enumeration of parasite eggs in small amounts of stool (usually 41.7 mg) and the estimation of the number of eggs per 1 g (EPG) of feces (S. japonicum or S. mansoni) or per 10 ml of urine (S. haematobium). The ERR measures the overall effect of treatment on the entire group of infected subjects treated (ignoring individual variability) and is expressed as the ratio between the mean of the pre- and post-treatment egg counts [3]. ERR is considered more suitable than CR to assess the impact of preventive chemotherapy on morbidity (which is commensurate to infection intensity) in the context of continuous risk of reinfection and in view of the low sensitivity of current diagnostic methods [12,14].
Which type of means (e.g., geometric mean (GM) or arithmetic mean (AM)) should be used to express treatment outcomes against helminthiases at the community level is an additional subject of debate [15,16]. Thus far, studies of treatment efficacy have predominantly reported results using GM egg counts [15] but recently, the use of AM egg counts has been advocated [17,18]. The issue is that egg counts are not normally distributed, even after logarithmic transformation, which would call for using GM [19]. However, GM hide extreme values (e.g., (i) a small proportion of individuals disproportionally contributing to total egg excretion, and (ii) individuals who do not respond to treatment), which are important when assessing the effects of interventions, and which are better captured by using AM.
A further complication when dealing with different studies is the diversity in methodologies, in particular: (i) which diagnostic method is used (e.g., single or multiple urine filtration for detection of S. haematobium eggs) [20] or the Kato-Katz technique for detection of S. japonicum or S. mansoni eggs in fecal samples, whose sensitivity depends upon the baseline infection intensity, the number of thick smears from a single sample, and the number of stool specimens examined [21–24]; and (ii) how many weeks post-treatment effects are measured, which also depends on Schistosoma species [25,26].
The objectives of this paper were to compare customary methods to assess the efficacy of praziquantel for treating schistosomiasis and to explain differences; to identify possible alternative approaches to express treatment effects on egg excretion; and to verify whether the size of treatment effects for intestinal schistosomiasis change when measured with a single or quadruplicate Kato-Katz thick smears. The overall aim of these analyses was to derive suggestions for standardizing outcome measures in future drug efficacy studies. For this purpose, we combined and analyzed available data from various studies where praziquantel was used to treat infections with different Schistosoma species.
All studies selected for the current secondary analyses had been approved by the relevant institutional review boards and ethics committees, and were conducted according to international ethics standards (for details, see individual publications [25–36]). Data received from the individual studies were completely anonymized.
We built a common database from 24 studies including 4,740 individuals assigned to three different treatment groups who had received either 40 mg/kg (18 studies, 3,713 individuals), 60 mg/kg (five studies, 690 individuals), or 80 mg/kg (one study, 337 individuals) praziquantel against S. haematobium [25,27–29,34,36], S. japonicum [31,32], and S. mansoni [21,23,26–28,30,31,33,35].
The main study characteristics, including Schistosoma species, praziquantel dose, age of participants, time-point of treatment follow-up, and diagnostic approach, are summarized in Table 1. Studies enrolled a total of 4,740 individuals, of whom 2,633 (55.5%) were infected with S. haematobium, 1,804 (38.1%) with S. mansoni, and the remaining 303 (6.4%) with S. japonicum. Studies generally enrolled children and adolescents except one study in the People’s Republic of China [32] and another one in Côte d’Ivoire [33]. The praziquantel dose was 40 mg/kg in 17 studies, 60 mg/kg in five studies, and 80 mg/kg in one study. Follow-up was within 3 weeks in 13 studies, four weeks in two studies, within 2 months in six studies, and longer in the remaining three studies.
For the detection of S. haematobium infection, two diagnostic approaches were employed: (i) a single urine filtration slide in five studies (one of them was carried out on the same sample but at two different time points) [25,26,28,29,34]; and (ii) duplicate urine filtration slides in one study [36]. For the diagnosis of S. japonicum, duplicate Kato-Katz thick smears from each of two stool specimens were subjected to microscopic examination in all studies. For the detection of S. mansoni, the most common diagnostic approach was duplicate Kato-Katz thick smears from each of two stool specimens. In one study, a single Kato-Katz thick smear was performed on five samples at baseline and triplicate Kato-Katz thick smears for four samples at follow-up [35,37].
Treatment response was assessed both at the overall group and the individual level. The AM and GM EPG values were calculated at pre- and post-treatment for S. mansoni and S. japonicum by multiplying the individual fecal egg counts (FECs) obtained by a single Kato-Katz thick smear (41.7 mg) by a factor of 24. For S. haematobium, egg counts are presented as eggs per 10 ml of urine. Drug efficacy was expressed as ERR and CR.
ERR (arithmetic (ERRAM) or geometric (ERRGM)) was calculated as the ratio of the difference between the (arithmetic or geometric) means of the pre- and post-treatment EPG or eggs per 10 ml urine to the pre-treatment (arithmetic or geometric) mean EPG or eggs per 10 ml urine: ERR = [(mean egg countpre-treatment − mean egg countpost-treatment) / mean egg countpre-treatment] x 100
GM egg counts were calculated as follows: exp[∑i=1i=njkln(xijk+c)njk]−c, where xijk is the observed egg count for individual host i, Schistosoma species j, and study k; njk is the number of hosts who provided a (pre- and post-treatment) sample for determination of infection intensity for each Schistosoma species and study, and c is a constant added to each count to allow inclusion of zero counts (negative test) [38]. Confidence intervals (CIs) for the ERR (calculated with AM and GM) were determined by using a bootstrap resampling method (with replacement) over 1,000 replicates and expressed as a univariate calculation of the 2.5th and 97.5th percentiles.
Individual ERR was calculated as the ratio of the difference between the pre- and post-treatment EPG or eggs per 10 ml urine to the pre-treatment EPG or eggs per 10 ml urine multiplied by 100.
CRs and 95% binomial CIs were the percentage of infected individuals negative for Schistosoma (in their urine or stool) at post-treatment follow-up. The distribution of individual responses in egg excretion was categorized as (i) negative (corresponding to CR), (ii) reduction, (iii) no change or increase, and further expressed in centiles to quantitate the fraction of poor responders.
We compared (i) the ERRAM versus ERRGM and (ii) the CR versus ERR. The results were presented graphically in modified L’Abbé plots with 95% CIs for both comparisons, and additionally in Bland and Altman plots for ERR using AM and GM. The coefficient of determination (R2) was also calculated.
A linear mixed model of the difference of the ERRs calculated as GM and AM (ΔERRg,a) was developed to estimate which parameter could better predict the difference between the two ERR calculations, with the average of the two mean ERRs (GM and AM) set as covariate. Such a model was extrapolated from the Bland and Altman regression by further including predictive factors. In order to evaluate the effect of different factors on ΔERRg,a, the ERR was calculated on the different strata defined by the combination of the categories of the following parameters: age, sex, treatment dose, and Schistosoma species. The same age categories were defined across all studies. The linear mixed model was estimated including these parameters as independent factors. 95%CIs of the difference between the two ERR calculations were calculated by using Tango’s score confidence interval which was shown in the literature to outperform other calculation methods in the case of correlated proportions [39,40].
Modeling was carried out through a shrinkage method of variable selection. Variables were first selected using the ElasticNet procedure, which is mixing a least absolute shrinkage and selection operator (LASSO) procedure and ridge regression [41]. Subsequently, a variance-covariance matrix structure was selected among unstructured, variance components, autoregressive, compound symmetry and Toeplitz structures, which minimized the Aikake information criterion corrected (AICC) for finite sample size. Post-hoc tests on each parameter were carried out with a Tukey adjustment. Pairwise differences in least square means (LSM) were calculated for fixed values of the average of the AM and GM ERRs (for the range 70–99%), thereby evaluating the influence of exogenous parameters on the bias between the two ERRs. This bias was evaluated in the Bland and Altman method by regressing the difference of the two methods by their mean.
All tests were two-tailed; a p-value of 5% was deemed significant. Only studies with treatment follow-up examination done within 90 days were included in the models (to minimize the effect of reinfection after praziquantel administration). Calculations and analyses were performed by using SAS system version 9.3 (SAS Institute, Cary, NC, United States of America).
For S. mansoni, we compared single (using the first thick smear on the first stool specimen) versus quadruplicate (using four thick smears of the same stool specimen) Kato-Katz thick smears for expressing CR and ERR (with AM and GM). In this sub-analysis, we only included individuals infected with S. mansoni whose first slide on the first fecal sample was positive.
Within this population, we calculated and compared the overall AM and GM pre- and post-treatment FECs and the respective ERRs and CRs based on single and quadruplicate Kato-Katz thick smears with 95% CIs calculated by boot-strapping. The results of the individual studies were presented graphically in Bland and Altman plots and in modified L’Abbé plots with 95% CIs, and the coefficient of determination (R2) was calculated.
The distributions of the raw egg counts at baseline by Schistosoma species for each study (including AMs and GMs) are presented in Fig 1. Efficacy outcomes were analyzed on a total of 4,375 individuals with pre- and post-treatment egg counts. Among them, 2,365 (54.1%) were infected with S. haematobium, 1,708 (39.0%) with S. mansoni, and the remaining 300 (6.9%) with S. japonicum. Details of baseline FECs and drug efficacy outcomes, including group-based means and individual responses by study and Schistosoma species, are presented in Table 2.
ERRAM ranged from 17.0% to 99.8% and ERRGM from 50.7% to 99.8% for S. haematobium; 81.5–100% (ERRAM) and 95.9–100% (ERRGM) for S. mansoni; 83.5–99.9% (ERRAM) and 86.9–99.9% (ERRGM) for S. japonicum. The 95% CIs estimated by boot-strapping tended to be wider with AM compared to GM. Among the 24 studies included in our analyses, six had an ERR <90% by AM (two if restricted to studies with 3 weeks’ follow-up), but only two by GM.
The modified L’Abbé plot (Fig 2A) indicates that ERRs tend to be higher when calculated using GM. The R2 of the linear regression showed a strong linear correlation for S. japonicum (R2 = 1.00, three studies) and S. haematobium (R2 = 0.88, eight studies), but a weaker linear correlation for S. mansoni (R2 = 0.46, 13 studies).
The mixed linear model found a significant relationship between the difference between ERRAM and ERRGM and their mean value. However, introducing the variables identified as predictors by the ElasticNet procedure, coupled with model averaging based on 1000 replicates (sampling with replacement)–mean baseline epg, year of study and species–rendered this relation non-significant (meaning that none of these covariates could explain the differences between ERRAM and ERRGM.)
LSM pairwise comparisons and ERR (individually for AM and GM) models showed a significantly better consistency between ERRs calculated with AM and GM for S. haematobium than for S. mansoni. Study participants’ age was found to have an effect only for ERRGM (higher for school-aged children and adults than for preschool-aged children; S1 Table).
Group means and individual responses are presented in Table 2. Individual responses are also displayed graphically in Fig 3 (panel A as bar diagrams; panel B as centile distributions). Overall, 6.3% (ranging in individual studies from 0 to 39%), 1.7% (0–5%); and 4.3% (0–11%) of the subjects treated for S. haematobium, S. japonicum, and S. mansoni infection, respectively, had no reduction in their egg counts (ERR = 0). The 5th, 10th, and 25th centiles of the subjects treated for S. haematobium had individual ERRs of 0% (ranging in individual studies from 0 to 99.7%), 49.3% (0–100%), and 96.5% (0–100%); the corresponding values for S. japonicum were 75.0% (0–97.4%), 99.0% (25.6–98.8%), and 99.0% (96.8–99%), and for S. mansoni 18.2% (0–97.4%), 65.3% (0–99.4%), and 99.8% (0–99.8%).
The centile distribution of iERRs in studies with ERRAM <90% was shifted to the right and clearly distinct from those with ERRAM ≥90% (Fig 4).
When considering only studies assessing outcomes within a maximum of 28 days (13 studies evaluated drug efficacy at 21 days, two at 28 days), only two (both treating S. mansoni with 40 mg/kg praziquantel with 21-day follow-up [27,30]) had an ERRAM <90%; in terms of individual responses, the 30th and 36th centile, respectively had ERRs <90%, and 10.8% and 7.1% of patients, respectively had no change in their egg counts (ERR = 0). Three additional studies, all with ERR AM >90%, had individual ERRs <90% in the 13th centile: one study on S. haematobium treated with 40 mg/kg and 21-day follow-up [27], and two studies on S. mansoni treated with 40 or 60 mg/kg with 28-day follow-up [35,37]. In these studies, 3.2%, 3.5% and 2.4%, respectively of individuals had no decrease in egg counts. Across these studies, ERRAM and iERR were highly correlated (R2 0.95).
For all three Schistosoma species, the CR was systematically lower than ERR, regardless of whether AM or GM was employed, except in a single study [29]. ERR >90% corresponded to CRs ranging from 51.4% to 99%. Only when ERRs were very high (range: 97.7–100%) there was a good agreement between both indicators. The CR ranged from 82.1% to 100% (Fig 5).
Among the 1,435 individuals enrolled in the studies who were found positive for S. mansoni eggs in their stool based on quadruplicate Kato-Katz thick smears, 1,167 (81.3%) were diagnosed positive on the first Kato-Katz thick smear. In this subset, we found that the results in the individual studies were highly correlated (Fig 6A–6B; R2 = 0.95 for AM and R2 = 0.86 for GM). The same number of studies (n = 2) had AM ERR of less than 90% with either approach.
On aggregate, the FECs at baseline were lower with quadruplicate than with a single Kato-Katz thick smear when using either AM (1,046 EPG versus 2,617 EPG) or GM (342 EPG versus 575 EPG); the ERR estimates were comparable only by GM (99.5% versus 99.8%) whereas by AM, the estimate was lower with quadruplicate than single examinations (86.9% versus 94.9%) (Table 3). In order to verify if using a single Kato-Katz thick smear selected for a different sample, we compared the baseline FECs and ERRs assessed with quadruplicate Kato-Katz thick smears on the overall sample of these studies (n = 1,435) to that of this subgroup with the first Kato-Katz thick smear positive (n = 1,167): while the overall average baseline FECs was approximately threefold lower for the latter, there was no difference in ERR between the two groups for either AM and GM.
In contrast to ERRs, CRs derived from single versus quadruplicate Kato-Katz thick smears were poorly correlated (R2 = 0.50) (Fig 7). With few exceptions, the CR assessed from single Kato-Katz thick smear (CR = 92.0%) was consistently higher than that from quadruplicate Kato-Katz thick smears (CR = 82.6%) in the individual studies and the difference between the two approaches increased with decreasing efficacy (Fig 7A and 7B). The overall CR was 88.0% applying a single and 79.0% using quadruplicate Kato-Katz thick smears (Table 4).
Preventive chemotherapy with praziquantel is the current backbone of the global strategy to control the morbidity caused by schistosomiasis in high-endemicity areas [2,4,6,8]. Monitoring of praziquantel efficacy should accompany schistosomiasis control programs in order to identify promptly suboptimal responders; to that effect, the WHO has issued standard procedures for control programs based on one single measurement around 3 weeks post-treatment [3]. At the same time, more work is needed to improve the current evidence-base for decision-making: in order to provide reliable information, it is important to agree on a robust statistical approach to assess drug efficacy, especially in clinical trials, and to use standardized, quality-controlled diagnostic methods that are comparable from one setting to another [18,42]. To date, both CR and ERR (the latter based on either AM or GM), are used for assessing drug efficacy. The stool and urine sampling and diagnostic approaches vary across studies, and these issues have important ramifications for drug efficacy estimates.
The controversy over the use of AMs or GMs to measure anthelmintic treatment efficacy as assessed by ERR in parasitic nematodes of cattle, and more recently in those of humans, has already been expounded in several studies, with contrasting results [43] favoring either AMs [17], or GMs (see e.g. [15]) or pointing to inadequacies of both [44]. The overall aim of this study was to compare and contrast different approaches to express treatment effects on Schistosoma infections in order to derive indications for standardizing future studies of drug efficacy.
The first specific objective of this paper was to compare customary measures of drug efficacy. We first considered whether efficacy assessment based on ERR changes when using AM or GM egg counts for the three predominant human Schistosoma species. While both means were in the same range for all species and showed a moderate level of correlation, the discrepancy between AM and GM became wider with decreasing drug efficacy. As previously reported by some [17,43,45]] but not all authors[17], GM estimates tended to be higher than AM. These findings suggest that the two means can be used interchangeably if drug efficacy is high (ERR >95%), but the difference between the two means is expected to increase as efficacy decreases. It is also worth noting that two out of the 13 studies with assessment of efficacy within 3 weeks would not meet the WHO threshold for acceptable efficacy of 90% [3] for ERR when calculated by AM, while only 8% would not meet the WHO threshold, if calculated by GM (2/24). Therefore, between the two, using AM, as suggested by WHO, appears to be more sensitive an approach to identify problems with the response to praziquantel.
We employed models to help further qualifying these findings using explanatory variables. We found that ERRs are more consistent between the AM and GM egg count values of S. haematobium and S. japonicum than of S. mansoni infection, and for school-aged children and adults than for preschool-aged children (but only when using GM). On the contrary, these findings are not accounted for by the baseline and post-treatment distributions of the raw egg counts (S2 Table), intensities of infection (S3 Table), or proportion of individuals with extreme values (S2 Table). Together, these findings are important to allow a meaningful comparison of newer studies using AM to older studies which would have used GM.
We then compared drug efficacy estimates by ERR (using AM or GM) versus CR. While generally used in the past, the CR is known to have some major limitations, and is no longer recommended by WHO for assessing the programmatic efficacy of drugs used in mass drug administration [12,43,45]. As expected, efficacy estimates by ERR and CR were hardly comparable, as they assess two different outcome measures (intensity versus presence of infection). At the onset of schistosomiasis control, the primary goal is to reduce morbidity, which is reflected by infection intensity, and hence, the ERR rather than CR might be the efficacy measure of choice. This choice is further justified by the relatively low sensitivity of widely used diagnostic methods, particularly the Kato-Katz technique for intestinal schistosomiasis [22,24,42] and the fact that current anthelmintic drugs only show low to moderate efficacies in terms of CR [13,46,47].
However, the situation is complex, and neither ERR nor CR alone provide a satisfactory description of the situation. The key questions about outcome measures are between effects on presence versus intensity of infection; and between measures of central tendency (for a group of individuals) versus individual subject responses. Our analyses indicate that the distribution of individual responses in egg excretion may be a better way of expressing results, as it comprises in one single measure drug effects on both presence and intensity of infection, and allows further detailing the distribution in centiles–which helps identifying and quantitating the presence of poor responders. ERRAM and iERR are correlated, but the latter appears to be more apt to detect the presence and to quantitate the magnitude of suboptimal responses to praziquantel. More than 10% of individuals had ERR <90% in five studies, only two of which were identified by applying the ERRAM 90% threshold. In these two studies, with ERRAM ~82%, <30% of individuals did not achieve a 90% reduction in their egg counts, and 11–17% had no reduction at all.
These observations raise important questions as to which approach is best suited to assess drug efficacy for which purpose: while ERR is, currently, the preferred measure at the program level, identifying poor responders is important in view of tracking trends in responses and signaling potential problem areas and emergence of drug resistance. Importantly, discussions should be held with a variety of stakeholders, including drug regulatory authorities, especially if new drug applications are forthcoming.
Lastly, we compared drug efficacy estimates (ERR calculated using AM and GM, and CR) obtained by a single versus quadruplicate Kato-Katz thick smears for S. mansoni. Obviously, a single Kato-Katz thick smear has a lower sensitivity than multiple Kato-Katz thick smears (here ~81%) [48], and the effect and ramifications thereof for estimating drug efficacy against Schistosoma and other helminth infections have been discussed [42,49–51]. At the same time, collecting single stool samples and examining single Kato-Katz thick smears is operationally more feasible and less expensive than multiple sampling and multiple thick smear examinations under a microscope. We found that, when single Kato-Katz thick smears are used, the initial intensities of infection are ~2.5 times higher than with quadruplicate thick smears; efficacy estimation by ERRs based on AM and CR are ~8–9% higher, respectively, whereas ERRs based on GM are similar. Taken together, these results reflect the lower sensitivity of a single Kato-Katz thick smear, which misses low-intensity infections both on enrolment and post-treatment; the bias appears to be proportionally greater for the initial infection intensity than for the treatment outcomes. Overestimation of treatment effects may be an issue with efficacies nearing the 90% ERR threshold. In order to ascertain whether this introduces a selection bias which could affect the estimation of efficacy [52] (i.e., if the sample positive on a single Kato-Katz thick smear is different from that positive on multiple Kato-Katz thick smears), we compared both baseline FECs and efficacy estimates in these two samples using the same diagnostic technique; we found that excluding the ~19% of subjects who were negative on a single Kato-Katz selects for subjects with marginally higher initial FECs, but has ultimately no effect on efficacy estimates. Currently there is no diagnostic ‘gold’ standard method to assess Schistosoma response to treatment [53]. Standards may need to be tailored to the study, whether a field survey (limited by practical imperatives) or a clinical trial (which could afford more complex conditions and costly diagnosis), and whether in high or low infection intensity setting. Part of the problem, however, is the limited sensitivity of the current diagnostic methods, particularly for the detection of low intensity infections, which cannot be corrected until and unless more reliable tests become available [6,42,54].
Using group means is practical when assessing sample effects, but may not be suited to detect small changes, especially those that may occur in early phases of decreasing drug efficacy. We estimate that the distribution of individual responses in Schistosoma egg excretion, which accounts for individual variability of responses to praziquantel treatment, allows measuring effects on both presence and intensity of infection, and helps identifying and quantitating poor responders. More research and larger databases will be required in order to identify meaningful thresholds–e.g., centile by which 90% ERR is achieved; ERR achieved by the lowest 5% or 10% centile–and also analyze in greater detail reasons for poor response.
Both approaches could be used in parallel and complement each-other. It is important to agree on standardized outcome measures that are tailored to specific purposes, such as epidemiological surveys, routine monitoring, clinical trials, morbidity control, or elimination settings. Hence, we invite other groups to contribute to this discussion and scientific inquiry.
|
10.1371/journal.pgen.1002681 | MicroRNA-277 Modulates the Neurodegeneration Caused by Fragile X Premutation rCGG Repeats | Fragile X-associated tremor/ataxia syndrome (FXTAS), a late-onset neurodegenerative disorder, has been recognized in older male fragile X premutation carriers and is uncoupled from fragile X syndrome. Using a Drosophila model of FXTAS, we previously showed that transcribed premutation repeats alone are sufficient to cause neurodegeneration. MiRNAs are sequence-specific regulators of post-transcriptional gene expression. To determine the role of miRNAs in rCGG repeat-mediated neurodegeneration, we profiled miRNA expression and identified selective miRNAs, including miR-277, that are altered specifically in Drosophila brains expressing rCGG repeats. We tested their genetic interactions with rCGG repeats and found that miR-277 can modulate rCGG repeat-mediated neurodegeneration. Furthermore, we identified Drep-2 and Vimar as functional targets of miR-277 that could modulate rCGG repeat-mediated neurodegeneration. Finally, we found that hnRNP A2/B1, an rCGG repeat-binding protein, can directly regulate the expression of miR-277. These results suggest that sequestration of specific rCGG repeat-binding proteins could lead to aberrant expression of selective miRNAs, which may modulate the pathogenesis of FXTAS by post-transcriptionally regulating the expression of specific mRNAs involved in FXTAS.
| Fragile X–associated tremor/ataxia syndrome (FXTAS) is an adult-onset neurodegenerative disorder, usually affecting males over 50 years of age. FXTAS patients are the carriers of fragile X premutation alleles. Using a FXTAS Drosophila model, we previously demonstrated that fragile X premutation rCGG repeats alone could cause neurodegeneration. Pur α and hnRNP A2/B1 were identified as specific premutation rCGG repeat-binding proteins (RBPs) that could bind and modulate fragile X permutation rCGG-mediated neuronal degeneration. MiRNAs are sequence-specific regulators of post-transcriptional gene expression. Here we show that fragile X premutation rCGG repeats could lead to aberrant expression of selective miRNAs, which may modulate the pathogenesis of FXTAS by post-transcriptionally regulating the expression of specific mRNAs involved in FXTAS.
| Fragile X syndrome (FXS), the most common form of inherited mental retardation, is caused by expansion of the rCGG trinucleotide repeat in the 5′ untranslated region (5′ UTR) of the fragile X mental retardation 1 (FMR1) gene, which leads to silencing of its transcript and the loss of the encoded fragile X mental retardation protein (FMRP) [1]–[6]. Most affected individuals have more than 200 rCGG repeats, referred to as full mutation alleles [7]. Fragile X syndrome carriers have FMR1 alleles, called premutations, with an intermediate number of rCGG repeats between patients (>200 repeats) and normal individuals (<60 repeats) [8]. Recently, the discovery was made that male and, to a lesser degree, female premutation carriers are at greater risk of developing an age-dependent progressive intention tremor and ataxia syndrome, which is uncoupled from fragile X syndrome and known as fragile X-associated tremor/ataxia syndrome (FXTAS) [9], [10]. This is combined with cognitive decline associated with the accumulation of ubiquitin-positive intranuclear inclusions broadly distributed throughout the brain in neurons, astrocytes, and in the spinal column [11], [12].
At the molecular level, the premutation is different from either the normal or full mutation alleles. Based on the observation of significantly elevated levels of rCGG-containing FMR1 mRNA, along with either no detectable change in FMRP or slightly reduced FMRP levels in premutation carriers, an RNA-mediated gain-of-function toxicity model has been proposed for FXTAS [13]–[17]. Several lines of evidence in mouse and Drosophila models further support the notion that transcription of the CGG repeats leads to this RNA-mediated neurodegenerative disease [11], [15], [17]–[19]. The hypothesis is that specific RNA-binding proteins may be sequestered by overproduced rCGG repeats in FXTAS and become functionally limited, thereby contributing to the pathogenesis of this disorder [15], [17], [19], [20]. There are three RNA-binding proteins found to modulate rCGG-mediated neuronal toxicity: Pur α, hnRNP A2/B1, and CUGBP1, which bind rCGG repeats either directly (Pur α and hnRNP A2/B1) or indirectly (CUGBP1, through the interaction with hnRNP A2/B1) [21], [22].
MicroRNAs (miRNAs) are small, noncoding RNAs that regulate gene expression at the post-transcriptional level by targeting mRNAs, leading to translational inhibition, cleavage of the target mRNAs or mRNA decapping/deadenylation [23], [24]. Mounting evidence suggests that miRNAs play essential functions in multiple biological pathways and diseases, from developmental timing, fate determination, apoptosis, and metabolism to immune response and tumorigenesis [25]–[31]. Recent studies have shown that miRNAs are highly expressed in the central nervous system (CNS), and some miRNAs have been implicated in neurogenesis and brain development [32]–[34].
Interest in the functions of miRNAs in the CNS has recently expanded to encompass their roles in neurodegeneration. Investigators have begun to reveal the influence of miRNAs on both neuronal survival and the accumulation of toxic proteins that are associated with neurodegeneration, and are uncovering clues as to how these toxic proteins can influence miRNA expression [35]. For example, miR-133b is found to regulate the maturation and function of midbrain dopaminergic neurons (DNs) within a negative feedback circuit that includes the homeodomain transcription factor Pitx3 in Parkinson's disease [36]. In addition, reduced miR-29a/b-1-mediated suppression of BACE1 protein expression contributes to Aβ accumulation and Alzheimer's disease pathology [37]. Moreover, the miRNA bantam is found to be a potent modulator of poly-Q- and tau-associated degeneration in Drosophila [38]. Other specific miRNAs have also been linked to other neurodegenerative disorders, such as spinocerebellar ataxia type 1 (SCA1) and Huntington's disease (HD) [39], [40]. Therefore, miRNA-mediated gene regulation could be a novel mechanism, adding a new dimension to the pathogenesis of neurodegenerative disorders.
Here we show that fragile X premutation rCGG repeats can alter the expression of specific miRNAs, including miR-277, in a FXTAS Drosophila model. We demonstrate that miR-277 modulates rCGG-mediated neurodegeneration. Furthermore, we identified Drep-2, which is associated with the chromatin condensation and DNA fragmentation events of apoptosis, and Vimar, a modulator of mitochondrial function, as two of the mRNA targets regulated by miR-277. Functionally, Drep-2 and Vimar could modulate the rCGG-mediated neurodegeneration, as well. Finally, we show that hnRNP A2/B1, an rCGG repeat-binding protein, can directly regulate the expression of miR-277. These data suggest that hnRNP A2/B1 could be involved in the transcriptional regulation of selective miRNAs, and fragile X premutation rCGG repeats could alter the expression of specific miRNAs, potentially contributing to the molecular pathogenesis of FXTAS.
Given the important roles of miRNAs in neural development and human neurological disorders, we investigated the role of miRNAs in rCGG-mediated neurodegeneration. To determine whether fragile X premutation rCGG repeats could influence the expression of miRNAs, we profiled the expression of 72 known miRNAs using rCGG repeat transgenic flies that we generated previously [15]. In rCGG repeat transgenic flies, the severity of their phenotype depends on both dosage and length of the rCGG repeat. Moderate expression of (CGG)90 repeats exclusively in the eyes have an effect on morphology and histology; however, expression of (CGG)90 repeats in the neurons leads to lethality at the embryonic stage, preventing analysis at the adult stage [15]. Therefore, we used a shorter repeat length, r(CGG)60, which allowed us to examine the gene expression in adults. To analyze the effect of rCGG repeats in adult brains, we used RNAs isolated from the age- and sex-matched brains of control flies (elav-GAL4) and flies expressing rCGG60 repeats in neurons (elav-GAL4;UAS-CGG60-EGFP) for miRNA profiling experiments (Figure 1). We identified a subset of miRNAs that consistently displayed altered expression in rCGG repeat flies versus the control group. Seven miRNAs with a ≥two-fold increase and two miRNAs with expression decreased by ≥1.5-fold have been found in rCGG repeat flies. These results suggest that fragile X premutation rCGG repeats could lead to the dysregulation of a subset of specific miRNAs.
To assess the potential involvement of the miRNAs that showed altered expression in FXTAS fly brain, we examined the genetic interaction between specific miRNAs and rCGG-mediated neuronal toxicity based on the fragile X premutation rCGG repeat-mediated neurodegenerative eye phenotype we observed previously [15]. We generated UAS fly lines that could overexpress Drosophila miR-277, bantam, let-7, or miR-1, as well as bantam mutant lines (ban12 and ban20) that we generated previously [41]. We then crossed these transgenic lines with gmr-GAL4, UAS-(CGG)90-EGFP transgenic flies that exhibit photoreceptor neurodegeneration to determine the role of specific miRNAs in rCGG-mediated neurodegeneration. As shown in Figure 2, flies co-expressing miR-277 and rCGG90 consistently showed an aggravated eye phenotype, with enhanced disorganized, fused ommatidia compared with flies expressing rCGG90 alone (Figure 2B and 2D). Flies overexpressing miR-277 alone displayed a very mild rough eye phenotype (Figure 2C). Alterations of the levels of bantam, let-7, or miR-1 by either a gain of function or loss of function had no effect on rCGG-mediated neurodegeneration (Figure 2E–2N). These data together suggest that miR-277 could be involved in rCGG-mediated neurodegeneration. The role of miR-277 in rCGG-mediated neurodegeneration seems specific, since the other miRNAs we found with altered expression in the presence of fragile X premutation rCGG repeats, including bantam, let-7, and miR-1, had no effect on the rCGG90 eye phenotype. The rest of our work focused on the role of miR-277 and its potential mechanisms in modulating rCGG-mediated neurodegeneration.
Our miRNA profiling and genetic interaction studies indicated that an increase in miR-277 expression in rCGG repeat flies could alter the expression of specific cellular mRNAs by miR-277, resulting in the enhanced rCGG-induced eye phenotype. To further explore the potential regulatory effect of miR-277 on rCGG-mediated neurodegeneration, we generated a transgenic miR-277 sponge (miR-277SP) line, which could block the activity of miR-277, to test for any blocking effect on the rCGG-induced neurodegenerative eye phenotype. We generated the miRNA sponge transgenic construct as described previously [42], [43]. In brief, we placed 10 repetitive sequences complementary to miR-277 with mismatches at positions 9–12 into the 3′ UTR of EGFP in a pUASP expression vector (Figure 3A). We crossed miR-277SP transgenic flies with the flies expressing 90 CGG repeats and found that the expression of miR-277 sponge could consistently suppress rCGG-mediated neuronal toxicity (Figure 3B). MiR-277 sponge alone or scramble control sponge had no effect on eye morphology (Figure 3B and Data not shown). This result suggests that blocking the activity of miR-277 could mitigate the neurodegeneration caused by fragile X premutation rCGG repeats.
To seek the mechanisms by which miR-277 modulates rCGG-mediated neurodegeneration, we searched the RNA network and referenced TargetScanFly 5.1 to identify potential miR-277 targets [44], [45]. We selected the top candidates for miR-277 target mRNAs with the mutant alleles available for further analyses (Table 1). We then carried out a genetic screen on the rCGG90 neurodegenerative eye phenotype to identify potential miR-277 targets that could modulate rCGG-mediated neuronal toxicity. We crossed gmr-GAL4, UAS-(CGG)90-EGFP transgenic flies with fly mutants in genes coding for the top candidates for miR-277 target genes (Table 1). The progenies were then tested for potential suppression or enhancement of the disorganized eye phenotype versus flies expressing rCGG90 alone. Through this screen, we identified two modifiers of rCGG-mediated neurodegeneration, Drep-2 and Vimar. As shown in Figure 4, partial loss of Drep-2 could enhance the rCGG90 eye phenotype by increasing ommatidial disorganization (Figure 4B). Overexpression of Drep-2 could suppress the rCGG-induced eye phenotype (Figure 4C). Flies carrying the Drep-2 mutation alone or Drep-2 overexpression alone displayed normal eyes (Figure 4D and 4E). We also found a heterozygous loss-of-function mutant of Vimar that aggravated the rCGG90 eye phenotype (Figure 4F). Eyes of control flies carrying the same Vimar mutation but no rCGG90 repeats are normal (Figure 4G). These data together indicate that Drep-2 and Vimar could modulate rCGG-mediated neurodegeneration.
Since both Drep-2 and Vimar were predicted to be regulated by miR-277, by introducing 3′-UTR dual luciferase assays, we tested whether miR-277 could indeed target to Drep-2 or Vimar. We cloned the 3′-UTR of Drep-2 or Vimar containing the predicted miR-137 target sites from fly cDNA into a dual luciferase (R-Luc and F-Luc) reporter construct, allowing for the assessment of protein translation of these targets regulated via their 3′-UTR (Figure 5A). These 3′-UTR dual constructs were transfected into HEK293 cells. We found that overexpression of miR-277 could suppress the R-Luc activity at 48 h post-transfection (Figure 5B and 5C). Furthermore, when we mutated the seed regions of miR-277 located within the Drep2-3′-UTR reporter and Vimar-3′-UTR reporter, we saw that the mutation alleviated the miR-277-mediated suppression of luciferase activity (Figure 5A–5C), suggesting the action of miR-277 is specific to the miR-277 seed region within the Drep-2 3′-UTR and Vimar 3′-UTR. Together, these data demonstrate that the effect of miR-277 on Drep-2 and Vimar expression is repressive and specific. Importantly, they also suggest that Drep-2 and Vimar are the functional targets of miR-277.
Next we went on to examine the steady-state levels of Drep-2 and Vimar mRNA in rCGG repeat flies (elav-GAL4;UAS-CGG60-EGFP), in which the expression of miR-277 is increased. We saw a significant reduction of endogenous Drep-2 mRNA in rCGG repeat flies relative to control flies (elav-GAL4), whereas the Vimar mRNA expression in rCGG repeat flies remained similar to control flies (elav-GAL4) (Figure 5D). Furthermore, ectopic expression of miR-277 or miR-277 sponge could alter the endogenous mRNA level of Drep2 while have no apparent effect on Vimar mRNA level (Figure 5E). These observations suggest that miR-277 could regulate Drep-2 and Vimar mRNAs differentially, with miR-277 regulating the expression of Drep-2 mainly at the mRNA level, and Vimar via translational suppression instead.
Two rCGG repeat-binding proteins, Pur α and hnRNP A2/B1, were previously found to bind rCGG repeats directly and modulate rCGG-mediated neuronal toxicity [21], [22]. Intriguingly, recent studies have shown that multiple heterogeneous nuclear ribonucleoproteins (hnRNPs) could interact with heterochromatin protein 1 (HP1) to bind to genomic DNA and modulate heterochromatin formation [46]. Thus we tested whether hnRNP A2/B1 could interact directly with genomic regions proximal to miR-277. We performed hnRNP A2/B1-specific chromatin immunoprecipitation (ChIP) followed by real-time quantitative PCR across a six-kb region surrounding miR-277. Immunoprecipitation of chromatin chemically cross-linked to DNA with an hnRNP A2/B1-specific antibody demonstrated that a region 1.5 kb upstream of miR-277 was enriched ∼seven-fold relative to IgG control and adjacent regions (Figure 6A and 6B). Furthermore, the ectopic expression of hnRNP A2/B1 in fly brain could reduce the expression of miR-277 (Figure 6C). These results together suggest that hnRNP A2/B1 could directly bind to the upstream region of miR-277 and regulate its expression. In the presence of fragile X premutation rCGG repeats, hnRNP A2/B1 will be sequestered, leading to the de-repression of the miR-277 locus.
Fragile X-associated tremor/ataxia syndrome (FXTAS) is a neurodegenerative disorder that afflicts fragile X syndrome premutation carriers, with earlier studies pointing to FXTAS as an RNA-mediated neurodegenerative disease. Several lines of evidence suggest that rCGG premutation repeats may sequester specific RNA-binding proteins, namely Pur α, hnRNP A2/B1, and CUGBP1, and reduce their ability to perform their normal cellular functions, thereby contributing significantly to the pathology of this disorder [15], [21], [22]. The miRNA pathway has been implicated in the regulation of neuronal development and neurogenesis [32], [47]–[49]. A growing body of evidence has now revealed the role of the miRNA pathway in the molecular pathogenesis of neurodegenerative disorders [35]. Here we demonstrate that specific miRNAs can contribute to fragile X rCGG repeat-mediated neurodegeneration by post-transcriptionally regulating target mRNAs that are involved in FXTAS. We show that miR-277 plays a significant role in modulating rCGG repeat-mediated neurodegeneration. Overexpression of miR-277 enhances rCGG repeat-induced neuronal toxicity, whereas blocking miR-277 activity could suppress rCGG repeat-mediated neurodegeneration. Furthermore, we identified Drep-2 and Vimar as the functional miR-277 targets that could modulate rCGG repeat-induced neurodegeneration. Finally, we show that hnRNP A2/B1, an rCGG repeat-binding protein, can directly regulate the expression of miR-277. Our biochemical and genetic studies demonstrate a novel miRNA-mediated mechanism involving miR-277, Drep-2, and Vimar in the regulation of neuronal survival in FXTAS (Figure 7).
Several lines of evidence from studies in mouse and Drosophila models strongly support FXTAS as an RNA-mediated neurodegenerative disorder caused by excessive rCGG repeats [11], [15], [17]–[19]. The current working model is that specific RNA-binding proteins could be sequestered by overproduced rCGG repeats in FXTAS and become functionally limited, thereby contributing to the pathogenesis of this disorder [15], [17], [19], [20]. Three RNA-binding proteins are known to modulate rCGG-mediated neuronal toxicity: Pur α, hnRNP A2/B1, and CUGBP1, which bind rCGG repeats either directly (Pur α and hnRNP A2/B1) or indirectly (CUGBP1, through the interaction with hnRNP A2/B1) [21], [22]; how the depletion of these RNA-binding proteins could alter RNA metabolism and contribute to FXTAS pathogenesis has thus become the focus in the quest to understand the molecular pathogenesis of this disorder. Nevertheless, the data we present here suggest that the depletion of hnRNP A2/B1 could also directly impact the transcriptional regulation of specific loci, such as miR-277. We know that hnRNPs can interact with HP1 to bind to genomic DNA and modulate heterochromatin formation [46]. Our results indicate that hnRNP A2/B1 could participate in the transcriptional regulation of miR-277; however, it remains to be determined whether other loci could be directly regulated by hnRNP A2/B1, as well. Identifying those loci will be important to better understand how the depletion of rCGG repeat-binding proteins could lead to neuronal apoptosis.
In recent years, several classes of small regulatory RNAs have been identified in a range of tissues and in many species. In particular, miRNAs have been linked to a host of human diseases. Some evidence suggests the involvement of miRNAs in the emergence or progression of neurodegenerative diseases. For example, accumulation of nuclear aggregates that are toxic to neurons have been linked to many neurodegenerative diseases, and miRNAs are known to modulate the accumulation of the toxic proteins by regulating either their mRNAs or the mRNAs of proteins that affect their expression. Moreover, miRNAs might contribute to the pathogenesis of neurodegenerative disease downstream of the accumulation of toxic proteins by altering the expression of other proteins that promote or inhibit cell survival [35]. Our genetic modifier screen revealed that miR-277 could modulate rCGG repeat-mediated neurodegeneration. By combining our genetic screen and reporter assays, we identified Drep-2 and Vimar as the functional targets of miR-277 that could modulate rCGG-mediated neurodegeneration. The closest ortholog of miR-277 in human is miR-597 based on the seed sequence. It would be interesting to further examine the role of miR-597 in FXTAS using mammalian model systems.
Drep-2 is associated with the chromatin condensation and DNA fragmentation events of apoptosis [50], [51]. Drep-2 is one of four Drosophila DFF (DNA fragmentation factor)-related proteins. While Drep-1 is a Drosophila homolog of DFF45 that can inhibit CIDE-A mediated apoptosis [50]. Drep-2 has been shown to interact with Drep-1 and to regulate its anti-apoptotic activity [50]. Vimar is a Ral GTPase-binding protein that has been shown to regulate mitochondrial function via an increase in citrate synthase activity [52]. In the presence of fragile X premutation rCGG repeats, overexpression of miR-277 will suppress the expression of both Drep-2 and Vimar, thereby altering anti-apoptotic activity as well as mitochondrial functions, which have been linked to neuronal cell death associated with neurodegenerative disorders in general (Figure 7). Interestingly, we saw a significant reduction of Drep-2 mRNA in the flies expressing rCGG repeats, while Vimar mRNA levels remained similar to control flies. This observed difference may be due to the fact that miRNA could be involved in different modes of action, including mRNA cleavage, translational inhibition and mRNA decapping/deadenylation its target mRNAs [23], [24].
In summary, here we provide both biochemical and genetic evidence to support a role for miRNA and its selective mRNA targets in rCGG-mediated neurodegeneration. Our results suggest that sequestration of specific rCGG repeat-binding proteins can lead to aberrant expression of selective miRNAs that could modulate the pathogenesis of FXTAS by post-transcriptionally regulating the expression of specific mRNAs involved in this disorder. Identification of these miRNAs and their targets could reveal potential new targets for therapeutic interventions to treat FXTAS, as well as other neurodegenerative disorders.
All flies were maintained under standard culture conditions. The rCGG repeat transgenic flies (UAS-CGG60-EGFP and UAS-CGG90-EGFP) were generated in our lab as described previously [15]. Flies mutant in genes coding for different candidate miR-277 targets were obtained from Bloomington Stock Center. The UAS-miR-277-Sponge transgenic flies were generated as described previously [43]. We introduced 10 repetitive miR-277 sponge sequences (TGTCGTACCAGGCGTGCATTTA) with a 4-nt linker between each repeat downstream of EGFP in a pUASP expression vector. Similarly a scramble control construct (GTTCACGGATAGTGCCTGTACT) was generated as well. Both constructs were confirmed by DNA sequencing and then injected in the w1118 strain by standard methods.
Total RNAs were isolated from the control (elav-GAL4) and rCGG60 fly heads using Trizol. TaqMan MicroRNA Assays detecting 72 known individual Drosophila miRNAs were obtained from ABI (ABI). cDNA was prepared with High-Capacity cDNA Reverse Transcription Kits (ABI; Cat#437496). The 15-µl reverse transcription reactions consisted of 10 ng of total RNA, 5 U MultiScribe Reverse Transcriptase, 0.5 mM of each dNTP, 1× reverse transcription buffer, 4 U RNase inhibitor, and nuclease-free water. This was performed at 16°C for 30 min and at 42°C for 30 min, terminated at 85°C for 5 min and 4°C until use in TaqMan assays. For real-time PCR of TaqMan MicroRNA Assays, we used 0.5 ul 20×TaqMan MicroRNA Assay Primer, 1.33 ul undiluted cDNA, 5 ul 2×TaqMan Universal PCR Master Mix, 3.17 ul nuclease-free water. Each PCR reaction was performed in triplicate with MicroAmp optical 96-well plates using a 7500 Fast Real-Time PCR System (ABI), with reactions incubated at 95°C for 10 min, followed by 40 cycles of 95°C for 15 s, and 60°C for 1 min. Fluorescence readings were taken during the 60°C step. RQs were calculated using the ΔΔCt method, with 2S RNA TaqMan miRNA control assay as the endogenous control, and calibrated to the control samples.
The fly heads from control (elav-Gal4) and rCGG60 flies were collected. Trizol (Invitrogen; Cat# 15596-026) was used to isolate total RNA from each genotype. RNA samples were reverse-transcribed into cDNA with oligo(dT)20 and SuperScript III (Invitrogen; Cat#18080051). Real-time PCR was performed with gene-specific primers and Power SYBR Green PCR Master Mix (Applied Biosystems; Cat# 4367660) using the 7500 Standard Real-Time PCR System (Applied Biosystems). RpL32 (Qiagen; Cat# QT00985677) was used as an endogenous control for all samples. Primers for Drep2 and Vimar transcripts were designed using Primer Express 3.0 software (Applied Biosystems) and were as follows. Drep2: forward, 5′-TGGAACGCCTCAACTCCAA-3′; and reverse, 5′-TCGGACTCGCGATCCAA-3′. Vimar: forward, 5′-GCACCCGCCGAACAGA-3′; and reverse, 5′-TGCGATCGTAGTCTTGCGTTA -3′. All real-time PCR reactions were performed in triplicate, and RQs were calculated using the ΔΔCt method, with calibration to control samples.
Drep-2 3′-UTR and Vimar 3′-UTR sequences were PCR-amplified directly from w1118 fly brain first-strand cDNA generated from 5 ug TRIZOL-isolated total RNA using oligo-dT SuperScript III reverse transcription according to the manufacture's protocol (Invitrogen; Cat. #1808-093). The PCR products were then cloned into psiCHECK-2 dual luciferase vector (Promega; Cat# C8021). The miR-277 target sites in the Drep-2 3′-UTR and Vimar 3′-UTR were deleted using QuikChange Site-Directed Mutagenesis Kits (Stratagene; Cat. #2000518). Target sites deletions were verified by Genewiz sequencing. Briefly, 293FT cells were co-transfected by Attractene transfection reagent (Qiagen; Cat. #301005) with psiCHECK-2-3′UTR or psiCHECK-2-3′ UTRΔmiR-277 and miR-277 duplex RNA (Qiagen; Cat# MSY0000338) or control miRNA duplex (Qiagen; Cat# 1027280). All co-transfections used a total of 600 ng of plasmid DNA and 120 nmol of duplex RNA. Luciferase expression was detected using the Dual-Luciferase Reporter 1000 System (Promega; Cat# E1980) according to the manufacturer's instructions. At 48 h after transfection, R-Luc activity was normalized to F-Luc activity to account for variation in transfection efficiencies, and miR-277-mediated knockdown of R-Luc activity was calculated as the ratio of normalized R-Luc activity in the miR-277 duplex treatments to normalized R-Luc activity in the negative control duplex treatments. Luciferase experiments were repeated three times.
ChIP was performed using a ChIP Assay Kit (Millipore). S2 cells were cross-linked with 1% formaldehyde (Sigma-Aldrich) for 10 min at room temperature. Chromatin was fragmented to an average size of 500 bp by sonication (Sonicator 3000; Misonix) and immunoprecipitated with anti-Flag M2 antibody (sigma). Immunoprecipitated and purified DNA fragments were diluted to 1 ng/µl in nuclease-free water. We used 8 ng of DNA in 20-µl SYBR Green real-time PCR reactions consisting of 1× Power SYBR Green Master Mix and 0.5 µM forward and reverse primers. Reactions were run on an SDS 7500 Fast Instrument (Applied Biosystems). Primers were designed using Primer Express 3.0 software (Applied Biosystems) and were as follows. 4.5 kb upstream: forward, 5′-CAGAAAACAGGCGTGCAAAC; and reverse, 5′-GAATTTGCATTGGCTTTGGAA. 3.5 kb upstream: forward, 5′- TTACAATTGGATGGGCTTCGT; and reverse, 5′-AAGCTGACGGCCTGACTAAAAA. 2.5 kb upstream: forward, 5′-GTTGGCTGCTGCGTCAATT; and reverse, 5′- GCCCCAGCGGCATTTATA. 1.5 kb upstream: forward, 5′- TTCTGGCACTGGCAGCTTT; and reverse, 5′- CATCGTGCTGGCCAACAC. 1.0 kb upstream: forward, 5′-TGTACGGGCATGTGTATGCA; and reverse, 5′- TCAACGAACACGCTGCGTAT. 0.5 kb upstream: forward, 5′- GGGCATTTTCATTTCATTCCA; and reverse, 5′- CGGGCAGCGTAATTTAAGCT. 0.5 kb downstream: forward, 5′- CGCCCACAAGAGCTTTTGA; and reverse, 5′- TTTCCACGGTATGCTGCTTTT. 1.5 kb downstream: forward, 5′- CGTTTCCATTTAGTTGGATTTTTGT; and reverse, 5′- GGCAAACCACACATTTTAACATACA. DNA relative enrichment was determined by taking the absolute quantity ratios of specific IPs to nonspecific IPs (normal mouse IgG only), IP/IgG, and normalizing to control (pUAST only). Independent chromatins were prepared for all ChIP experiments, and real-time PCR reactions were performed in triplicate for each sample on each amplicon.
For scanning electron microscopy (SEM) images, whole flies were dehydrated in gradient concentration ethanol (25%, 50%, 75%, 100%), dried with hexamethyldisilazane (Sigma; Cat# 16700), and analyzed with an ISI DS-130 LaB6 SEM/STEM microscope.
|
10.1371/journal.pgen.1007316 | ZMYND10 stabilizes intermediate chain proteins in the cytoplasmic pre-assembly of dynein arms | Zinc finger MYND-type-containing 10 (ZMYND10), a cytoplasmic protein expressed in ciliated cells, causes primary ciliary dyskinesia (PCD) when mutated; however, its function is poorly understood. Therefore, in this study, we examined the roles of ZMYND10 using Zmynd10–/–mice exhibiting typical PCD phenotypes, including hydrocephalus and laterality defects. In these mutants, morphology, the number of motile cilia, and the 9+2 axoneme structure were normal; however, inner and outer dynein arms (IDA and ODA, respectively) were absent. ZMYND10 interacted with ODA components and proteins, including LRRC6, DYX1C1, and C21ORF59, implicated in the cytoplasmic pre-assembly of DAs, whose levels were significantly reduced in Zmynd10–/–mice. LRRC6 and DNAI1 were more stable when co-expressed with ZYMND10 than when expressed alone. DNAI2, which did not interact with ZMYND10, was not stabilized by co-expression with ZMYND10 alone, but was stabilized by co-expression with DNAI1 and ZMYND10, suggesting that ZMYND10 stabilized DNAI1, which subsequently stabilized DNAI2. Together, these results demonstrated that ZMYND10 regulated the early stage of DA cytoplasmic pre-assembly by stabilizing DNAI1.
| Dynein arm defects are linked to primary ciliary dyskinesia (PCD). ZMYND10 increased the stability of its interacting proteins and specifically regulated intermediate chain protein assembly, revealing tightly regulated mechanisms underlying dynein arm assembly and PCD-related pathogenesis. Increasing protein stability could be useful for developing PCD therapeutics.
| Primary ciliary dyskinesia (PCD) is an autosomal-recessive disorder caused by defective motile cilia or flagella that is characterized by respiratory distress, impaired mucociliary clearance, chronic cough, sinusitis, bronchiectasis, male infertility, laterality defects, and cardiac anomalies in term neonates [1, 2]. To date, mutations in about 30 genes have been linked to PCD in approximately 50–70% of cases [3].
In motile cilia or flagella, the outer dynein arm (ODA) and inner dynein arm (IDA) are attached to the peripheral microtubules of the 9+2 axoneme with a fixed periodicity and generate ATP-dependent motion. Dynein arms are large, multisubunit protein complexes comprised of light, intermediate, and heavy chains [4]. The latter have ATPase activity, which provides power for the sliding between microtubules in the beating cilia. The capacity of the dynein arm to function as molecular motors depends on the integrity of its components. ODA consists of two to three heavy chains, two or more intermediate chains, and a cluster of four to eight light chains, whereas IDA has a more diverse composition [4, 5]. As such, several genes linked to PCD encode dynein arm components, including dynein axonemal light chain 1 (DNAL1), dynein axonemal intermediate chain 1 (DNAI1), DNAI2, dynein axonemal heavy chain 5 (DNAH5), and DNAH11 [6–10]. Mutations in DNAH6, a heavy chain of IDA, were detected in individuals with heterotaxy and ciliary dysfunction [11].
Dynein arms are pre-assembled in the cytoplasm and transported into motile cilia, where they are docked onto peripheral microtubules; however, the underlying mechanisms are poorly understood [12]. Dynein axonemal assembly factors (DNAAFs) are involved in the pre-assembly of dynein arms, and their mutation is linked to PCD [13–16]. There are five known DNAAFs—leucine-rich repeat-containing 50 (LRRC50, DNAAF1) [13], kintoun (KTU, DNAAF2) [14], DNAAF3 [15], dyslexia susceptibility 1 candidate 1 (DYX1C1, DNAAF4) [16], and HEAT repeat-containing protein 2 (HEATR2, DNAAF5) [17]. KTU and DYX1C1 interact with chaperone proteins, including heat shock protein 70 (HSP70), HSP90, and T-complex chaperonin complex [14, 16]. DNAAFs work with a chaperone complex to facilitate the proper folding of heavy chains and their assembly with intermediate chains [15]. Defects in DNAAFs result in the loss of ODAs and IDAs from axonemes [13–16].
In addition to DNAAFs, mutations in several cytoplasmic proteins, including LRRC6 [18, 19], zinc finger MYND-type-containing 10 (ZMYND10) [20, 21], chromosome 21 open reading frame 59 (C21ORF59) [22], PIH1 domain-containing 3 (PIH1D3) [23, 24], and armadillo repeat-containing 4 (ARMC4) [25], have been identified as those causing PCD when defective. Mutations in LRRC6, ZMYND10, C21ORF59, and PIH1D3 cause ODA and IDA defects [18–24], whereas those in ARMC4 cause ODA defects [25]. Thus, these proteins likely function at different stages of pre-assembly or have ODA- or IDA-specific roles. Overall, these proteins are expected to be involved in the pre-assembly of dynein arms based on their cytoplasmic localization and consequences upon loss of expression. However, the specific functions of these proteins and their relationships with DNAAFs are not well understood.
ZMYND10 (also known as BLU) has a myeloid, nervy, and DEAF-1 (MYND)-type zinc finger domain at its C-terminus that engages in protein-protein interactions [26]. ZMYND10 is highly enriched in ciliated cells compared with that in nonciliated cells [27] and is expressed in motile ciliated tissues in mice [21]. ZMYND10 has been shown to interact with LRRC6 [20, 21], although its function in motile ciliated cells is not known. Therefore, in this study, we generated and characterized Zmynd10−/− mice and found that they recapitulate phenotypic aspects of human PCD, including the absence of ODA and IDA without defects in ciliogenesis or 9+2 axonemal structure. We also found that the levels of DNAI1 and DNAI2 in ODA were reduced in Zmynd10−/− mice. ZYMND10 binds to and stabilizes DNAI1, thereby facilitating the assembly of intermediate chains. Our data suggested that ZMYND10 may play a role in the pre-assembly of dynein arms by regulating the expression of dynein arm components at the protein, and not the mRNA level and promoting their assembly into a cytoplasmic protein complex.
To investigate the function of ZMYND10, we generated mice with targeted deletion of the Zmynd10 gene locus (S1A and S1B Fig) using a lacZ-containing targeting cassette (Zmynd10tm1[KOMP]Wtsi). β-Galactosidase activity staining of Zmynd10+/− lung tissues on postnatal day (P)1 revealed Zmynd10 expression in the bronchus and bronchioles, but not in the alveoli (S1C and S1D Fig). This expression pattern was consistent with that in the previous in situ hybridization results in mouse lung tissues [21]. Zmynd10 expression was also observed in the testes of Zmynd10+/− mice at P28 as well as in the spermatids and earlier-stage germ cells (S1E and S1F Fig). Deletion of coding exons 2–11 yielded a Zmynd10-null allele (Zmynd10−/−), and western blotting and immunofluorescence analyses of the testis lysates and tracheal tissue, respectively, confirmed the absence of ZMYND10 (S2 Fig).
Zmynd10−/− mouse litters conformed to Mendelian ratios, and neonates showed no gross abnormalities, indicating that loss of Zmynd10 did not cause embryonic lethality; however, mutant mice exhibited growth retardation and were visibly smaller at P10, with all eventually dying within 1 month of birth (S3A–S3C Fig), with a mean survival of 14 days. Zmynd10−/− mice developed hydrocephaly and subsequent abnormal head morphology, with complete penetrance (Fig 1A and 1B), and the cerebral ventricles were dilated with cortical tissue thinning (Fig 1C–1F). In addition, 42% of the Zmynd10−/− mice showed laterality defects, including reversal of the heart apex, stomach, liver, or spleen (S3E and S3F Fig). Alcian Blue staining of the paranasal cavities revealed mucosal congestion in Zmynd10−/− mice but not in wild-type littermates, suggesting defective mucociliary clearance (Fig 1G and 1H), which is a prominent manifestation of PCD leading to recurrent airway infection. Inflammation in the lung tissue was never observed in Zmynd10−/− mice that died before P20; however, some of the mice that survived past P25 showed severe pulmonary inflammation in mutants, as evidenced by loss of alveolar architecture, thickening of alveolar septae, collapse of the alveolar space, and infiltration of inflammatory cells and fibroblasts (Fig 1I–1J and S3D Fig). Fibrosis was not observed in Masson trichrome staining (Fig 1K and 1L). Taken together, these data indicated that loss of Zmynd10 induced defects consistent with PCD.
Since the Zmynd10−/− phenotype suggested defects in motile cilia, we examined their ultrastructure by transmission electron microscopy (TEM). The tracheal cilia and basal bodies were abundant in the tracheal and lung epithelial cells in both Zmynd10+/+ and Zmynd10−/− mice (Fig 2A–2D and S4 Fig); however, in the latter, some areas of the tracheal epithelium were surrounded by cellular debris and mucus (S4 Fig). Analysis of the tracheal cilia cross-sections revealed a typical 9+2 microtubular structure in both wild-type and Zmynd10−/− mice (Fig 2E and 2F). In contrast, cilia in Zmynd10−/− mice lacked both ODA and IDA structures (Fig 2H), which were present in the peripheral microtubules of Zmynd10+/+ mice (Fig 2G). This is in agreement with the previous results of studies on humans subjects with ZMYND10 mutations who lacked ODA and IDA in the respiratory epithelium [20, 21]. The observed defects in the ciliary structure resulted in a loss of ciliary motility and beating in the ventricles of the Zmynd10−/− brain (S1 Movie and S2 Movie). However, ciliogenesis itself was normal, as demonstrated by the observation that the numbers of cilia and basal bodies were not different between Zmynd10+/+ and Zmynd10−/− mice (Fig 2I and 2J).
The subcellular distribution of the ODA intermediate chain protein DNAI2 in mouse tracheal epithelial cells (mTECs) cultured at the air-liquid interface (ALI) for 14 days was examined by immunofluorescence microscopy. In mTECs from Zmynd10+/+ mice, DNAI2 was detected in both cilia and cytoplasm and was found to colocalize with the ciliary marker acetylated α-tubulin (Fig 2K and S5A Fig). In contrast, DNAI2 was only weakly expressed in the cytoplasm of mTECs from Zmynd10−/− mice and did not colocalize with acetylated α-tubulin, suggesting that DNAI2 was significantly decreased and failed to translocate to motile cilia in the absence of Zmynd10 (Fig 2L and S5B Fig). Thus, the phenotypes observed in Zmynd10−/− mice resulted from a ciliary motility defect associated with the loss of axonemal ODA and IDA.
The mRNA levels of DNAH5 and DNALI1—ODA heavy and IDA light-intermediate chain proteins, respectively—were downregulated by ZMYND10 knockdown in cultured human tracheal epithelial cells [20]. This was examined in Zmynd10+/+ and Zmynd10−/− mice by transcriptional profiling/RNA sequencing using total RNA extracted from the testis, lung, and brain tissues (S1 Table). The transcript levels of ODA and IDA components in mutants were similar to those in wild-type mice (Fig 3), suggesting that ZMYND10 did not regulate the transcription of dynein arm components. Gene ontology analysis showed that genes involved in muscle function were significantly upregulated, whereas those involved in ion transport were reduced in Zmynd10-/- mice compared with those in Zmynd10+/+ mice (S6 Fig).
We and others previously reported that ZMYND10 interacts with LRRC6 [20, 21]. Given that both proteins are cytoplasmic, we examined whether they interacted with other cytoplasmic proteins that are known to be defective in PCD by co-immunoprecipitation in human embryonic kidney (HEK) 293T cells. We also obtained potential interactors from a web-based protein-protein interaction database, PrePPI (https://honiglab.c2b2.columbia.edu/PrePPI/index_old.html). ZMYND10 was found to interact with C21ORF59 and DYX1C1 (DNAAF4), IQ motif and ubiquitin domain-containing protein (IQUB), Tctex1 domain-containing D1 (TCTEX1D1), and DNAI1, but not with DNAI2 (S7 Fig). TCTEX1D1 is a dynein light chain of ODA [5], and one of the intermediate chain proteins of ODA, DNAI1, interacts with ZMYND10. Interactions with LRRC6 and REPTIN, which are essential for cilia motility [28], as well as interactions with C21ORF59, DNAI1, and IQUB were confirmed by glutathione S-transferase (GST) pulldown assays using lysates from mTECs (Fig 4A). In addition, we found that ZMYND10 interacted with heat shock cognate protein 70 (HSC70), a constitutively expressed molecular chaperone (S8 Fig), suggesting that ZMYND10 played a role in the folding and assembly of dynein arms through cooperation with HSC70. These protein-protein interactions are illustrated in Fig 4B. REPTIN and C21ORF59 expression was also examined in mTECs at ALI 14 by immunofluorescence microscopy. REPTIN and C21ORF59 signals were reduced in Zmynd10−/− mTECs (Fig 4C–4E and S9 Fig). Interestingly, C21ORF59 was recently shown to interact with LRRC6 and Dishevelled (Dvl) and is implicated in planar cell polarity as well as the correct localization of ODAs to motile cilia [29]. Our results demonstrated that ZMYND10 interacted with cytoplasmic proteins associated with PCD and that some interaction partners were downregulated in the absence of Zmynd10.
Our results in mTECs suggested that protein levels of dynein arm components (Fig 2K and 2L) and their interaction partners (Fig 4C–4E) were altered in Zmynd10−/− mice. To investigate this in detail, immunoblotting was carried out using the testis lysates. The levels of proteins that interacted with ZMYND10, such as C21ORF59, IQUB, and LRRC6, as well as dynein arm subunits, including DNAI1, DNAI2, and DNAH7, a heavy chain of inner dynein arm, were downregulated in mutant mice (Fig 5A and 5B). This was also confirmed by immunofluorescence analysis of the mouse tracheal tissue. DNAH5, DNAI2, and IQUB signals were weaker in Zmynd10−/− mice than in Zmynd10+/+ mice (Fig 5C–5E). A previous study in individuals with a homozygous truncating ZMYND10 mutation showed that DNAI2 was absent from bronchial epithelial cells, whereas DNAH5 and DNALI1 remained in the cytoplasm [30]. There were no differences in the mRNA levels of dynein arm components between Zmynd10+/+ and Zmynd10−/− mice (Fig 3). These data suggested that ODA and IDA were unstable, likely due to improper assembly in the absence of Zmynd10. We speculated whether the decrease in protein levels of ZMYND10-interacting factors resulted from dysregulated transcription in Zmynd10−/− mice. To investigate this possibility, we examined the transcript levels of these factors and of DNAAFs, which are involved in dynein arm assembly [3], by RNA sequencing. There were no differences in the mRNA levels of these proteins between Zmynd10+/+ and Zmynd10−/− mice (Fig 5F–5H). These results suggested that ZMYND10 stabilizes dynein arm components and their interaction partners at the protein level, but not at the transcript level.
To determine whether ZMYND10 regulated its interaction partners and dynein arms at the protein level, we investigated the effects of ZMYND10 on the stability of LRRC6, DNAI1, and DNAI2 in a heterologous system. HEK 293T cells overexpressing LRRC6 and/or ZMYND10 were treated with cycloheximide (100 μg/mL) for up to 48 h to block protein synthesis (Fig 6A). Only 7.8% of LRRC6 remained after 48 h of treatment. However, this value was increased to 44.4% when LRRC6 was co-expressed with ZMYND10 (Fig 6B), suggesting that ZMYND10 prevented LRRC6 degradation. Similarly, the amount of DNAI1 protein was increased from 30.9% to 64.1% by co-expressing ZMYND10 (Fig 6C and 6D), but not by co-expressing ZMYND10-p.Gln366* mutant protein, which lacked the MYND domain and was identified in a PCD patient [20] (S10 Fig), indicating that the MYND domain was necessary for the stabilizing effect. In contrast, the stability of DNAI2, which did not interact with ZMYND10 (S7K and S7L Fig), was unaffected by ZMYND10 overexpression (S11 Fig). Given that both DNAI1 and DNAI2 levels were downregulated in Zmynd10−/− mice, we speculated that ZMYND10 stabilized DNAI1, which in turn stabilized DNAI2. To evaluate this possibility, we compared the protein levels of DNAI2 upon co-expression with DNAI1 without or with ZMYND10. DNAI2 was more stable in the presence of both DNAI1 and ZMYND10 than in the presence of DNAI1 alone (Fig 6E and 6F, and S12 Fig). These results demonstrated that ZMYND10 stabilized some of its interaction partners at the protein level and modulated the pre-assembly of intermediate chains by stabilizing DNAI1.
In this study, we generated and characterized Zmynd10−/− mice as a model for human PCD. The mice exhibited loss of ciliary motility and ODA and IDA components without disruption of ciliogenesis, thereby recapitulating the phenotypes associated with ZMYND10 mutations in humans and serving as an appropriate model to study PCD pathogenesis.
The assembly of dynein arms into motile cilia is a complex process involving many regulatory factors, including DNAAFs and chaperones, that contribute to the stabilization, folding, and pre-assembly of dynein arm components [15, 24, 31] into multiprotein complexes that undergo intraflagellar transport into the axoneme for attachment to peripheral microtubules [31]. In Chlamydomonas, dynein arms exist as intermediate-chain/heavy-chain (IC-HC), light chain, and docking complexes [12]. During ODA pre-assembly, HCs, such as DNAH5 or DNAH11, are attached to ICs, DNAI1 (IC1), and DNAI2 (IC2). This process is facilitated by DNAAF1, DNAAF2, and DNAAF4, while DNAAF3 is proposed to act during the final stages of chaperone dissociation [15]. IC-HC assembly fails in the absence of the IC subunit [12]. Biochemical analyses of Zmynd10−/− mice showed that DNAI1 and DNAI2 are downregulated, suggesting that ZMYND10 stabilizes these two proteins or mediates their assembly. Given that formation of the IC complex precedes IC-HC assembly, the reduced levels of DNAI1 and DNAI2 may account for the observed decrease in an ODA HC, DNAH5.
In this study, we demonstrated that ZMYND10 formed a cytoplasmic protein network comprised of LRRC6, C21ORF59, DYX1C1, IQUB, REPTIN, and HSC70. The interaction between LRRC6 and REPTIN is essential for cilia motility in zebrafish, although this function is independent of its known role as a transcriptional regulator [28]. Similarly, although ZMYND10 binds to REPTIN, the expression levels of various dynein arm components and interactors of ZMYND10 were not diminished in Zmynd10−/− mice, ruling out transcriptional regulation of these factors as a mechanism underlying ODA and IDA defects. We previously showed that DNAH5 and DNALI1 mRNAs were downregulated in human tracheal epithelial cells in which an shRNA targeting ZMYND10 was delivered by lentivirus. This discrepancy between in vivo mouse and cell line data may be due to artefacts resulting from lentiviral vector integration [32]. C21ORF59 interacts with LRRC6, DNAAF1, and Dvl to regulate polarization as well as ciliary motility [29]. DYX1C1 interacts with KTU, HSP70, HSP90, and T-complex chaperonin [14, 16], whereas REPTIN interacts with PIH1D1 [33], which contains a PIH (protein interacting with HSP90) domain implicated in the pre-assembly of dynein arms [34]. PIH1D3 interacts with KTU, DYX1C1, and HSP90 [23, 24] and is implicated in the formation of the IC complex, as evidenced by its interaction with DNAI2 and the downregulation of DNAI1 and DNAI2 in Pih1d3−/− mouse sperm [24, 35]. ZMYND10 is functionally similar to PIH1D3 in that both proteins interact with DYX1C1 and heat shock proteins; moreover, ICs are reduced in mice lacking Zmynd10 or Pih1d3 [35]. ZMYND10 interacts with HSC70, a member of the HSP70 family. HSC70 is involved in diverse cellular processes, including protein folding and protein degradation, and exerts its chaperone activity by cooperation with cochaperones and by binding to nascent or unfolded polypeptides through the substrate binding domain in an ATP-dependent manner [36]. Therefore, it is possible that ZYMND10 affects the stability of DNAI1 through cooperation with HSC70.
Currently, there is no curative therapy for PCD. For PCD resulting from a defective dynein arm component, the component should be replaced with a normal one. However, this will be challenging considering the huge size of some dynein components. For example, the coding region of DNAH5, which is most frequently mutated in PCD [17], is about 15.6 kb, encoding a protein of 529 kDa. In this regard, PCD resulting from defects in DNAAFs or other cytoplasmic proteins is different in that dynein arm components are not compromised, but their cytoplasmic assembly or trafficking is defective. In this study, we demonstrated that the protein levels of DNAI1 and DNAI2 were reduced in Zmynd10−/− mice due to the decreased stability of DNAI1 in the absence of ZYMND10. Therefore, increasing protein stability of DNAI1 can be considered as a potential treatment.
In conclusion, our results demonstrated that several cytoplasmic proteins, including ZMYND10, formed a protein network in motile ciliated cells that, in conjunction with chaperone proteins, modulated various aspects of dynein arm pre-assembly. ZMYND10 specifically functioned in the early steps of this process by regulating DNAI1 stability or folding, thereby controlling IC assembly (S13 Fig). These findings provide insights into the molecular mechanisms involved in dynein arm assembly and the pathogenic basis for PCD-associated defects. This will also help to develop pharmacological interventions for PCD caused by defects in the cytoplasmic nonaxonemal components of motile cilia.
The animal protocol was reviewed and approved by the Institutional Animal Care and Use Committee of University of Michigan (#08619), Boston Children's Hospital (#13-01-2283), and Yonsei University College of Medicine (#2015–0178). All animals were handled in accordance with the Guidelines for the Care and Use of Laboratory Animals.
Targeted Zmynd10tm1(KOMP)Wtsi embryonic stem cells were obtained from the Knockout Mouse Project Repository and injected into blastocysts. Chimeric mice were bred with C57BL/6J mice to establish germline transmission. Wild-type littermates were used as controls for Zmynd10−/− mice. Genotyping was performed by standard PCR using the primers Zmynd10-ex2F (5′-TGGAGGAGCTTGGAACTGAC-3′), Zmynd10-ex2R (5′-GGAGGCAGACACAGTTAGGC-3′), and CSD-RAF5-F (5′-ACACCTCCCCCTGAACCTGAAA-3′, SR1 (5′-TGCTTTATTGTGCGAAAGGAAGAGGG-3′).
P1 or P28 mice were sacrificed, and the testes and lungs were dissected. After three washes with phosphate-buffered saline (PBS), the tissues were fixed in 4% paraformaldehyde (PFA)/0.02% Nonidet (N)P-40 for 2 h at room temperature and permeabilized with 0.02% NP-40 in PBS for 1 h. Samples were incubated overnight at 37°C in X-gal staining solution composed of 5 mM K3Fe(CN)6, 5 mM K4Fe(CN)6, 2 mM MgCl2, 0.01% sodium deoxycholate, 0.02% NP-40, and 1 mg/mL X-gal in PBS. They were then washed three times with PBS for 5 min each and post-fixed with 4% PFA for 24 h before embedding within paraffin. Sections (10 μm thick) were deparaffinized and rehydrated through a graded series of ethanol concentrations followed by counterstaining with Nuclear Fast Red (Vector Laboratories, Burlingame, CA, USA).
The lung and snout tissue specimens were fixed using 10% formalin for 24 h. The tissues were sectioned (5 μm thickness) and stained with hematoxylin and eosin, or periodic acid-Schiff for histological examination.
The tracheas of Zmynd10+/+ and Zmynd10−/− mice at P14 were dissected and fixed using 2.5% glutaraldehyde, 1.25% PFA, and 0.03% picric acid in 0.1 M sodium cacodylate buffer (pH 7.4) overnight at 4°C. Samples were then processed for TEM analysis using standard techniques.
P16 mice were deeply anesthetized and then decapitated. The brain was rapidly removed and immersed in ice-cold Dulbecco’s modified Eagle’s medium (DMEM; Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS; Sigma-Aldrich, St. Louis, MO, USA). Sagittal sections of 150-μm thickness were cut using a vibratome (VT1200S; Leica, Wetzlar, Germany). Sections from the third ventricle were visualized on an Axio Observer A1 microscope using a 63× phase contrast objective lens (LD Plan-Neofluor 0.75 Corr Ph2 M27; Carl Zeiss, Jena, Germany) equipped with a high-speed charge-coupled device camera (optiMOS sCMOS; QImaging, Surrey, BC, Canada). Movies were acquired at 100 frames/s.
A polyclonal antibody recognizing the C-terminal sequence (amino acids 339–362, DRLERENKGKWQAIAKHQLQHVFS) of mouse ZMYND10 was recovered from rabbits injected with the corresponding antigen (AbFrontier, Seoul, Korea). LRRC6 and DNAH5 antibodies were previously described [37, 38]. Antibodies against DNAI2 (H00064446-M01; Abnova, Taipei, Taiwan); REPTIN (ab89942; Abcam, Cambridge, UK); DNAH7 (NBP1-93613) and DNAI1 (SAB4501181; both from Novus Biologicals, Littleton, CO, USA); IQUB (HPA020621) and TCTEX1D1 (HPA028420; both from Sigma-Aldrich); acetylated α-tubulin (T7451 from Sigma-Aldrich and 5335S from Cell Signaling Technology, Danvers, MA, USA); FLAG (#8146) and Myc (#2276; both from Cell Signaling Technology); and C21ORF59 (sc-365792) and β-actin (sc-1615; both from Santa Cruz Biotechnology) were purchased from commercial sources. Secondary antibodies were purchased from Invitrogen and Santa Cruz Biotechnology for immunofluorescence and immunoblotting analyses, respectively.
mTECs grown on inserts were fixed using 4% PFA for 10 min and permeabilized with 0.1% Triton X-100 for 20 min at room temperature. The tracheal tissue was fixed in 4% paraformaldehyde overnight at 4°C, embedded in a paraffin blocks, and cut into 5-μm-thick sections. The sections were then mounted on slides, deparaffinized, and rehydrated through a graded series of ethanol concentrations. After rehydration, antigen retrieval was performed by boiling sections for 30 min using a Retrieve-All Antigen unmasking system 1 (pH 8; BioLegend, San Diego, CA, USA). Sections were permeabilized with 1% sodium dodecyl sulfate for 10 min at room temperature. mTECs and trachea samples incubated in blocking buffer containing 10% donkey serum and 1% bovine serum albumin for 1 h at room temperature. Samples were incubated overnight at 4°C with primary antibodies diluted in blocking buffer. After washes with PBS, samples were incubated with secondary antibodies and 4′,6-diamidino-2-phenylindole for 30 min at room temperature, washed, and covered with mounting medium and cover slips. Images were acquired using an SP5X laser scanning microscope (Leica) or LSM 700 microscope (Carl Zeiss).
mTECs were isolated from Zmynd10+/+ and Zmynd10−/− mice at P14 as previously described [39]. Briefly, cells were isolated by overnight digestion with pronase (Roche Diagnostics, Indianapolis, IN, USA) at 4°C and then separated from contaminating fibroblasts by incubation in mTEC basal medium on a Primaria cell culture plate (Corning Inc., Corning, NY, USA) for 3–4 h. mTECs were seeded in collagen-coated apical chambers of transwell permeable supports (0.4-μm polyester membrane; Corning Inc.). Proliferation medium was applied to the apical and basal chambers of the wells, and cells were cultured at 37°C in 5% CO2 [40]. For ALI culture conditions, the medium was removed from the apical chamber when mTECs became confluent, and differentiation medium was added to the basal chamber.
Total RNA was isolated from the brain (P14), lung (P14), and testis (P21) tissues obtained from Zmynd10+/+ and Zmynd10−/− mice using a Qiagen RNA extraction kit (Qiagen, Valencia, CA, USA). RNA sequencing was performed by Theragen Etex (Suwon, Korea). Libraries were constructed with a TruSeq RNA Library Sample Prep kit (Illumina, San Diego, CA, USA), and the enriched library was sequenced on an Illumina HiSeq 2500 system. Sequence reads were mapped against the mouse reference genome (NCBI GRCm38/mm10) and analyzed using CLC Genomics Workbench v.9.0.1 software (CLC Bio, Cambridge, MA, USA).
HEK 293T cells were maintained in DMEM supplemented with 10% FBS and penicillin (50 IU/mL)/streptomycin (50 μg/mL). The cells were transfected with plasmids using Lipofectamine PLUS reagent (Invitrogen).
Experiments were performed as previously described [41]. Immunoblotting was quantified by densitometry using ImageJ software (National Institutes of Health, Bethesda, MD, USA). Immunoprecipitation was performed using EZview Red anti-FLAG M2 or anti-c-Myc affinity gels (Sigma-Aldrich). Pulldown assays with GST-ZMYND10 and GST-MYND were performed as previously described [20].
Cycloheximide chase was used to assess the stability of LRRC6, DNAI1, and DNAI2. HEK293T cells were transfected with Myc-tagged LRRC6, DNAI1, or DNAI2 with or without FLAG-tagged ZMYND10; at 24 h post-transfection, cells were treated with 100 μg/mL cycloheximide (C4859; Sigma-Aldrich) to inhibit new protein synthesis. Cells were harvested at predetermined time points, and LRRC6, DNAI1, and DNAI2 levels were detected by western blotting.
Results are presented as means ± standard errors or standard deviations for the indicated number of experiments. Statistical analysis of continuous data was performed with two-tailed Student’s t-test or one-way analysis of variance, with Dunnet’s, Bonferroni, or Dunn post hoc test, as appropriate. Results with P values of less than 0.05 were considered statistically significant.
|
10.1371/journal.pgen.1002653 | Mutations in MITF and PAX3 Cause “Splashed White” and Other White Spotting Phenotypes in Horses | During fetal development neural-crest-derived melanoblasts migrate across the entire body surface and differentiate into melanocytes, the pigment-producing cells. Alterations in this precisely regulated process can lead to white spotting patterns. White spotting patterns in horses are a complex trait with a large phenotypic variance ranging from minimal white markings up to completely white horses. The “splashed white” pattern is primarily characterized by an extremely large blaze, often accompanied by extended white markings at the distal limbs and blue eyes. Some, but not all, splashed white horses are deaf. We analyzed a Quarter Horse family segregating for the splashed white coat color. Genome-wide linkage analysis in 31 horses gave a positive LOD score of 1.6 in a region on chromosome 6 containing the PAX3 gene. However, the linkage data were not in agreement with a monogenic inheritance of a single fully penetrant mutation. We sequenced the PAX3 gene and identified a missense mutation in some, but not all, splashed white Quarter Horses. Genome-wide association analysis indicated a potential second signal near MITF. We therefore sequenced the MITF gene and found a 10 bp insertion in the melanocyte-specific promoter. The MITF promoter variant was present in some splashed white Quarter Horses from the studied family, but also in splashed white horses from other horse breeds. Finally, we identified two additional non-synonymous mutations in the MITF gene in unrelated horses with white spotting phenotypes. Thus, several independent mutations in MITF and PAX3 together with known variants in the EDNRB and KIT genes explain a large proportion of horses with the more extreme white spotting phenotypes.
| White spotting coat color phenotypes are the result of aberrations in the development of melanocytes. The analysis of domestic animals with heritable white spotting phenotypes thus helps to better understand the complicated genetic network controlling the proliferation, migration, differentiation, and survival of pigment producing cells. We analyzed the so-called splashed white phenotype in horses, which is characterized by a very distinctive large blaze, extended white markings on the legs, and blue eyes. Splashed white horses are also frequently deaf. However, the phenotype is quite variable and, in some horses with minimal expression, the splashed white phenotype cannot be unambiguously discriminated from the “common” white markings. We studied horses from various breeds and found one mutation in the PAX3 gene and two mutations in the MITF gene that cause the splashed white phenotype. A third mutation in the MITF gene, which we found in a single Franches-Montagnes horse, causes a new coat color phenotype, termed macchiato. Similar mutations in humans cause either Waardenburg or Tietz syndrome, which both are characterized by pigmentation defects and a predisposition for deafness. Our study reveals the molecular basis for a significant proportion of white spotting phenotypes that are intermediate between completely unpigmented horses and common white markings.
| Coat color is a well-studied model trait for geneticists. Coat color phenotypes are relatively easy to record, which facilitates their analysis. In mammals melanocytes cover the entire body surface and are responsible for the pigmentation of skin, hairs, and eyes. Melanocytes are formed during fetal development from melanoblasts, which originate in the neural crest and migrate across the developing fetus in order to reach their final position on the body [1]. This developmental program requires a delicate level of regulation to ensure that the correct amount of cells reaches their final destination [2]–[5]. An over-proliferation of cells that have left their surrounding tissue might have fatal consequences for the developing fetus [6]. If however too few of the migrating melanoblasts survive, this will lead to partially or completely unpigmented phenotypes [7]. Domestic animals with such unpigmented phenotypes have been highly valued due to their striking appearance and have often been actively selected in animal breeding. Consequently, our modern domestic animals provide a large repertoire of spontaneous mutants that allow the dissection of the contributions of individual genes to migration, proliferation, differentiation, and survival of melanocytes.
From a genetic point of view, white spotting is considered a complex trait [8]. Phenotypes range from tiny white spots at the extremities of the body to large unpigmented areas, in symmetrical or asymmetrical patterns, up to completely unpigmented animals [9]. Different combinations of alleles at several loci can interact and it is generally impossible to precisely predict the genotype of an animal based only on a given white spotting phenotype.
Splashed white is a distinctive but nevertheless quite variable white spotting pattern in horses, which is primarily characterized by extensive depigmentation of the head. Splashed white horses often have blue eyes and they are sometimes deaf [9]. This phenotype has not yet been characterized at the molecular level.
Mutations in two genes have been identified to cause pronounced depigmentation phenotypes in horses thus far. A total of 19 different mutant alleles at or near the KIT gene were reported to cause either completely white horses or horses with pronounced depigmentation, such as the dominant white, tobiano, and sabino-1 spotting patterns [10]–[14]. The frame overo spotting pattern, which is phenotypically overlapping with the splashed white phenotype, is caused by a missense mutation in the EDNRB gene [15]–[17]. EDNRB mutations are also found in a small fraction of human patients with Waardenburg syndrome (WS). WS is characterized by pigmentation abnormalities, such as a typical white forelock or white skin patches, strikingly blue or heterochromatic irises, and varying degrees of sensorineural deafness and skeletal dysmorphologies. Other forms of human Waardenburg syndrome are caused by mutations in EDN3, MITF, PAX3, SNAI2, and SOX10 [18]. Mouse mutants for these genes are available and show largely similar although not completely identical phenotypes [19].
In this study we report the identification of several independent mutations in horses with striking depigmentation phenotypes that show parallels to human Waardenburg syndrome.
We obtained samples from a large Quarter Horse family segregating for a striking white spotting pattern termed splashed white (Figure 1). In our material this phenotype was characterized by a large blaze of variable size and shape. Many splashed white horses had an extremely large blaze extending over the eyes or even covering parts of the cheeks; such a head pattern is termed baldface. Splashed white horses also typically had high markings that extend high up on their legs and occasionally little white belly spots. Most of the splashed white horses had blue eyes or iris heterochromia. Although many splashed white horses have a very characteristic appearance, the phenotype is variable and in some splashed white horses the unpigmented areas are so small that we could not reliably distinguish these horses from horses with other subtle depigmentation phenotypes. Some, but not all of the sampled splashed white horses were deaf.
In this Quarter Horse family the splashed white coat color appeared to be inherited as an autosomal dominant trait. We therefore genotyped 24 cases and 7 controls from this family on the equine 50k SNP array and performed a linkage analysis (Figure S1). Parametric analysis with a fully dominant model of inheritance gave a maximum LOD score of 1.6 on ECA 6 with a corresponding maximum α of 0.78. These results suggested locus heterogeneity within the family. We also performed a case-control genome-wide association study (GWAS) with the SNP genotypes and found the strongest signal again on ECA 6 (praw = 5.6×10−7; pgenome = 0.003). The chromosomes with the next best associations were 11 and 16, each having SNPs associated at praw = 6.6×10−5, which is not genome-wide significant after permutation analysis (pgenome = 0.34).
The linked and associated region on ECA 6 contained PAX3 as a strong functional candidate gene. We sequenced the nine exons of the PAX3 gene in two cases and two controls and identified seven polymorphisms including one missense mutation, PAX3:c.209G>A (Table 1, Table S1). We then genotyped all remaining animals of the Quarter Horse family and other unrelated horses for this variant. We identified a total of 29 splashed white horses carrying the variant A-allele in heterozygous state. All these horses traced back to a female Quarter Horse born in 1987, whose genomic DNA from a hair-root sample tested homozygous wild-type. Thus, the mutation most likely arose in the germline of this animal. We did not detect the variant A-allele in 21 solid-colored Quarter Horses nor in 112 horses from 7 other breeds. We also did not find any horse with the homozygous variant A/A genotype (Table 2).
The PAX3:c.209G>A variant is predicted to result in a non-conservative amino acid exchange p.C70Y in the so-called paired domain, which together with the homeobox domain mediates the DNA-binding of the transcription factor PAX3. The wild-type cysteine at this position is conserved across all known PAX paralogs in animals including Drosophila and Caenorhabditis elegans (Figure 2). Based on the X-ray structure of the paired domain of the human PAX6 protein, the side-chain of this cysteine is predicted to form a direct contact to the DNA backbone [20].
The initial GWAS indicated potential secondary signals on ECA 11 and 16. We did not detect an obvious candidate gene on ECA 11. However, the MITF gene on ECA 16 represents a strong functional candidate for a white spotting phenotype. As the PAX3:p.C70Y variant did not explain the phenotype in all splashed white horses from the Quarter Horse family, we sequenced all exons and the known promoter elements from the MITF gene in two splashed white horses that did not have the PAX3C70Y allele and in two controls. We identified a total of 28 polymorphisms (Table S1). A variant in the proximal melanocyte-specific MITF promoter replacing a thymine with 11 nucleotides stood out as clear candidate for a non-coding regulatory mutation (ECA16:g.20,117,302Tdelins11; Figure 3). This variant interrupts a highly conserved binding site for PAX3 [21]–[23] and is located close to the mutation causing white spotting in dogs [24]. The variant allele was absent from 21 solid-colored Quarter Horses and from 112 horses with minimal white markings from 7 additional breeds. We found 19 splashed white horses that carried only the MITFprom1 allele and 20 splashed white horses that carried both the MITFprom1 and PAX3C70Y allele (Table 2).
We quantitatively analyzed the proportion of depigmented skin in the face area of splashed white Quarter Horses in relation to their underlying MITF and PAX3 genotypes as well as their base coat color (Figure S2, Figure S3). This analysis indicated that one copy of either the MITFprom1 or the PAX3C70Y allele has a similar effect on the face pigmentation. The presence of both the MITFprom1 and the PAX3C70Y alleles leads to a slightly more pronounced depigmentation of the face on average than either splashed white allele alone. This analysis also showed that the average white face area is more extended in splashed white chestnut horses compared to bay horses with the same splashed white genotype (Figure 4; Table S2).
We sequenced the coding regions and proximal promoter elements of the functional candidate genes KIT, MITF, and PAX3 in other horses with white spotting phenotypes in additional horses from various breeds. We then noticed that the MITF promoter polymorphism described above is much more widespread across breeds than we had originally hypothesized. We found the insertion allele in 58 Quarter Horses and American Paint Horses with either the splashed white or more pronounced other white spotting phenotypes (see below). We also found this allele in five Trakehner horses with white spotting phenotypes. A detailed pedigree analysis revealed that all these horses are descendants from the Thoroughbred stallion Blair Athol born in 1861. This stallion also had a very large white blaze (Figure S4). The MITFprom1 allele was also present in a Miniature Horse, a Shetland Pony, and 11 Icelandic Horses with either splashed white or more pronounced other white spotting phenotypes. Thus, the MITFprom1 allele is probably several hundred years old and arose before the foundation of the modern horse breeds.
In our study we defined horses with ≥20% white face area and ≤10% white area on the body as “splashed white horses”. If a horse had >10% white body area, it was considered to have a more pronounced “other white spotting” phenotype. We noticed that several horses with more pronounced “other white spotting phenotypes” were offspring from two “splashed white” parents.
We identified a total of 24 horses from 5 breeds that were homozygous for the MITF promoter variant. All these MITFprom1/prom1horses showed very pronounced but still variable depigmentation phenotypes. They had at least a white belly in addition to white legs and the white head (Figure 5A–5C). One of these horses had only small pigmented areas along the dorsal midline (Figure 5A). One horse homozygous for the MITFprom1 allele and heterozygous for the PAX3C70Y allele was completely white and also deaf (Figure 5D).
In our mutation analysis we identified one horse with a splashed white phenotype including a white belly, which was wild-type for both the MITFprom1 and the PAX3C70Y alleles (Figure 5E). This horse carried a small deletion in exon 5 of the MITF gene (c.837_841del5). The variant leads to a frameshift and a severely truncated MITF protein (p.C280Sfs*20), which might act in a dominant negative fashion. A completely white offspring of this horse was a compound heterozygote for the MITFprom1 and the MITFC280Sfs*20 alleles (Figure 4F). The MITFC280Sfs*20 allele was absent from 21 solid Quarter Horses and 112 control horses from 7 additional breeds (Table 2).
In a few unrelated horses with large blazes we did not find any candidate causative mutations in the KIT, MITF, or PAX3 candidate genes. These horses comprised 7 Quarter Horses, 1 Hanoverian, 2 Oldenburger, and 8 Thoroughbreds (Table 2).
In 2008 a colt with a striking white-spotting phenotype was born out of two solid-colored bay Franches-Montagnes parents. The coat color resembled a combination of white-spotting and coat color dilution (Figure 6). The parentage of this colt was experimentally verified using 13 microsatellite markers. Therefore, we assumed the coat color of this colt to be the result of a spontaneous de novo mutation and subsequently termed it “macchiato”. In 2010, we performed a detailed clinical and spermatological examination, which revealed that the two year old macchiato stallion was deaf and had a low progressive sperm motility.
We sequenced the coding regions of six functional candidate genes in the macchiato stallion and his solid-colored parents. The investigated candidate genes are involved in white spotting (KIT, MITF) or coat color dilution (MLPH, PMEL, SLC36A1, and SLC45A2). We identified a de novo missense mutation in exon 6 of the MITF gene in the macchiato stallion (c.929A>G). We found the mutant allele at approximately 50% intensity in DNA samples from blood, hair roots and sperm, indicating that the macchiato stallion is not a mosaic. The mutant allele was not present in DNA from blood samples from either the mother or the father. The mutation affects a highly conserved amino acid of the basic DNA binding motif of the transcription factor MITF (p.N310S). We analyzed the DNA binding properties of the mutant MITF in an electrophoretic mobility shift assay and found that DNA binding activity was reduced by about 80%, probably through changes to the kinetics of binding (Figure 7, Table S3). A mutation at the same residue in the human MITF protein leads to Tietz syndrome, which is characterized by a more generalized depigmentation and profound obligate hearing loss compared to the slightly milder Waardenburg syndrome 2A, which is caused by many other mutations in the human MITF gene [25]. The MITF:p.N310S variant was absent from 96 solid-colored Franches-Montagnes horses.
As the splashed white phenotype is rather distinctive among the many different variations of white spotting phenotypes in horses, we started this investigation with the expectation of finding only a single causative mutation. However, the linkage/GWAS data suggested locus heterogeneity even within a family of closely related Quarter Horses. Using a positional candidate gene approach we subsequently identified three causative mutations for splashed white phenotypes and one causative mutation for another similar phenotype that we termed macchiato (Table 1).
Although we do not have functional proof for all of the mutations, we obtained sufficient ancillary evidence to claim their causality. For the PAX3C70Y allele the arguments are that it (1) occurs only in horses with the splashed white phenotype, (2) the PAX3 locus was linked to this phenotype in a family, (3) mutations within the PAX3 gene cause similar phenotypes in humans and mice, and (4) the mutation affects a highly conserved cysteine residue in the paired domain, that is predicted to participate in DNA binding. We were able to determine that this variant arose in 1987 and to identify the specific founder animal.
For the three MITF mutations the support is even better than for the PAX3 mutation: (1) Multiple mutations within the same gene lead to comparable phenotypes, (2) the three variants occur exclusively in horses with characteristic depigmentation phenotypes, and (3) mutations within the MITF gene cause similar phenotypes in humans, mice, and dogs [18], [19], [24]. MITFC280Sfs*20 is the result of a frameshift mutation, which is extremely likely to affect the normal function of MITF as it truncates about half of the protein. The MITFprom1 variant affects a region of the melanocyte-specific promoter that has been shown to function as a PAX3 binding site in humans and mice [20]–[22]. For MITFN310S we demonstrated that it represents a spontaneous de novo mutation in the macchiato horse that was born out of solid-colored parents free of this allele. We also provide functional evidence that the MITFN310S protein has a reduced DNA binding capability. Taken together these data strongly argue for the causality of the four mutations in PAX3 and MITF that we report in this study.
Our findings will be of relevance to horse breeders. The MITFprom1 allele arose at least several hundred years ago, it is relatively common, and it occurs in several modern horse breeds. Horses homozygous for this allele are viable and typically have a more pronounced depigmentation phenotype than heterozygous horses. The PAX3C70Y allele is only 24 years old and occurs exclusively in Quarter and Paint Horses. We did not find a horse homozygous for this mutation, and based on data from mice it is unlikely that a homozygous PAX3C70Y/C70Y horse would be viable. As PAX3 is required for several key steps in neural development, homozygosity for this allele will most likely result in embryonic or fetal lethality [26]. Therefore, the mating of two heterozygous PAX3+/C70Y horses is not recommended in order to avoid the accidental production of an embryo homozygous for this allele. The MITFC280Sfs*20 and MITFN310S alleles are extremely rare. Data from mice again suggest that these alleles will most likely result in severe clinical phenotypes such as e.g. microphthalmia in the homozygous state [27]. Therefore, horses with such alleles should also not be mated to each other.
Depigmentation or white spotting phenotypes represent a complex trait, ranging from horses with a wild-type phenotype (no unpigmented skin areas) up to completely white horses. Previously, a number of single gene mutations in KIT and EDNRB have been shown to cause either dominant white or extreme white spotting phenotypes [10]–[17, Figure S5]. With this study we now report a series of mutations that lead to milder depigmentation phenotypes, which phenotypically overlap with the common white markings seen at the head and legs of many domestic horses. The newly reported mutations interact with other genetic factors. Chestnut horses carrying a mutation in MITF and/or PAX3 show a more pronounced depigmentation phenotype than bay horses with the same mutations. Thus, the basic coat color or more specifically the genotype at MC1R modifies the phenotypic expression of the reported mutations. The correlation between chestnut coat color and increased size of white markings has been reported before [28], [29]. The detailed analysis of depigmentation phenotypes may help us to better understand complex genetic networks on a molecular and functional level. In such networks, non-coding regulatory mutations such as MITFprom1 will probably play a major role and need to be taken into consideration.
In conclusion, we report four different mutations leading to splashed white or the new macchiato coat color phenotype in horses, which show similarities to the human Waardenburg or Tietz syndromes, respectively. This study highlights the potential of coat color genetics to gain molecular insights into complex regulatory gene networks.
All animal work was conducted in accordance with the relevant local guidelines (Swiss law on animal protection and welfare - permit to the Swiss National Stud Farm no. 2227). The only animal experiment in our study was the collection of blood samples from horses by certified veterinarians.
We analyzed a total of 239 horses, 106 animals with white spotting-phenotypes (cases) and 133 solid-colored controls. The cases consisted of 70 horses with the splashed white phenotype and another 36 horses with more extreme white spotting phenotypes (77 Quarter or American Paint Horses, 1 Hanoverian, 11 Icelandic Horses, 1 Miniature Horse, 2 Oldenburger, 1 Shetland Pony, 8 Thoroughbreds, 5 Trakehner). The 133 solid-colored controls consisted of 21 Quarter Horses, 1 American Standardbred, 96 Franches-Montagnes horses, 1 Haflinger, 4 Icelandic Horses, 7 Noriker, 1 Trakehner, and 2 European Warmblood horses. We considered Quarter Horses and American Paint Horses to represent one joint population as several horses in our study had double registrations and as it is quite common that offspring of Quarter Horse parents with white spotting phenotypes are registered as American Paint Horses.
We considered the proportion of white face area as the primary phenotypic criterion. To quantify this proportion we drew the heads of the horses in a standard perspective from photos (Figure S2, Figure S3). We then measured the proportion of unpigmented face area in comparison to a horse that we considered to have a 100% white face. Horses with ≥20% white face area were considered to have a white spotting phenotype. If a horse had ≥20% white face area and ≤10% white area on the body, it was considered “splashed white”. If a horse had ≥20% white face area and >10% white body area, it was considered to have an “other white spotting” phenotype. Horses with ≤3% white face area and 0% white body area were considered as solid-colored controls. Horses with a white face area ranging from 3–20% were classified as unknown phenotype.
We isolated genomic DNA from either EDTA blood or hair root samples from all horses. We genotyped 31 horses with the illumina equine 50K SNP beadchip containing 54,602 SNPs. We used the PLINK v1.07 software for pruning of the genotype data set [30]. We removed 28,381 SNPs that did not have genotype calls in every animal and 10,753 SNPs that had minor allele frequencies below 5%. For the final analysis 31 horses and 19,319 SNPs remained. We used the Merlin software [31] and a fully dominant model of inheritance to analyze the data for parametric linkage. We used the PLINK software for genome-wide association analyses. Empirical genome-wide significance levels were determined by performing 100,000 random permutations of the assigned phenotypes.
We amplified exons, flanking intron regions, and proximal promoter sequences of the KIT, MITF, and PAX3 genes as well as exon 2 of the EDNRB gene (Table S4). We subsequently sequenced the PCR products on an ABI 3730 capillary sequencer (Life Technologies). We analyzed sequencing data with the Sequencher 4.9 software for polymorphisms (Gene Codes). In the macchiato horse and its parents we sequenced all exons of the KIT, MITF, MLPH, PMEL, SLC36A1, and SLC45A2 genes.
We expressed recombinant wild-type and N310S MITF from in E. coli Rosetta 2 cells in LB medium supplemented with 1% glucose (amino acid residues 112–207 from uniprot acc. Q95MD1). We lysed the cells by sonication in 20 mM Tris, 500 mM NaCl, 50 mM imidazole, 0.5 mM phenylmethylsulfonyl fluoride (PMSF), 0.1% (v/v) β-mercaptoethanol, pH 9. We applied the soluble fraction to Ni-NTA resin and eluted MITF with an imidazole step gradient (0.2–0.6 M) and further purified the proteins by reverse phase HPLC on a C18 column, using a gradient of 5–95% acetonitrile (0.1% TFA) over 20 ml at 1 ml/min. We lyophilised the protein-containing fractions and stored them at −20°C. The protein was refolded by resuspension in 20 mM Tris (pH 7.4), and we confirmed the correct folding by circular dichroism (CD) spectropolarimetry.
We performed the EMSA using an M-Box containing oligonucleotide (5′-GGAAAGTTAGTCATGTGCTTTTCAGAAGA-3′) as previously described [32]. The EMSA reactions contained 20 mM Tris, 50 mM NaCl, 5 mM MgCl2, 1 mM DTT, and 33 µg/ml BSA at pH 7.4. After incubation on ice for 30 min, we separated the samples on a 6% (w/v) non-denaturing polyacrylamide gel, in 0.5× TBE (45 mM Tris, 45 mM boric acid, 2.5 mM EDTA, pH 8.3) and analyzed the gels using a PhosphorImager (Molecular Dynamics). We quantified the bands from three repeated experiments and calculated average values and standard deviations.
|
10.1371/journal.ppat.1005099 | Microbial Regulation of p53 Tumor Suppressor | p53 tumor suppressor has been identified as a protein interacting with the large T antigen produced by simian vacuolating virus 40 (SV40). Subsequent research on p53 inhibition by SV40 and other tumor viruses has not only helped to gain a better understanding of viral biology, but also shaped our knowledge of human tumorigenesis. Recent studies have found, however, that inhibition of p53 is not strictly in the realm of viruses. Some bacterial pathogens also actively inhibit p53 protein and induce its degradation, resulting in alteration of cellular stress responses. This phenomenon was initially characterized in gastric epithelial cells infected with Helicobacter pylori, a bacterial pathogen that commonly infects the human stomach and is strongly linked to gastric cancer. Besides H. pylori, a number of other bacterial species were recently discovered to inhibit p53. These findings provide novel insights into host–bacteria interactions and tumorigenesis associated with bacterial infections.
| This review focuses on a novel aspect of host–bacteria interactions: the direct interplay between bacterial pathogens and tumor suppression mechanisms that protect the host from cancer development. Recent studies revealed that various pathogenic bacteria actively inhibit the major tumor suppression pathway mediated by p53 protein that plays a key role in the regulation of multiple cellular stress responses and prevention of cancerogenesis. Bacterial degradation of p53 was first discovered in the context of Helicobacter pylori infection, which is currently the strongest known risk factor for adenocarcinoma of the stomach. This phenomenon, however, is not limited to H. pylori, and many other bacterial pathogens inhibit p53 using various mechanisms. Inhibition of p53 by bacteria is linked to bacterial modulation of the host cellular responses to DNA damage, metabolic stress, and, potentially, other stressors. This is a dynamic area of research that will continue to evolve and make important contributions to a better understanding of host–microbe interactions and tumorigenesis. These studies may offer new molecular targets and opportunities for drug development.
| p53 protein has been receiving significant attention for more than 30 years. This interest originates from the protein’s prominent role in tumor suppression that was eloquently paraphrased in the scientific literature as “the guardian of the genome” [1]. p53 is a key component of the cellular mechanisms controlling cellular responses to various cellular stresses, including DNA damage, aberrant oncogene activation, loss of normal cell–cell contacts, nutrient deprivation, and abnormal reactive oxygen species (ROS) production. Following cellular stresses, p53 is activated and primarily functions as a transcriptional regulator of expression of multiple effector proteins and miRNAs, which, in turn, regulate key cellular processes such as apoptosis, cellular proliferation, and autophagy. Since regulation of cellular stress responses is tightly intertwined with metabolic regulation, there is an interplay between p53 and multiple pathways involved in regulation of metabolism and cellular homeostasis that is complex and not fully understood. One prominent example is a reciprocal signaling between p53 and mTOR [2]. The latter pathway plays a key role in cell growth and proliferation. p53 is also directly involved in regulation of the cellular energy metabolism and the redox balance regulating glycolysis, oxidative phosphorylation, and the pentose phosphate pathway (PPP). Through multiple mechanisms, p53 can dampen glycolysis and the PPP and promote oxidative phosphorylation. The metabolic functions of p53 are likely to significantly contribute to its tumor suppression activity (Fig 1).
Inactivation of p53 is a hallmark of tumorigenic changes. More than half of all tumors carry p53 mutations, rendering the p53 gene (tp53) the most mutated gene in human tumors. p53 can also be inhibited by mutation-independent mechanisms. Inhibition of wild-type p53 by the SV40 virus was one of the first reported examples. SV40 is a small DNA tumor polyomavirus that induces cellular transformation in cell culture and an array of different tumors in animals. In infected cells, viral protein (SV40 large T antigen [T-Ag]) binds p53 and inhibits p53-dependent transcription, resulting in accumulation of inactivated p53 protein [3,4]. Inhibition of p53 by large T-Ag is closely linked to the ability of the SV40 virus to induce tumorigenic transformation; SV40 mutants, which are defective in inhibition of p53, are also defective in cellular immortalization and transformation [5,6].
p53 by itself was originally identified as a protein binding to SV40 large T-Ag [7,8]. Later studies have shown that SV40 T-Ag is not unique in this sense, and other small tumor DNA viruses (adenoviruses and papillomaviruses) also produce similar proteins (E1B-55K and E6) that interact with p53 [9,10]. Although adenoviral protein E1B-55K and human papillomavirus (HPV) protein E6 are different in their amino acid sequences, they converge at the same function, forming protein complexes with p53 to inhibit its activity. HPV and adenovirus (Ad) can also induce ubiquitination and proteasomal degradation of p53 [11]. The ability to degrade p53 varies among viruses. For example, high-risk genital HPV types 16 and 18, which cause around 70% of cervical cancers, efficiently degrade p53, while low-risk viruses such as HPV types 6 and 11 are unable to do so [12,13]. Similarly, p53 is degraded by human adenovirus serotypes 12 and 5 (Ad12, Ad5), while Ad9 and Ad11 do not have this ability [14,15]. To degrade p53, both HPV and Ad use the host protein degradation machinery. HPV E6 protein interacts with the host E3 ubiquitin ligase, E6AP, causing its substrate specificity to be altered so that it ubiquitinates p53 and induces its degradation by the 26S proteasomes [16]. In Ad-infected cells, viral proteins E1B-55K and E4orf6 interact with cellular proteins Cullin5 (or Cullin2), Rbx1, and Elongins B and C to form a Cullin-containing E3 ubiquitin ligase that targets p53 for proteasomal degradation [14,17,18]. A similar degradation strategy is also used by the Epstein–Barr virus (EBV), which forms a complex containing viral protein BZLF1 and cellular Cullin2/5-containing E3 ubiquitin ligase to degrade p53 [19].
Due to a relatively simple organization of the viral genomes, viruses have to rely on host resources for most aspects of their life cycle. In the process of interacting with host cells, they alter the intracellular environment to make it suitable for viral replication. These drastic alterations, however, may cause cellular stress and activate p53, resulting in cell cycle arrest or apoptosis of host cells; both outcomes are detrimental to viral replication. It is plausible that inhibiting p53 may provide advantages to viruses that have evolved to do so. Recently, this concept was further expanded to include additional microorganisms. These novel data are discussed in this review, focusing on specific mechanisms of bacterial inhibition of p53.
Recent studies have found that it is not only viruses, but also some pathogenic bacteria, that actively inhibit p53 and induce its degradation. This phenomenon was initially described in gastric cells co-cultured with Helicobacter pylori [20]. H. pylori is a gram-negative, spiral-shaped pathogen that lives in the stomachs of approximately half of the world’s population. The infection is typically acquired during childhood and causes lifelong chronic infection. Because of the association between H. pylori infection and the incidence of gastric cancer, the International Agency for Research on Cancer (IARC) has classified this bacterium as a Group 1 carcinogen. H. pylori infection is considered to be the strongest known risk factor for gastric cancer, and epidemiological studies have estimated that, in the absence of H. pylori, 75% of gastric cancers would not occur [21].
Pathogenesis associated with H. pylori infection is determined by interactions between bacterial factors and host cells. The most well characterized bacterial virulence determinants are the vacuolating cytotoxin A (vacA) and the cag pathogenicity island (cag PAI). The cag PAI is a 40 kb region of DNA that encodes a type IV secretion system (T4SS) that forms a syringe-like pilus structure used for the injection of a bacterial protein CagA (cytotoxin-associated gene A) into gastric cells. Following the delivery, intracellular CagA is localized to the plasma membrane and triggers complex alterations of the host signaling pathways [22], including activation of cellular oncogenes (Fig 2). CagA itself functions as an oncoprotein. In laboratory tests, CagA promoted anchorage-independent growth and, when transgenically expressed in mice, led to spontaneous development of gastrointestinal and hematopoietic neoplasms [23,24]. Oncogenic potential of CagA has also been demonstrated using Drosophila and zebrafish experimental models [25,26].
H. pylori infection results in conditions of cellular stress because the bacteria induce DNA damage and disturb normal cellular homeostasis (including aberrant activation of multiple oncogenic pathways), all of which are conditions that typically activate p53 [27,28]. However, initial studies of the p53 stress response revealed that H. pylori is able to dampen activity of p53 protein by inducing its rapid degradation [20]. The ability of H. pylori to suppress the p53 response was also demonstrated when DNA damage was experimentally induced by DNA-damaging agents [20,29,30]. The bacteria specifically target p53, as p73—another member of the p53 protein family, which has significant functional and structural similarities to p53—is not down-regulated by H. pylori but rather induced [31]. The ability to induce degradation of p53 varies between H. pylori strains, with CagA-positive bacteria being more potent [20,29]. Although CagA likely does not directly bind to p53, it induces its degradation [29]. Notably, ectopic transfection of CagA is sufficient to inhibit p53 activity and induce its degradation [20,30]. Recent studies pointed out a complex nature of CagA–p53 interactions. It was shown that levels and natural variability of CagA protein highly affect p53 degradation [32]. Among other bacterial factors, VacA was also reported to regulate p53 [33–35]. Down-regulation of p53 was found to facilitate autophagy in infected cells [35].
The kinetics of p53 in infected cells in vivo appears to be complex. In infected Mongolian gerbils, which are commonly used for studies of H. pylori infection, expression of p53 was changed in a bimodal fashion, with an accumulation after initial infection that was followed by a rapid down-regulation of p53 protein in gastric epithelial cells. A second peak of p53 was observed later, when gastritis (inflammation of the lining of the stomach) developed. These findings led to a hypothesis that, at a certain time, levels of p53 reflect a balance between p53 degradation induced by the bacteria and p53 induction caused by cellular stress [20]. A down-regulation of p53 protein, but not p53 mRNA, was observed in H. pylori-infected mice [36].
In contrast to small DNA tumor viruses, H. pylori takes advantage of host mechanisms normally regulating p53 [20,35]. The bacteria enhance proteasomal degradation of p53 mediated by E3 ubiquitin ligase HDM2 by increasing its phosphorylation at serine 166. An increased phosphorylation of HDM2 was found in gastric epithelial cells co-cultured with H. pylori in vitro and H. pylori-infected animals and humans in vivo [20,35,37]. Inhibition of HDM2 activity with siRNA or chemical inhibitor Nutlin3 suppresses bacterial degradation of p53 [20,35,38]. A similar effect can be achieved by inhibition of Akt and Erk kinases, showing that these enzymes mediate phosphorylation of HDM2 protein in infected cells [35,38]. Expression of HDM2 was found to correlate with phosphorylated Akt (pAkt) in patients infected with H. pylori [37]. In addition to HDM2, recent studies reported that another cellular E3 ubiquitin ligase, Mule/ARF-BP1, is involved in degradation of p53 in H. pylori-infected cells [32]. It remains unclear how this enzyme is activated by the bacteria.
p14ARF tumor suppressor (termed p19ARF in rodents and p14ARF in humans), which functions upstream of p53, was found to be a critical modulator of p53 protein stability in infected cells [32], as ARF inhibits activities of both HDM2 and ARF-BP1 proteins [39–41]. It was shown that cells expressing functional ARF are significantly more resistant to degradation of p53 (Fig 3). However, when ARF protein levels are decreased due to hypermethylation or deletion of the ink4a/ARF locus, H. pylori efficiently degrades p53 [32]. Loss of ARF occurs during gastric tumorigenesis and can be found in gastric precancerous lesions. Methylation of the p14ARF gene is also increased with age [42]. Given these findings, it was hypothesized that older people with gastric precancerous lesions, who are infected with H. pylori, may be particularly vulnerable to degradation of p53 [32].
Among other cellular factors, ASPP2 protein (apoptosis-stimulating protein of p53), which normally activates p53, was identified to regulate p53 in H. pylori-infected cells [29]. Buti et al. showed that binding of CagA protein to ASPP2 results in inhibition of transcriptional and proapoptotic activities of p53 and induction of proteasomal degradation of p53.
Recent studies suggest that bacterial degradation of p53 may contribute to gastric tumorigenesis. It was reported that clinical isolates of H. pylori varied greatly in their ability to degrade p53, but that, generally, isolates associated with a higher gastric cancer risk more strongly affect p53 when compared to low-risk counterparts [32].
H. pylori inhibits p53 through multiple mechanisms, implying that inhibition of p53 activity is an important factor for successful infection. The bacteria not only induce degradation of p53, but also alter the expression profile of p53 isoforms [43]. Interaction of H. pylori with gastric epithelial cells, mediated via the cag PAI, induces N-terminally truncated Δ133p53 and Δ160p53 isoforms, which inhibit transcriptional and proapoptotic activities of p53, resulting in activation of NFkB. Induction of proinflammatory cytokine Macrophage Migration Inhibitory Factor (MIF) by H. pylori was suggested to inhibit p53 by decreasing its phosphorylation [44]. It was also shown that H. pylori can facilitate mutagenesis of the p53 gene. Infection with H. pylori leads to aberrant induction of activation-induced cytidine deaminase (AID), which deaminates cytosine residues, leading to accumulation of p53 mutations in gastric tissues [45]. Interestingly, AID and other cytidine deaminases are induced by a number of viruses such as HPV, HTLV-1, HCV, and others [46–48]. SV40 and influenza A viruses have been shown to affect expression of p53 isoforms [49,50].
A new and exciting development in this area is that other bacteria induce degradation of p53 using a similar mechanism to that of H. pylori (Fig 4). Two research groups have recently reported that the intracellular bacterial pathogen Chlamydia trachomatis, and potentially other Chlamydia species, induces degradation of p53 by activating HDM2 protein [51,52]. C. trachomatis is a common cause of bacterial sexually transmitted disease (STD) and blinding trachoma. Similar to H. pylori, C. trachomatis activates the PI3K/Akt pathway and increases phosphorylation of HDM2 (Ser166), leading to activation of HDM2 and proteasomal degradation of p53. Down-regulation of p53 allows Chlamydia to enhance activity of the PPP that provides bacteria with necessary metabolites, such as nucleotides precursors, and protects against oxidative stress by increasing the cellular NADPH pool [52]. Enforced expression of p53 in infected cells results in strong inhibition of chlamydial growth, while overexpression of glucose-6-P-dehydrogenase, a key enzyme in the PPP that is inhibited by p53, rescues the bacterial growth. The authors reported that degradation of p53 by Chlamydia interferes with the host’s response to genotoxic stress and may contribute to cancerogenesis in the female genital tract [51,52].
Inhibition of p53 through the HDM2-dependent mechanism is also employed by enteropathogen Shigella flexneri, which causes bacillary dysentery in humans. Infection with Shigella is accompanied by strong genotoxic stress and cellular damage [53]. To prevent activation of p53, Shigella causes rapid degradation of p53 using two distinct mechanisms. During the early phase of infection, the bacterial virulence effector IpgD promotes activation of the host PI3K/Akt pathway and phosphorylation of HDM2 at serines 166 and 186, causing activation of HDM2 and degradation of p53. The second mechanism for p53 inhibition comes into play during the late phase of infection. p53 is proteolytically cleaved by the calpain protease system, in which activation is facilitated by the Shigella virulence effector VirA. The VirA activates calpain by promoting proteolysis of the calpain inhibitor calpastatin. Bergounioux et al. suggested that Shigella inhibits p53 to prevent apoptotic cells death that saves energy and preserves its own epithelial niche [53]. Interestingly, not all enteric pathogens inhibit p53. Activation of p53 was reported in the context of Salmonella typhimurium infection [54]. Outside the Enterobacteriaceae family, down-regulation of p53 protein was reported in studies of Neisseria gonorrhoeae, which is responsible for the sexually transmitted gonorrhea that may increase the risk of genital neoplasms [55]. Similar to the aforementioned pathogens, N. gonorrhoeae causes strong genotoxic stress and induces both single and double strand DNA breaks. The mechanism of p53 down-regulation is not fully understood, but Vielfort et al. reported that the bacteria can inhibit transcription of the p53 gene [56].
Inhibition of p53 may provide certain benefits to bacteria. One particular mechanism that may be targeted by bacteria is the p53 DNA damage response. Inhibition of p53 may allow bacteria to subvert the host cell cycle control and apoptosis mechanisms, resulting in inhibition of cell death and survival of host cells damaged by infection. This is in agreement with the findings of antiapoptotic and prosurvival effects produced by bacterial pathogens, which inhibit p53 [20,29,52,53]. In the case of H. pylori, expression of the CagA virulence factor is sufficient to inhibit p53 and extend short and long term survival of gastric epithelial cells that underwent DNA damage [20]. Besides the DNA damage response, bacteria may also target the metabolic control of p53. Inhibition of the p53 metabolic regulation may be particularly important for obligatory intracellular pathogens such as Chlamydia. As described above, degradation of p53 allows C. trachomatis to release inhibition of the PPP elicited by p53. When bacterial degradation of p53 was experimentally inhibited, the development and formation of infectious progeny was blocked, suggesting that metabolic control of p53 provides antibacterial protection. It is possible to draw a parallel between Chlamydiae and viruses since both are obligatory intracellular pathogens, which strictly rely on the host resources. Similar to viruses, inhibition of p53 allows Chlamydia to reprogram the host cell signaling to create a metabolic environment necessary for chlamydial survival and growth. To some extent, this may also be applied to obligate parasitic Mycoplasma bacteria, which inhibit activity of p53 [57]. A more complex picture emerges in regards to the role of the p53 signaling in the context of chronic infections with extracellular pathogens such as H. pylori. One proposed possibility is that inhibition of p53 helps H. pylori to compromise the gastric epithelial barrier, allowing the bacteria to acquire nutrients from the host or get access to the lamina propria. This concept is supported by recent findings showing that H. pylori inhibits activation of p53 induced by disruption of the adherens junctions, which stabilize cell–cell adhesion [38]. It was also suggested that suppression of p53 responses may help H. pylori adapt during the early phase of infection and prevent the host immune response [20]. The p53 pathway is known to affect immune response [58]. Among direct transcription targets of p53 are a number of proteins regulating innate immunity and cytokine and chemokine production. p53 is also known to affect NF-κB activity and pro-inflammatory signaling. Although immunomodulatory function may play a role, there is no direct evidence yet that bacterial inhibition of p53 affects the host immune response. Additional studies are needed to further explore these mechanisms.
Interaction of bacterial pathogens with the host cells induces DNA
damage, alters intracellular signaling, and profoundly affects normal cellular homeostasis. To prevent the cellular stress response, which may be detrimental to a successful infection, some bacteria have evolved to inhibit p53, a key component of the stress response machinery. Bacteria inhibit p53 through multiple mechanisms, including protein degradation, transcriptional inhibition, and post-translational modifications. Current research revealed that p53 has a role in controlling the bacterial infections and that inhibition of p53 may confer certain selective advantages to bacteria. Unfortunately, this may have grave consequences for the hosts, increasing the risk of tumor development. It is particularly relevant to prolonged chronic infections. Initial experiments with inhibition of protein degradation of p53 demonstrate that p53 activities can be restored in infected cells using specific chemical inhibitors. These findings may offer new and exciting opportunities for therapeutic targeting of p53 in infected cells. Future studies of the bacterial regulation of p53 hold the promise of a better understanding of pathogenesis and tumorigenesis associated with bacterial infections.
|
10.1371/journal.pntd.0007231 | A computational method for the identification of Dengue, Zika and Chikungunya virus species and genotypes | In recent years, an increasing number of outbreaks of Dengue, Chikungunya and Zika viruses have been reported in Asia and the Americas. Monitoring virus genotype diversity is crucial to understand the emergence and spread of outbreaks, both aspects that are vital to develop effective prevention and treatment strategies. Hence, we developed an efficient method to classify virus sequences with respect to their species and sub-species (i.e. serotype and/or genotype). This tool provides an easy-to-use software implementation of this new method and was validated on a large dataset assessing the classification performance with respect to whole-genome sequences and partial-genome sequences. Available online: http://krisp.org.za/tools.php.
| Dengue (DENV), Chikungunya (CHIKV) and Zika (ZIKV) are considered major public health challenges. In addition to the epidemic caused by DENV, which has been described in many tropical countries, the introduction of CHIKV and ZIKV in these countries is a major public health concern. These arboviruses are primarily transmitted by mosquitoes of the species Ae. Aegypti and its related diseases result in increased financial costs associated with diagnosis and treatment. To support the design of efficient diagnosis, prevention and treatment strategies, a bioinformatics tool has been developed for the genotyping of these viruses based on appropriate evolutionary models in an automatic, accurate and rapid manner. A set of virus reference sequences was obtained from GenBank and used for the development of the tools. This process involved the alignment of the reference sequences followed by phylogenetic tree reconstructions. To assign the genotypes uploaded by the user, the tool analyses the sequences one by one, genotypes through identification, alignment and phylogenetic reconstruction. This computational method allows the high-throughput classification of these virus species and genotypes in seconds. As shown experimentally, genotypes are classified most confidently using the envelope gene or complete genome sequences.
| In the recent years, an increasing number of outbreaks of Dengue (DENV), Chikungunya (CHIKV) and Zika (ZIKV) viruses have been reported in Asia and the Americas [1–3]. The predominant mosquito species transmitting DENV, CHIKV and ZIKV, are Aedes aegypti and Aedes Albopictus, which are widely distributed in tropical and sub-tropical regions [4]. In the past few years, several studies have reported concurrent outbreaks of DENV, CHIKV and ZIKV in the same geographical area [5, 6]. Currently, unprecedented outbreaks of DENV, CHIKV and ZIKV are co-occurring in Brazil. In 2017, the Brazilian Ministry of Health estimated that approximately 251,000 suspected cases of DENV, 185,000 suspected cases of CHIKV and close to 18,000 suspected ZIKV cases had occurred in Brazil [7].
Monitoring virus genotype diversity is crucial to understand the emergence and spread of outbreaks, both aspects that are vital to develop effective prevention and treatment strategies. Both DENV and CHIKV epidemics are associated with a mortality and morbidity that puts a significant economic burden on the affected regions [8,9]. While infections with ZIKV are rarely fatal, as stated before, ZIKV infections may result in Guillain-Barré syndrome and congenital malformations [10,11]. Genomic surveillance of epidemics at the appropriate resolution and consistently classifying the reported genetic sequences, also enables the identification of strains associated with greater epidemic potential [12] or disease severity [13].
However, methods that consistently classify arbovirus sequences at the level of species and sub-species (i.e. serotype and/or genotype) are currently lacking. Additionally, whole genome sequences are often not available in routine clinical settings, forcing the use of shorter gene sequences to classify at viral species or sub-species level. It has however insufficiently been explored which genomic regions are most suitable for accurate classification.
A new computational method for the identification of DENV/CHIKV/ZIKV sequences, with respect to species and sub-species (i.e. serotype and/or genotype), is presented. The classification method is implemented in the Genome Detective software tool, which was validated on a large dataset by assessing the classification performance of whole-genome sequences, partial-genome sequences and products from next-generation sequencing methods. Furthermore, the suitability of different genomic regions for virus classification was evaluated.
An efficient method to classify virus sequences with respect to their species and sub-species (i.e. serotype and/or genotype) was developed. This method was implemented in Java and this implementation was integrated in an easy-to-use web interface. A detailed description of the method and its implementation can be found in the ‘Classification method and implementation’ Methods subsection.
Two different methods were used to verify the suitability of sub-genomic regions for genotyping purposes: a boot-scanning method and a likelihood-mapping method (see Methods).
For DENV, the only sub genomic region that supports confident genotype assignment across the four different serotypes was the envelope gene. For CHIKV, the envelope region E1 was the only region that allowed consistent assignment. The boot-scanning analysis showed that for ZIKV, segments of around 1,200–1,500 base pairs support the genotype assignment with bootstrap > 70% (Fig 3). This was the case over the entire genome, with the exception of the end of the genome (i.e. the non-coding region) and near the NS3 region, where bootstraps fell below 60%.
Our likelihood-mapping analyses show that for DENV, the envelope, NS1, NS3 and NS5 had good phylogenetic signal across all four serotypes. For CHIKV, the envelope E2 gene had the best signal but this region did not provide good boot-scanning support for the classification of the ECSA genotype (Fig 3). For ZIKV, the envelope, NS1, NS2A, NS3, NS4A, NS4B and NS5 regions had good phylogenetic signal. A detailed overview of the results of the likelihood-mapping analysis can be found in the S2 Table of the Supporting Information.
In summary, these analyses show that the envelope genes of the reference datasets of the three pathogens (DENV, 1,485 nucleotides; CHIKV, 1,317 nucleotides; ZIKV, 1,525 nucleotides) are the most suitable targets for reliable genotype classification.
Our automated method provided specificity, sensitivity and accuracy of 100% for the identification of complete genomes for all viral species and genotypes compared to the gold standard, a manual classification. For a detailed overview of the DENV, CHIKV and ZIKV assignment performance, we refer to the Supporting Information S3 Table.
Only ten of 4118 DENV whole-genomes could not be classified at the genotype level, either by manual phylogenetic analysis or by our automated method. Notably, the seven sequences (AF298807, KF864667, EU179860, JQ922546, KF184975, KF289073, EF457905 of DENV-Sero1 were outliers in the phylogenetic tree (see Supporting Information, S1 Fig). We tested all ten sequences for recombination using boot-scanning (see Supporting Information, S2 Fig) and the recombination detection program RDP4 [36]. We only found sequence AY496879 to be a clear recombinant of DENV genotype 3I and 3II. The two other sequences (DENV-Sero2 KF744408 and DENV-Sero3 JF262783) were also identified as a divergent outlier.
Our analysis shows that the classification results for the envelope sub-genomic region at the species and genotype level were similar to that obtained using whole-genome sequences and largely in agreement with the gold standard, a manual classification.
For DENV, most of the genotypes were classified with great accuracy (i.e. specificity and sensitivity > 99%) using the envelope gene. The exception was DENV-sero2 genotype IV, of which 41 envelope sequences were available and for which 33 were correctly identified (i.e. sensitivity 80.49%, specificity 100%). The CHIKV sequences covering the E1 region were accurately classified for all three genotypes (i.e. 100% sensitivity and specificity). All the ZIKV envelope sequences were classified with 100% sensitivity and specificity. For a detailed overview of the DENV, CHIKV and ZIKV assignment performance refer to Supporting Information S4 Table.
Since a good phylogenetic signal was reported for the DENV and ZIKV NS5 region and the CHIKV E2 region, a classification analysis was performed for these regions as well. For the DENV NS5 region a sensitivity of 57,48% and specificity of 31,35%) was observed. Nearly all ZIKV NS5 sequences were correctly assigned to the African genotype (i.e. sensitivity of 97.72% and specificity of 100%). This indicates that the ZIKV NS5 region might also be used for genotype classification. For CHIKV, the E2 region showed perfect accuracy, similar to the E1 region (i.e. specificity and sensitivity of 100%). However, our previous boot-scanning support showed that the genetically variable E2 region may cause problems for some strains to be correctly identified as ECSA genotype.
In summary, our results suggest that the envelope region of DENV and ZIKV and the E1 envelope region of CHIKV are suitable for genotyping purposes. In addition, these regions contain the largest number of sequences in public databases, which easily allows for a wide range of comparative analyses and validation experiments.
Emerging infectious diseases caused by viral pathogens still represent a major threat to public health worldwide, as recently demonstrated by outbreaks of Ebola, Zika, Middle East Respiratory Syndrome (MERS) and Yellow Fever virus. Fast and accurate real-time monitoring of outbreaks and surveillance of on-going epidemics is crucial to anticipate viral spread and to design effective prevention or treatment strategies. To this end, an accurate and reliable method for the classification of ZIKV/DENV/CHIKV arboviruses was developed: The ArboTyping tool.
The ArboTyping tool implements a classification pipeline that consists of a BLAST-based species assignment and phylogenetic assessment to identify subspecies (i.e. genotypes) with respect to a set of reference strains, as exemplified for other virus species by previous work [29–31]. To enable accurate classification, a set of reference sequences that cover the extent of diversity within species and subspecies, was carefully selected.
The classification performance of the ArboTyping tool was assessed on a dataset of whole-genome sequences. All whole-genome sequences in this dataset that could be confidently assigned a species and genotype with the gold standard, a manual classification procedure, were concordant with the typing tool.
There were, however, 10 sequences that could not be classified using the manual classification procedure: further analyses show that these 10 sequences consist out of 3 outlier sequences, 2 clades of outlier sequences (3 sequences in each outlier clade) and 1 recombinant sequence. As these outliers have been previously identified [43], these results need to be further investigated to assess whether these outliers form new genotypes [44].
However, whole-genome sequences are currently not routinely available and the suitability of the different genomic regions was evaluated with respect to their use for classification. Since the envelope gene is a popular target for phylogenetic classification, there is a large availability of envelope sequences in public databases. Therefore, the performance of the ArboTyping tool was evaluated on a large dataset of envelope sequences (i.e. Global-ENV dataset). For these envelope sequences, a classification performance close to the tool’s performance on whole-genome sequences was reported.
While the availability of sequence products originating from other genomic regions is currently low, it can be expected that these regions will increase in relevance given the interest in developing antiviral agents that target non-structural proteins. Therefore, more detailed studies to assess the classification performance of other genomic regions are warranted [44].
In this manuscript, we focus on the classification of consensus sequences on the species and sub-species level. However, Genome Detective, the framework in which our tools are integrated, is also a virus discovery toolchain [41]. Genome Detective’s user interface allows users to supply raw next-generation sequence reads that can be automatically assembled into a consensus and passed to the ArboTyping tool. Details on the methods used to assemble reads in Genome Detective and an extensive validation using raw NGS reads can be found in [41].
In conclusion, the new method presented here allows the fast, accurate and high-throughput classification of DENV, CHIKV and ZIKV species and genotypes. Species can be classified using different sequencing products (i.e. whole-genome sequences, envelope sequences and individual next-generation sequencing reads) and genotypes can be classified most confidently when using envelope sequences or whole-genome sequences. This method accommodates the need to consistently and accurately classify DENV/CHIKV/ZIKV sequences, which is essential to implement epidemic tracing and to support outbreak surveillance efforts. Additionally, we present a solid framework that has the potential to serve as the foundation for many other arbovirus classification tools. These tools are also useful to be integrated in data management environments [45].
Our method is implemented in the Genome Detective software framework, suitable for many virus typing tools. The web application that makes our tool available through an easy-to-use web interface is available online via a dedicated server that is hosted at http://www.krisp.org.za/tools.php.
|
10.1371/journal.pntd.0001264 | A Multi-Center Randomized Trial to Assess the Efficacy of Gatifloxacin versus Ciprofloxacin for the Treatment of Shigellosis in Vietnamese Children | The bacterial genus Shigella is the leading cause of dysentery. There have been significant increases in the proportion of Shigella isolated that demonstrate resistance to nalidixic acid. While nalidixic acid is no longer considered as a therapeutic agent for shigellosis, the fluoroquinolone ciprofloxacin is the current recommendation of the World Health Organization. Resistance to nalidixic acid is a marker of reduced susceptibility to older generation fluoroquinolones, such as ciprofloxacin. We aimed to assess the efficacy of gatifloxacin versus ciprofloxacin in the treatment of uncomplicated shigellosis in children.
We conducted a randomized, open-label, controlled trial with two parallel arms at two hospitals in southern Vietnam. The study was designed as a superiority trial and children with dysentery meeting the inclusion criteria were invited to participate. Participants received either gatifloxacin (10 mg/kg/day) in a single daily dose for 3 days or ciprofloxacin (30 mg/kg/day) in two divided doses for 3 days. The primary outcome measure was treatment failure; secondary outcome measures were time to the cessation of individual symptoms. Four hundred and ninety four patients were randomized to receive either gatifloxacin (n = 249) or ciprofloxacin (n = 245), of which 107 had a positive Shigella stool culture. We could not demonstrate superiority of gatifloxacin and observed similar clinical failure rate in both groups (gatifloxacin; 12.0% and ciprofloxacin; 11.0%, p = 0.72). The median (inter-quartile range) time from illness onset to cessation of all symptoms was 95 (66–126) hours for gatifloxacin recipients and 93 (68–120) hours for the ciprofloxacin recipients (Hazard Ratio [95%CI] = 0.98 [0.82–1.17], p = 0.83).
We conclude that in Vietnam, where nalidixic acid resistant Shigellae are highly prevalent, ciprofloxacin and gatifloxacin are similarly effective for the treatment of acute shigellosis.
Controlled trials number ISRCTN55945881
| The bacterial genus Shigella is the most common cause of dysentery (diarrhea containing blood and/or mucus) and the disease is common in developing countries with limitations in sanitation. Children are most at risk of infection and frequently require hospitalization and antimicrobial therapy. The WHO currently recommends the fluoroquinolone, ciprofloxacin, for the treatment of childhood Shigella infections. In recent years there has been a sharp increase in the number of organisms that exhibit resistance to nalidixic acid (an antimicrobial related to ciprofloxacin), corresponding with reduced susceptibility to ciprofloxacin. We hypothesized that infections with Shigella strains that demonstrate resistance to nalidixic acid may prevent effective treatment with ciprofloxacin. We performed a randomized controlled trial to compare 3 day ciprofloxacin therapy with 3 days of gatifloxacin, a newer generation fluoroquinolone with greater activity than ciprofloxacin. We measured treatment failure and time to the cessation of individual disease symptoms in 249 children with dysentery treated with gatifloxacin and 245 treated with ciprofloxacin. We could identify no significant differences in treatment failure between the two groups or in time to the cessation of individual symptoms. We conclude that, in Vietnam, ciprofloxacin and gatifloxacin are similarly effective for the treatment of acute dysentery.
| Dysentery is an infection of the gastrointestinal tract characterized by diarrhea containing blood and/or mucous, abdominal cramping and tenesmus. The major cause of dysentery is the bacterial genus Shigella. Humans are the only known reservoir of Shigella, which are transmitted from person to person by the fecal-oral route, the infectious dose is low [1], and children under five years of age are at the greatest risk of infection [2]. The precise burden of Shigellosis is difficult to estimate, particularly in developing countries, as many cases go unreported and accurate diagnosis relies on bacterial culture, which is not always feasible or reliable. World Health Organization (WHO) estimates in 1999 put the number of cases at 164.7 million per year worldwide, with 99% of these infections occurring in developing countries [3]. The WHO also estimated that Shigellosis was responsible for over a million deaths per year [4]. Shigella are endemic in Vietnam [5], and in Nha Trang, in central Vietnam, an incidence of 490/100,000 in children under five has been estimated [6].
Antimicrobials are routinely administered to Shigellosis patients and correspond with clinical improvement [7], [8], however, many infecting organisms from various regions are now resistant to multiple antimicrobials [9]–[11]. In southern Vietnam we have recently documented a transition from S. flexneri to S. sonnei which has been concurrent with a dramatic shift in antimicrobial resistance patterns [12]. The majority (> 90%) of Shigella strains we now isolate demonstrate resistance to nalidixic acid [12]. Whilst nalidixic acid is no longer considered as an effective therapeutic agent for shigellosis, the fluoroquinolone, ciprofloxacin, is the current recommendation of the WHO [4], [13]. Ciprofloxacin is considered to be efficacious, cost effective and generally safe for use in a pediatric population. However, resistance to nalidixic acid generally correlates with a decreased susceptibility to ciprofloxacin and other older generation fluoroquinolones [14]. Mutations within the DNA gyrase gene (gyrA), the topoisomerase gene (parC) and multiple plasmid mediated quinolone resistance (PMQR) determinants can elevate the minimum inhibitory concentration (MIC) to ciprofloxacin, and may hinder effective treatment [15]. In fact, some Shigella strains that have been isolated from patients with dysentery have already been reported as being resistant to ciprofloxacin, with an MIC > 0.25 µg/mL [16]. Alternative regimens are obviously needed for those patients infection with organisms that may exhibit resistance to ciprofloxacin.
Gatifloxacin is a fourth generation fluoroquinolone. It has a broad spectrum of activity, with potent activity against many gram-positive and gram-negative organisms [17]. Gatifloxacin has been shown to be able be distributed extensively throughout human tissues and can actively penetrate phagocytic cells in vitro [18]. Significant intracellular accumulation means that gatifloxacin demonstrates bactericidal activity against susceptible intracellular pathogens, such as Shigella [19]. Gatifloxacin has been demonstrated to be highly efficacious in the treatment of enteric fever caused by nalidixic acid resistant organisms in trials performed in Nepal and Vietnam [20], [21]. A differential binding motif of gatifloxacin to the DNA gyrase (gyrA) with respect to ciprofloxacin, means that this agent is less prone to the inhibition induced by the common resistance associated mutations [15].
To date there very few published trials studying the efficacy of fluoroquinolones in the treatment of shigellosis in children [22]–[24] and we hypothesized that a therapy with gatifloxacin would be effective in treating shigellosis, particularly in an area with a high level of nalidixic acid resistance. We conducted a randomized controlled trial (RCT) evaluating the efficacy of gatifloxacin versus ciprofloxacin in pediatric patients with uncomplicated dysentery in two locations in southern Vietnam where nalidixic acid resistant Shigella predominate.
Controlled trials number ISRCTN55945881.
This study was conducted according to the principles expressed in the Declaration of Helsinki. This work was approved by the institutional ethical review boards of the Hospital for Tropical Diseases, Ho Chi Minh City, Huu Nghi Hospital, and The Oxford Tropical Research Ethics Committee (OXTREC) (number 010–06 (2006)), United Kingdom. The parent or guardian of all enrollees was required to provide written informed consent prior to entrance into the study.
An open label randomized comparison of gatifloxacin (10 mg/kg/day once-a-day orally) for 3 days versus ciprofloxacin (30 mg/kg/day in twice-a-day orally) for 3 days for the treatment of uncomplicated bacillary dysentery in children. This trial was design to demonstrate the superiority of gatifloxacin over ciprofloxacin.
The study physicians enrolled patients who were admitted to Pediatric Ward B, The Hospital for Tropical Diseases, Ho Chi Minh City, Vietnam or the Department of Infectious Diseases, Huu Nghi Hospital, Cao Lanh, Dong Thap, Vietnam from June 22nd, 2006 to March 13th, 2009. The study included children under the age of 15 years with a history of passing bloody or mucoid stools, with or without abdominal pain, tenesmus or fever (defined as a temperature >37.8°C) for less than 72 hours prior to admission. Exclusion criteria were any signs of a severe infection, including shock, jaundice, extensive gastrointestinal bleeding, a previous history of hypersensitivity to either of the trial drugs, known previous treatment with any (fluoro) quinolone during the current bout of disease prior to hospital admission, or a coexisting infection requiring antimicrobial therapy. Additionally, all children that had a trophozoite or Entamoeba histolytica present in their stool on microscopic examination were excluded from enrolment.
Each patient recruited into the study was randomized to receive one of two regimens, treatment with either gatifloxacin (Stada pharmaceuticals, Vietnam) 10 mg/kg/day in a single daily dose for 3 days or treatment with ciprofloxacin (OPV manufacturer, Vietnam) 30 mg/kg/d in two divided doses for 3 days. Other treatments, including fluid (ORS, parenteral) and antipyretics (paracetamol) were given to enrollees in both groups according to the treating clinicians' discretion. Seizures were treated with diazepam (OPV manufacturer, Vietnam) 0.25 mg/kg intravenously. No additional anti-diarrheal drugs, including smectite, loperamide or probiotics were permitted or used.
On admission (study day 0), a detailed history of the present illness was documented on a standard report form which recorded the duration of illness prior to admission to hospital (days), the presence of fever (defined as a prolonged temperature > 37.8°C), abdominal discomfort, vomiting, bloody or mucoid diarrhea (defined as ≥3 loose stools with obvious blood or mucus), estimated number of episodes of diarrhea before attending hospital, convulsions believed to be related to fever and/or infection and the administration of any known pre-treatment with antimicrobials and or other treatment. Baseline data including age (in months), sex, location of residence and weight (kg) were also collected.
A physical examination was performed on admission and daily thereafter until discharge. Particular note was taken of any bone or joint symptoms. The axillary temperature, pulse rate, respiratory rate, blood pressure, and frequency and character of stools were recorded every six hours. A full blood cell count was performed on all patients and stools were examined by microscopy (HPF (× 400)) to identify parasites, and to count white blood cells and red blood cells; the cell counts were scored on scale from zero to three, scale 0 = 0 cells/HPF, scale 1 = 1 to 10 cells/HPF, scale 2 = 11 to 20 cells/HPF, scale 3 = 21 to 30 cells/HPF and scale 4 = >40 cells/HPF. A blood sample was taken on Day 0 (on enrolment) and on day 3 (at the same time at the blood sample on Day 0) and glucose test was performed on those in the gatifloxacin group to assess any potential dysglycemia (hypoglycemia or hyperglycemia) in this portion of the study population over the duration of the therapy. Time from initial investigation in hospital to the cessation of bloody/mucoid and watery diarrhea (in hours) was recorded. Duration of hospital stay was recorded in days post admission; patients were only discharged when all clinical symptoms had resolved completely.
Enrolled patients were additionally required to attend a follow up visit, which was scheduled at seven days after discharge. During the follow up visit, the parents or guardians of the enrolled were questioned about their general health after discharge. Data regarding shigellosis-related symptoms were recorded in standard report form and a stool sample was collected for microbiological culture.
Patients stool samples were cultured overnight in selenite F broth (Oxoid, Basingstoke, UK) and onto MacConkey and XLD agar (Oxoid) at 37°C. Colonies suggestive of Salmonella or Shigella (non-lactose fermenting) were sub-cultured on to nutrient agar and were identified using a ‘short set’ of sugar fermentation reactions (Klinger iron agar, urea agar, citrate agar, SIM motility-indole media (Oxoid)). After incubation for 18–24 h at 37°C, the test media were read for characteristic Shigella reactions and API 20E test strips of biochemical reactions (Biomerieux, Paris, France) were used to confirm the identity of Shigella spp. Serologic identification was performed by slide agglutination with polyvalent somatic (O) antigen grouping sera, followed by testing with available monovalent antisera for specific serotype identification as per the manufacturers recommendations (Denka Seiken, Japan). Antimicrobial susceptibility testing of all Shigella isolates against nalidixic acid (NAL), ciprofloxacin (CIP) and gatifloxacin (GAT) was performed by disk diffusion following standardized Clinical and Laboratory Standards Institute methods [25]. The minimum inhibitory concentrations (MICs) were additionally calculated for all isolates by E-test, according to manufacturer's recommendations (AB Biodisk, Solna, Sweden) and were compared to control strain E. coli ATCC 25922 and an in house fully sensitive E. coli control. Isolates were stored on Protect beads (Prolabs, Oxford, United Kingdom) at −70°C.
The pre-defined primary endpoint of this study was the composite endpoint of in-hospital treatment failure defined as the occurrence of any clinical or microbiological treatment failure (or both). A clinical treatment failure was defined as fever (≥ 37.8°C), or the persistence of any signs or symptoms of the disease. These symptoms were defined as vomiting, abdominal pain or tenesmus with or without ≥3 loose stools with our without blood, mucus or blood and mucus after 120 hours of start of either treatment. Clinical treatment failures were subsequently treated with ceftriaxone at 50 mg/kg/day intravenously in single doses for 3 days at the discretion of the treating physician. A microbiological treatment failure was defined as a positive stool culture for the original infecting pathogen after the antimicrobial course was complete (day 3 of treatment or later).
The secondary endpoints were defined to describe the time from study enrolment (i.e. from the first dose of study treatment) until resolution of specific symptoms. As Shigellosis can be self-limiting, the time to cessation of all symptoms was additionally calculated from the time of disease onset, i.e. accounting for the duration of disease prior to hospital admission. The secondary endpoints were fever clearance time (FCT), i.e. time until the temperature was for the first time ≤37.8°C and subsequently remained ≤37.8°C for at least 48 hours; bloody diarrhea clearance time (BDCT), i.e. time until the first non-bloody stool recorded; diarrhea clearance time (DCT), i.e. time until the first non-diarrheal stool was recorded; and total duration of illness, i.e. the time till cessation of all symptoms. The follow-up visit was scheduled to occur 7 days after hospital discharge and information at this time point was not deemed to be as quantifiable as outcomes during hospitalization and, therefore, was not included in the primary outcome but the proportion of patients with failure at the follow-up visit (overall combined failure, diarrhea and other symptoms of shigellosis) was reported as a secondary endpoint.
This trial was designed as a superiority trial of gatifloxacin versus ciprofloxacin with the hypothesis that in patients with Shigella positive stool culture, gatifloxacin reduces the proportion of patients failing treatment from 25% to 5%. To detect such a reduction with 80% power at the two-sided 5% significance level, each group required 58 culture confirmed enrollees. We estimated the rate of Shigella positive stool culture to be approximately 33% of all children with bloody of mucoid diarrhea and the dropout rate to be approximately 5% and thus aimed to enroll a total of 366 cases with acute dysenteric syndrome. However, when 366 cases had been enrolled, only 80 culture positive patients with Shigella were identified (culture rate 22%), and the study was extended to enroll 134 more cases (500 in total). When 500 patients had been enrolled, 107 cases were stool culture positive for Shigella. Because of time and budget limitations, recruitment was terminated at 500. Of note, the original sample size calculation was based on a conservative sample size formula; a more precise power estimate based on simulation revealed that the power to detect the target effect with the recruited number of Shigella patients was 81%.
Patients were randomly allocated to one of two treatments. An administrator otherwise not involved in the trial performed randomization in blocks of 10. The random allocations were placed in sealed opaque envelopes, which were opened once the patient was enrolled into the trial after meeting the inclusion and exclusion criteria. Consecutive patients who presented to the clinic and were eligible to be enrolled were randomized and enrolled in the same sequence as they presented. Blinding was not carried out in this trial due to different frequencies of administration and other logistical reasons.
The proportion of patients reaching the primary endpoint of overall treatment failure and its components, respectively, were compared between the two arms using χ2-tests (p-values were calculated by Monte Carlo simulation if the expected count in any cell was <5) and 95% confidence intervals for absolute risk differences were calculated according to the method of Agresti and Caffo [26].
We analyzed the secondary time-to-event endpoints using survival methods and comparisons were based on Cox regression models with treatment as the sole covariate. We also summarized the proportion of patients with failure at the follow-up visit as for the primary endpoint. Only patients who attended the follow-up visit were included, i.e. we did not perform any imputation of missing data.
The main analysis population included all patients who received at least one dose of the intervention analyzed according to the randomized treatment group (intention-to-treat) and analyses were repeated in the per-protocol population, which included only Shigella spp. stool culture positive patient. In addition, the primary endpoint was analyzed in the subgroups defined by the following grouping criteria: culture positive vs. negative, Shigella pathogen vs. Salmonella pathogen and Nalidixic acid sensitive Shigella vs. Nalidixic acid resistant Shigella. Heterogeneity of the treatment effect was tested with a likelihood ratio test based on a logistic regression model that included an interaction between treatment effect and the sub-grouping criteria. P-values for two-sided tests and 95% confidence intervals are reported throughout. All analyses were performed with the statistical software R version 2.9.1 (R Foundation for Statistical Computing, Vienna, Austria, www.r-project.org.)
Between June 22nd, 2006 and March 13th, 2009 551 pediatric patients hospitalized with acute dysentery were assessed for enrolment in this RCT. The flow chart of the trial is presented in Figure 1, and shows that out of the eventual 500 enrolled patients randomized (300 at Huu Nghi Hospital, Dong Thap and 200 at the Hospital for Tropical Diseases), 494 received at least one dose of either treatment, corresponding with 245 patients in the ciprofloxacin group and 249 patients in the gatifloxacin group.
The baseline characteristics of the intention to treat patients (ITT) are shown in Table 1. The median age of the 494 patients in the intention-to-treat population (ITT) was 19 months, with a range of 2 to 144 months, 58.9% were Male. Patients had a median duration of dysentery prior to admission of 24 hours, with a range of 1 to 72 hours. The most common clinical observation was fever, as 87.4% of patients had a temperature > 37.8°C at randomization. Forty two point five percent (n = 210) of patients were enrolled with a history of blood in their stools and 57.5% of enrollees (n = 284) had a history of mucus without blood in their stools. Baseline characteristics, in particular disease severity (signified by blood in the stool, number of diarrheal episodes, fever, vomiting and abdominal pain), were comparable in both groups (Table 1).
There were more patients with a bacterial pathogen positive stool culture in the gatifloxacin group (n = 86, 34.5%) than in the ciprofloxacin group (n = 63, 25.7%), of which 61 and 46 were Shigella spp. respectively. Sixty seven point three percent (n = 72) of the Shigella isolated were resistant to nalidixic acid (MIC ≥16 µg/ml), additionally, the same proportion (n = 72) of Shigella isolates were S. sonnei, 33 were S. flexneri and 2 were S. boydii.
Results for the primary and secondary endpoints in the ITT population are shown in Table 2. The primary endpoint (composite overall treatment failure), demonstrated that fifty-seven (11.5%) of all ITT patients failed treatment, of which, 30 were in gatifloxacin group and 27 were in ciprofloxacin group (p = 0.72). The majority of patients that failed therapy had clinical failure, which was principally associated with persistent diarrhea for longer than 120 hours after initiating treatment. There were four patients with prolonged fever and nine with microbiological failures in the ciprofloxacin group and six microbiological failures in the gatifloxacin group. There were no significant difference in secondary endpoints related to time to resolution of symptoms (Table 2).
In the per-protocol analysis (Shigella spp. stool culture positive patients), the risk of treatment failure was as 4/61 (6.6%) for gatifloxacin and 5/46 (10.9%) for ciprofloxacin (p = 0.49). Three out of sixty one (4.9%) and 2/46 (4.3%) in the gatifloxacin group and the ciprofloxacin group, respectively, had persistent diarrhea for longer than 120 hours post treatment. In the per-protocol analysis, the time to resolution of symptoms was also similar in the gatifloxacin and ciprofloxacin groups, respectively: time from study enrolment to fever clearance (median 18 vs. 18 hours, p = 0.45) or bloody diarrhea clearance (median 24 vs. 24 hours, p = 0.74) and total illness duration (median duration from illness onset 77 vs. 71 hours, p = 0.48). The risk of overall treatment failure in selected subgroups is displayed in Table 3, which shows no evidence of treatment effect heterogeneity.
The proportion of patients that attended the follow up health visit was 87.4%. None of the 17 (3.9%) patients attending follow up which still demonstrated evidence of disease had a Shigella spp. positive stool culture, yet 15 (3.5%) still had persistent diarrhea (Table 2). There was no significant difference in prolonged symptoms synonymous with dysentery between the two study groups in the ITT analysis or the PP analysis.
There is a concern about the use of gatifloxacin, as it has been associated with dysglycemia (hypoglycemia and hyperglycemia) in elderly adults (both diabetic and non-diabetic) [27]. We compared the blood glucose levels to detect dysglycemia on day zero (prior to the first dose) (n = 244) and on day three (blood sample taken on same time of day as the primary sample) (n = 234) in patients in gatifloxacin group. On day zero, 180/244 (73.8%) had normal blood glucose (65–115 mg/dl), 49/244 (20%) had mild hyperglycemia (116–160 mg/dl), 6/244 (2.5%) had moderate hyperglycemia (161–250 mg/dl), 5/244 (2%) had mild hypoglycemia (55–64 mg/dl) and 4/244 (1.6%) had moderate hypoglycemia (40–54 mg/dl). On Day three, 193/234 (82.5%) had glucose within the normal range, 33/234 (14.1%) had mild hyperglycemia, 5/234 (2.1%) had moderate hyperglycemia and 3/234 (1.3%) had mild hypoglycemia. Paired glucose data were for 234 patients treated with gatifloxacin, the means of blood glucose levels on day zero and day three were 100.7 mg/dl and 98.4 mg/dl, respectively (p = 0.18; 2 tailed t-test). There were limited number of non-severe events, these included, one patient in the gatifloxacin arm with night sweats and 18 children vomiting after taking the medication (12 in ciprofloxacin group and 6 in the gatifloxacin group). In the case of vomiting the treatment drug were re-administered one hour after the vomiting episode. There was no death or severe clinical complications during the course of the trial. The clinical status of two children (one in each group) after three days of the study treatments warranted the use of intravenous ceftriaxone.
Shigellosis remains a common infection in low and medium human index countries and in countries undergoing industrialization with areas of inadequate sanitation [28]. In the pediatric proportion of the population that suffers shigellosis an effective antimicrobial remains important for the management of infection. Despite shigellosis being self-limiting, antimicrobials are generally considered to be associated with clinical improvement and a shortened duration of disease. The findings of a recent Cochrane review support antimicrobial use in Shigella infections, although the overall strength of data on antimicrobial therapy for shigellosis is weak and requires further consideration [29]. One potential effect of a suitable antimicrobial therapy for all Shigella patients is reducing fecal carriage of the organisms after infection, thus reducing transmission to close contacts, such as household members or children attending the same day-care setting [7], [23], [24].
A three day course of ciprofloxacin is the treatment currently recommended by the WHO for shigellosis, including S. dysenteriae type 1 infections [13]. In the study presented here, 494 children hospitalized for acute dysentery (in which 107 cases had a Shigella positive stool culture) were treated by either current WHO recommendations, or three days of one daily dose of oral gatifloxacin. Our study could not demonstrate the superiority of gatifloxacin. Rather, it showed a similar efficacy of both drugs in the treatment of childhood dysentery, including those with a stool culture confirmed Shigella infection. However, gatifloxacin has a longer half-life than ciprofloxacin and the once-a-day administration may be considered a little more convenient than twice-a-day regime of ciprofloxacin. In total, 67.3% of Shigella strains isolated were nalidixic acid resistant, but both antimicrobials appeared to clear Shigella from stools equally effectively (as tested by microbiological culture).
Our work has some caveats, firstly, blinding was not performed due to financial limitations and differing regimens, thus there was the potential of introducing investigator bias, yet, this was minimal due as result of the randomization process. Secondly, the primary endpoint was a composite endpoint of both clinical symptoms and microbiological failure, which was deemed more appropriate due to the nature of the infection. Furthermore, it was not possible to identify an infecting agent in the majority of cases, and we know from our currently unpublished work that more than one etiological agent may cause the infection. Therefore, whilst cessation of symptoms is the most relevant clinical end point (for the treating clinician and the patient) for shigellosis, it may be difficult to assess accurately and consistently. We cannot rule out that the persistence of symptoms may be caused by an additional infecting pathogen. Finally, it would be difficult to conduct such a trial in the community in this setting, etiological diagnostics for diarrheal disease are not routinely performed in this location and we are uncertain of the burden of shigellosis in the community. Our study may represent the most severe end of the disease spectrum for shigellosis, yet it is these individuals that may require the most apposite antimicrobial therapy.
Our data show similar overall risks of treatment failure in the two treatment groups (11% in the ciprofloxacin group versus 12% in gatifloxacin group). The overall risk of treatment failure rate of 8.4% in culture confirmed shigellosis cases is considerably lower than the 31% to 35% reported in the treatment of dysentery in a 2002 study [30]. This disparity in failure rate may be reflected by a difference in the principal Shigella serotype isolated. The most commonly isolated Shigella species here was S. sonnei, whilst S. dysenteriae was the most common isolated in the aforementioned 2002 study. S. dysenteriae causes a considerably more severe syndrome than S. sonnei, which is largely associated with the secretion of shiga toxin [31]. In an additional shigellosis treatment trial, the failure rate in adults treated with ciprofloxacin was 18% [32]. In dysentery treatment trial conducted in Ho Chi Minh City, S. flexneri was the dominant serogroup and the failure rate with the fluoroquinolone, ofloxacin, was comparable to our current data, at 10% [24]. A possible explanation for a reduced treatment failure rate, despite a pattern of reduced susceptibility to fluoroquinolones, is that shigellosis caused by S. sonnei is gradually becoming a more benign infection, an opinion that is shared by others [33]. Alternatively, due to improvements in education and healthcare, patients are now admitted to hospital earlier. Therefore, case management has improved and children receive therapy more rapidly after the onset of symptoms. Here, we observed similar failure rates in both treatment groups, despite a relationship between nalidixic acid resistance and reduced susceptibility to ciprofloxacin. These data suggest that whilst there has been a notable increase in MIC to nalidixic acid in Shigella in Vietnam over the last ten years, it may not yet be substantial enough to hinder the bactericidal effect of ciprofloxacin in vivo. A similar effect of both antimicrobial agents, despite gatifloxacin having greater in vivo activity supports the theory of a less severe infection, which may not, in all cases, require an antimicrobial for the cessation of symptoms. Alternatively, Shigella may respond in atypical manner to gatifloxacin, which respect to other Gram-negative organisms, and mutations in the gyrA and parC gene may have a greater effect on reducing the potency of the antimicrobial agent.
The use of fluoroquinolones in children has been contraindicated for many years because of the fear of arthropathy, however there is limited evidence to support this theory. A two year of follow-up of children treated with a short course ciprofloxacin or ofloxacin for typhoid fever in Vietnam did not find any adverse effects on growth and there was no evidence or arthropathy [34]. A more recent study conducted in a sheep model showed that neither ciprofloxacin nor gatifloxacin affected growth velocity when administered with using dosing regimen suitable for children [35]. The WHO, after considering the risks and the benefits, has recommended ciprofloxacin as the primary treatment of shigellosis in adults and children [13], our data suggests that this choice remains effective and we consider fluoroquinolones safe for use in the pediatric population. Gatifloxacin is an effective treatment against many Gram-negative infections, yet was voluntarily withdrawn from the US and Canadian drug market in May 2006 because of concerns about severe glucose disturbances. Reports of gatifloxacin related hypoglycemia and hyperglycemia in clinical trials are rare, yet severe dysglycemic events (mainly involved elderly patients with diabetes) were reported and discussed in the medical literature during the post-marketing period [27], [36], [37]. To date, clinically relevant dysglycemia has not been reported in young adults or in children treated with gatifloxacin [38], [39], we found no evidence of dysglycemia in the gatifloxacin treatment group. Treatment options for many Gram-negative infections in developing countries, including Shigella are clearly becoming increasingly limited. Gatifloxacin is a highly efficacious antimicrobial for the treatment of many gastrointestinal infections in the young and otherwise healthy patients and should be available for such indications in locations (i.e. in industrializing countries) where these diseases are endemic. However, in view of the potential risk of dysglycemia, it may be prudent to not treat patients over 50 years of age or with additional co-morbidities with such infections with gatifloxacin.
Our data demonstrate that despite a substantial increase in the number of organisms demonstrating resistance to nalidixic acid in the preceding years, ciprofloxacin remains an effective therapy for acute bacterial dysentery. Furthermore, we demonstrate that gatifloxacin is similarly effective as ciprofloxacin in the cessation of symptoms of shigellosis. Our current knowledge of the mechanisms and the global distribution of Shigella with reduced susceptibility to older fluoroquinolones (ciprofloxacin and ofloxacin) remain limited and require further investigation. We conclude that in locations where nalidixic acid resistant Shigellae are highly prevalent, ciprofloxacin and gatifloxacin are similarly effective for the treatment of acute shigellosis.
|
10.1371/journal.pgen.1000565 | Origin of an Alternative Genetic Code in the Extremely Small and GC–Rich Genome of a Bacterial Symbiont | The genetic code relates nucleotide sequence to amino acid sequence and is shared across all organisms, with the rare exceptions of lineages in which one or a few codons have acquired novel assignments. Recoding of UGA from stop to tryptophan has evolved independently in certain reduced bacterial genomes, including those of the mycoplasmas and some mitochondria. Small genomes typically exhibit low guanine plus cytosine (GC) content, and this bias in base composition has been proposed to drive UGA Stop to Tryptophan (Stop→Trp) recoding. Using a combination of genome sequencing and high-throughput proteomics, we show that an α-Proteobacterial symbiont of cicadas has the unprecedented combination of an extremely small genome (144 kb), a GC–biased base composition (58.4%), and a coding reassignment of UGA Stop→Trp. Although it is not clear why this tiny genome lacks the low GC content typical of other small bacterial genomes, these observations support a role of genome reduction rather than base composition as a driver of codon reassignment.
| The genetic code, which relates DNA sequence to protein sequence, is nearly universal across all life. Examples of recodings do exist, but new instances are rare. Genomes that exhibit recodings typically have other extreme properties, including reduced size, reduced gene sets, and low guanine plus cytosine (GC) content. The most common recoding event, the reassignment of UGA to Tryptophan instead of Stop (Stop→Trp), was previously known from several mitochondrial and one bacterial lineage, and it was proposed to be driven by extinction of the UGA codon due to reduction in GC content. Here we present an unusual bacterial genome from a symbiont of cicadas. It exhibits the UGA Stop→Trp reassignment, but has a high GC content, showing that reduction in GC content is not a necessary condition for this recoding. This symbiont genome is also the smallest known for any cellular organism. We therefore propose gene loss during genome reduction as the common force driving this code change in bacteria and organelles. Additionally, the extremely small size of the genome further obscures the once-clear distinction between organelle and autonomous bacterial life.
| The GC content of bacterial genomes has been known to vary widely since at least the 1950s [1]. Currently sequenced genomes range from 17–75% GC and show a strong correlation between genome size and GC content [2]–[4] (Figure 1). The tiny genomes of symbionts of sap-feeding insects are extreme exemplars of this relationship: Carsonella ruddii [5], Sulcia muelleri [6], and Buchnera aphidicola Cc [7], which represent three independently evolved endosymbiont lineages, have the smallest and most GC-poor genomes yet reported (Figure 1). These bacteria have a strict intracellular lifestyle, and this shift from a free-living state to an obligate intracellular one greatly reduces the effective population size of the bacteria, in part by exposing them to frequent population bottlenecks as they are maternally transmitted during the insect lifecycle [2],[3],[8]. This population structure leads to an increase in genetic drift, and this increase, combined with the constant availability of the rich metabolite pool of the insect host cell, is thought to explain the massive gene loss and high rate of sequence evolution seen in intracellular bacteria [2],[3]. Sequence evolution is also likely accelerated by an increased mutation rate, stemming from the loss of genes involved in DNA repair during genome reduction [4]. This loss of repair enzymes may contribute to the AT bias of small bacterial genomes since common chemical changes in DNA, cytosine deaminations and guanosine oxidations, both lead to mutations in which an AT pair replaces a GC pair, if left unrepaired [9],[10]. Indeed, the properties of all symbiont genomes published to date fit well within this framework (Figure 1).
The UGA Stop→Trp recoding, found in the mycoplasmas and several mitochondrial lineages, is associated with both genome reduction and low GC content [11]–[13]. Under the “codon capture” model, a codon falls to low frequency and is then free to be reassigned without major fitness repercussions. Applying this model to the UGA Stop→Trp recoding, mutational bias towards AT causes each UGA to mutate to the synonym UAA without affecting protein length [14],[15]. When the UGA codon subsequently reappears through mutation, it is then free to code for an amino acid [14],[15]. While some have argued that codon capture is insufficient to explain many recoding events [11],[12], the fact that all known UGA Stop→Trp recodings have taken place in high AT genomes [11],[16] makes the argument attractive for this recoding.
Here we describe the genomic properties of an α-Proteobacterial symbiont (for which we propose the name Candidatus Hodgkinia cicadicola) from the cicada Diceroprocta semicincta (Davis 1928) [17]. We show that at only 143,795 bps it has the smallest known cellular genome, but has a high GC content of 58.4% and a recoding of UGA Stop→Trp. We hypothesize that gene loss associated with genome reduction is a critical step in this recoding, rather than mutational pressure favoring AT. Specifically, we suggest that loss of translational release factor RF2, which recognizes the UGA stop, was the unifying force driving the recoding in Hodgkinia as well as in certain other small AT-rich genomes.
Previous work revealed that some cicadas had Sulcia as symbionts [18], but the identity of other symbionts, if any, was unknown. To identify any coexisting symbionts, we amplified and sequenced 16S rRNA genes from cicada bacteriomes (organs containing symbiotic bacteria). A second bacterial type was discovered and found to have large and irregularly shaped cells (Figure 2). Unusual cell morphologies have been observed in other bacteria with tiny genomes [5],[18], suggesting that this symbiont species might also have a small genome. Preliminary analysis using the Naive Bayesian rRNA Classifier [19] at the Ribosomal Database Project website [20] placed the new 16S rDNA sequence in the α-Proteobacteria with 100% confidence and, more specifically, within the Rhizobiales with 86% confidence. Because all other endosymbiotic α-Proteobacteria with small genomes are members of the Rickettsiales (e.g. Wolbachia, Rickettsia, and Erhlichia), we were interested in obtaining genomic data to further characterize this seemingly strange bacterium.
Genome sequencing revealed that Hodgkinia had some properties that were similar to other endosymbiont genomes, such as high coding density and shortened open reading frames (Table 1). But other aspects of the Hodgkinia genome suggested a highly atypical bacterial genome structure. In particular, the genome was only 144 kb, and thus even smaller than other known symbiont genomes, but had an unusually high GC content of about 58%. To our knowledge, this is an unprecedented combination of genome size and base composition (Figure 1). Additionally, initial rounds of gene prediction revealed that many protein-coding regions were interrupted by putative stop codons. Our previous experience [6] suggested that this could be due to errors in homopolymeric run lengths predicted by Roche/454 sequencing technology. However, the addition of Illumina/Solexa data indicated that the interrupted reading frames were not caused by sequencing errors. We noticed that computational translation of the genome with the NCBI genetic code 4 (UGA Stop→Trp) afforded full-length protein sequences, which immediately suggested that Hodgkinia might use an alternative genetic code.
Analysis of the gene complement of Hodgkinia revealed that the genome contains a homolog of prfA, encoding translational Release Factor RF1, which recognizes the stop codons UAA and UAG, but does not contain a homolog of prfB (RF2), which recognizes UAA and UGA. RF2 is dispensable if UGA is not used as a stop codon, and the loss of RF2 combined with recoding of UGA Stop→Trp is known in Mycoplasma species [13],[21],[22]. Additionally, the anticodon of the sole tRNA-Trp gene in Hodgkinia (trnW) has mutated from CCA to UCA, which allows recognition of both the normal tryptophan codon (UGG) and the putatively recoded UGA stop codon under Crick's wobble rules for codon-anticodon pairing [23]. This tRNA-Trp mutation has also been observed in mitochondrial genomes that have the UGA Stop→Trp recoding [24]. Additionally, it was observed that UGA codons in Hodgkinia open reading frames correspond to the position of conserved tryptophan residues in homologous proteins of other bacteria (Figure 3). Cumulatively, these data strongly suggested that UGA encodes tryptophan in Hodgkinia.
The long branch lengths for the Hodgkinia lineage in both rDNA and protein trees (Figure 4, Figure 5, and Figure S1) indicate a fast substitution rate, a situation typical of reduced bacterial genomes. Because the average percent identity of Hodgkinia proteins to their top hits in the GenBank non-redundant database was only 39.5%, it was difficult to rule out other recoding events based solely on sequence comparisons. To eliminate the possibility of other such changes in the genetic code, and to experimentally verify the UGA Stop→Trp recoding, shotgun protein sequencing by mass spectrometry [25] was used to sequence peptides derived from cicada bacteriomes. These peptide sequences ruled out any other codon reassignments, and experimentally confirmed the predicted UGA Stop→Trp code change (Figure 6 and Table S1).
Phylogenetic analysis of 16S rDNA sequences, including two newly acquired sequences from symbionts of other cicada species, shows that the cicada symbionts form a highly supported clade that falls within the α-Proteobacteria but outside of the Rickettsiales (Figure 4). The complete genome allowed additional phylogenetic analysis to further establish the placement of Hodgkinia within the α-Proteobacteria. Phylogenetic trees based on protein sequences (Figure 5 and Figure S1) support the grouping of Hodgkinia in the Rhizobiales, although the support was not always strong and trees made with some individual protein sequences placed it within the Rickettsiales with weak support (data not shown). We therefore looked for additional evidence in the form of gene order to further resolve the placement of Hodgkinia. The “S10” region (corresponding to the genomic region flanking ribosomal protein rpsJ) is a highly conserved cluster of genes that shares blocks of gene order conserved between Bacteria and Archaea [26]. The Rickettsiales have gene rearrangements and broken colinearity in this region that are unique within the α-Proteobacteria ([27] and Figure 7). Hodgkinia does not share these genomic signatures, instead showing perfect colinearity with genomes in the Rhizobiales and Rhodobacteraceae (Figure 7). These data rule out Hodgkinia's grouping within the Rickettsiales, but do not entirely preclude a common ancestor with them, as Hodgkinia could have diverged from other Rickettsiales before the S10 region rearrangement.
The accurate placement of Hodgkinia within the α-Proteobacteria is confounded by both long branch attraction (LBA) and large differences in GC contents between different members of the α-Proteobacteria. LBA is expected to incorrectly associate Hodgkinia with the Rickettsiales, since these two lineages have the longest branches on the tree. Therefore, the fact that most analyses place Hodgkinia outside the Rickettsiales is significant. Conversely, the GC content bias is expected to incorrectly group sequences that are similar in GC content but that are not truly related by ancestry, and this artifact might tend to place Hodgkinia outside of the Rickettsiales, since Hodgkinia and most other non-Rickettsial α-Proteobacteria have high GC contents. We therefore tested all possible permutations in the placement of the Hodgkinia clade shown in Figure 4 under a model that does not assume nucleotide composition homogeneity among taxa [28],[29]. Hodgkinia did not group with the Rickettsiales in any of the highest scoring trees (Figure 4), suggesting that Hodgkinia's grouping in the Rhizobiales was not a function of GC content bias. Overall, the results from the phylogenetics of proteins and 16S rDNA, as well as from gene order comparisons, strongly argue for the grouping of Hodgkinia with the Rhizobiales.
All previously confirmed UGA Stop→Trp recoding events have occurred in genomes with low GC content: the mitochondria of Metazoa and Fungi, some Protist mitochondria, and certain bacteria in the Firmicutes [11]. (This same recoding may have occurred in the nuclear genomes of some Ciliates, but information on those genomes is limited [16]). Proposed evolutionary mechanisms for genetic code reassignments fall into three groups: the codon capture hypothesis [14],[15], involving the extinction and reassignment of codons; the genome reduction hypothesis, under which the pressure to minimize genome content drives the recoding of some codons, reducing the number of tRNAs [30]; and the ambiguous translation hypothesis, under which a single codon is temporarily read in two different ways, with a subsequent loss of the original meaning of the code [12],[31]. These hypotheses are not mutually exclusive and may apply more to some recoding events than to others [12]. For example, the pioneering ideas of Osawa and Jukes on this topic [14] involved loss of the corresponding tRNA following the extinction of a codon. Also, ambiguous translation, which is known for Bacillus subtilis [32], could facilitate a transition through the codon extinction route or the genome reduction route.
Codon capture requires the changing of one codon to another synonym though an initial codon extinction step potentially resulting from biases in nucleotide base composition. All previously described cases of UGA Stop→Trp recoding occur in GC-poor genomes, and this recoding has been proposed to result from genome-wide replacement of UGA by UAA, due to AT-biased mutational pressure [14],[15]. Under this explanation, the extinction of UGA Stop allows UGA to later reappear, recoded as an amino acid. Several arguments weigh against the codon capture hypothesis [11],[12]; most relevant is the fact that, in mitochondrial genomes, there is no association between the codons that undergo a reassignment and those that are expected to potentially disappear due to GC content bias [12]. Tallying stop codons in α-Proteobacteria with complete genomes also weighs against codon extinction as an initial step in this recoding event: although UGA codons are fewest in small and AT-biased genomes, in no case does UGA approach extinction. Among previously sequenced α-Proteobacteria (excluding Hodgkinia), even the smallest and most AT-biased genomes retain over 100 genes using UGA as Stop (e.g., there are 137 UGA Stop codons in the 1.11 Mb genome of Rickettsia prowazekii, which has a GC content of only 29%). In α-Proteobacteria with GC-rich genomes, UGA is the most frequent of the three stop codons and is typically used in a majority of genes (typically 50%–70% of coding genes end in UGA). Thus, the combination of phylogenetic evidence, which places Hodgkinia in the GC-rich Rhizobiales, and UGA usage patterns in extant α-Proteobacteria weigh strongly against UGA extinction as a causal step in the observed recoding.
We suggest an alternative hypothesis, implicating genome reduction as the primary driver of the UGA recoding, to explain the coding change observed in Hodgkinia (Figure 8). As in the ambiguous translation hypothesis, the recoding would first be enabled by the relaxed codon recognition of a mutated tRNA-Trp as promoted by structural changes in the tRNA [31] (Figure 8, step 1). For example, point mutations in either the D- or anticodon-arms of tRNA can induce C-A mispairing at the third codon position [33],[34]. In the presence of such alternative coding, RF2 is no longer essential and thus can be lost through the ongoing process of genome reduction (step 2). This is similar to the scenario envisioned in the codon capture hypothesis, except that in our case UGA does not need to have gone extinct before RF2 is lost. The further changes observed in Hodgkinia would evolve readily since they involve single base changes driven by positive selection; these include a change in the tRNA-Trp anticodon (step 3) and shifts in stop codon usage (step 4).
Since UGA Stop→Trp has evolved independently in other small genomes such as Mycoplasma and mitochondria, the case of Hodgkinia weighs in favor of genome reduction, and specifically loss of RF2, as the common force driving UGA Stop→Trp recoding events. Some of the Mollicutes, including Mycoplasma, and certain mitochondrial lineages are the other clear cases of this recoding event, and these genomes also have been characterized by a history of ongoing gene loss [22]. Of course, some small genomes do not show this recoding, and we do not expect the consequences of genome reduction to be predictable in each case. For example, the highly reduced genome of Carsonella ruddii, which retains UGA Stop and RF2, exhibits an unusual feature of having many overlapping genes with the most common overlap consisting of ATGA, in which ATG is the start of the downstream genes and TGA is the stop of the upstream gene [35], a situation that might act to conserve UGA Stop and RF2 in the genome.
At the initial loss of RF2, the additional C-terminal length imposed on UGA-ending proteins might be expected to impose some deleterious effects. It is possible that the functionality of proteins with such extensions could be enhanced in Hodgkinia due to an abundance of protein-folding chaperonins, similar to the high levels of GroEL seen in other symbiotic bacteria with small genomes [36],[37]. Indeed, analysis of the shotgun proteomic data for Hodgkinia shows that homologs of GroEL and DnaK are the two most abundant proteins in the cell (Table 2). Additionally, the shortened gene lengths observed in Hodgkinia relative to homologs in other genomes (Table 1) indicate that, if UGA-ending proteins were once extended due to recoding, they have since been reduced in length by the generation of new UAG and UAA stop codons. Other models are possible, such as the loss of RF2 effected by a change in the tRNA-Trp anticodon from CCA to UCA instead of distal mutations. Similarly, it is formally possible that Hodgkinia went through a period of AT bias under which the recoding occurred, with a subsequent shift to GC bias as is seen in the present genome. Because phylogenetic evidence favors placement of Hodgkinia's in the Rhizobiales and not within any group characterized by AT rich genomes, we consider this scenario unlikely. Regardless of the recoding mechanism, however, this example provides a rare case in which the loss of an “essential” gene (RF2) in a highly reduced bacterial genome can be compensated by a few simple steps, namely the adaptive fixation of several point mutations.
The mechanisms that give rise to GC-content differences in bacterial genomes are unclear, although variations in the replication and/or repair pathways are often suggested as candidates [38]–[40]. Various lines of evidence support this idea, including a correlation between genome GC content and the types of DNA polymerase III, α subunit (DnaE) encoded in a genome [41] and the discovery of point mutations affecting the repair enzyme MutT that can detectably change the GC content of Escherichia coli [38]. One mechanistic clue is the correlation between genome size and GC content, a universal pattern in previously studied bacterial and archaeal genomes (Figure 1). Until now, this tendency has been especially pronounced in obligate intracellular bacterial genomes. Two (not necessarily mutually exclusive) hypotheses have been forwarded to explain this base composition bias in genomes of intracellular organisms. The first is an adaptive argument, based on selection for energy constraints [42]: synthesis of GTP and CTP require more metabolic energy, and ATP is the most common nucleotide in the cell because of its ubiquitous role in cellular processes. Therefore, competition for scarce metabolic resources has been hypothesized to force intracellular genomes to low GC values. The second hypothesis relates to mutational pressure resulting from altered capacity for DNA repair [43]. Small intracellular genomes typically lose many repair genes, and these organisms therefore are expected to be deficient in their ability to repair damage caused by spontaneous chemical changes. This is particularly expected in organisms such as endosymbionts in which genetic drift plays a major role in sequence evolution [43]. Indeed, recent experiments in Salmonella strongly support this hypothesis [44].
Our results weigh against the energetic hypothesis because Sulcia, living in the same bacteriome and presumably exposed to the same metabolite pool, has a GC content of 22.6% (J.P.M, B.R.M, and N.A.M., unpublished data), almost identical to the GC content of 22.4% for the previously published Sulcia genome from Glassy-winged sharpshooter [6]. One would expect that if the metabolite pool caused an increase in GC content in Hodgkinia, the same trend would be observed in Sulcia. Additionally, the GC content of the third position in 4-fold degenerate sites (which should be under little or no selective pressure) in the Hodgkinia genome is 62.5% (Table S2), consistent with mutational pressure as a cause of elevated genomic GC content.
Collectively, these data suggest that the replicative process or mutagenic environment of Hodgkinia differ from those of other small-genome α-Proteobacteria and other small genome insect symbionts. Hodgkinia has only two genes involved in replication (dnaE, DNA polymerase III, α subunit; and dnaQ, DNA polymerase III, ε subunit), implicating them as primary targets for future study of the source of GC bias. Regardless of the mechanisms involved in shifting genomic GC contents, our results indicate that low GC content is not an inevitable consequence of loss of repair enzymes, since Hodgkinia has no detectable repair enzymes (and is thus more extreme in this regard than previously sequenced symbiont genomes, which show partial loss of repair enzymes).
Our finding that two other cicada species contained symbionts belonging to the same clade, based on 16S rDNA genes (Figure 4) suggests that this symbiont infected an ancestor of cicadas and subsequently has been transmitted maternally, a typical history for bacteriome-dwelling insect symbionts [45],[46]. In such cases, the symbiont is restricted to its particular group of insect hosts, and restriction to cicada hosts is highly likely for this case. We propose the candidate name Candidatus Hodgkinia cicadicola for this α-Proteobacterial symbiont of cicadas, with the genus name referring to the biochemist Dorothy Crowfoot Hodgkin (1910–1994), and the species name referring to presence only in cicadas. Distinctive features include restriction to cicada bacteriomes, large tube-shaped cells, a high genomic GC content, a recoding of UGA Stop→Trp, and the unique 16S rDNA sequence ACGAGGGGAGCGAGTGTTGTTCG (positions 535–557, E. coli numbering).
Female cicadas were collected in and around Tucson, Arizona, USA. Tissue for genome sequencing was prepared from bacteriomes dissected in 95% ethanol and cleaned up in Qiagen's DNeasy Blood and Tissue Kit. DNA was prepared for the Roche/454 GS FLX pyrosequencer [47] following the manufacturer's protocols. Sequencing generated 523,979 reads totalling 116,176,938 bases, and these were assembled using the GS De novo Assembler (version 1.1.03) into 1029 contigs. Contigs expected to be from the Hodgkinia genome were identified by BLASTX [48] against the GenBank non-redundant database and the associated reads were extracted and reassembled to construct the Hodgkinia genome. Eleven contigs with an average depth of 73× were generated representing 143,582 nts of sequence with an average GC content of 58.4%. The order and orientation of the 11 contigs were predicted using the “.fm” and “.to” information appended to read names encoded in the 454Contigs.ace file and these joins were confirmed by PCR and Sanger sequencing.
Illumina/Solexa sequencing [49] generated 12,965,640 reads totalling 505,659,960 nts. These data were mapped to the Hodgkinia genome using MUMMER [50] (nucmer -b 10 -c 30 -g 2 -l 12; show-snps -rT -×30) to an average depth of 43×. Forty-five homopolymeric nucleotide runs were adjusted in length based on the Illumina data. Annotation was carried out as described previously [6], except that NCBI genetic code 4 (TGA encoding tryptophan) was used to computationally translate the predicted protein-coding genes. The Candidatus Hodgkinia cicadicola genome has been deposited in the GenBank database with accession number CP001226.
D. semicincta bacteriomes were dissected in PBS and gently disrupted with a mortar and pestle. Cells were fixed as described [51] and imaged on a Zeiss 510 Meta microscope. The probe sequences were Cy3-CCAATGTGGGGGWACGC (Sulcia) and Cy5-CCAATGTGGCTGACCGT (Hodgkinia). The scale bar in Figure 2 generated by the microscope software was overlaid with a plain white bar for legibility.
The PCR conditions used to amplify Magicicada cassini (Brood XIII, Chicago, Illinois) and Diceroprocta swalei (Tucson, Arizona) 16S rDNA were 94°C for 30 seconds followed by 35 cycles of 94°C 15 seconds, 58°C 30 seconds, 72°C 2 minutes using the primers 10F_ALPHA (AGTTTGATCCTGGCTCAGAACG) and 1507R (TACCTTGTTACGACTTCACCCCAG). Amplicons were cloned into Invitrogen's TOPO TA PCR2.1 kit and sequenced. The D. swalei and M. cassini 16S rDNA sequences have been deposited in the GenBank database with accession numbers FJ361199 and FJ361200, respectively.
The initial assignment of the Hodgkinia 16S rRNA sequence was based on the Naive Bayesian classifier [19] at the Ribosomal Database Project (RDP) [20]; this uses a bootstrapping procedure involving resampling of sequence fragments with replacement and assignment of individual fragments to taxonomic units represented in this large database. The three Hodgkinia 16S rDNA sequences, sampled from bacteriomes of D. semicincta and two additional cicada species (M. cassini and D. swalei), were aligned to the Bacterial 16S rDNA model at the RDP, and the remaining sequences used in the generation of Figure 4 were also obtained from the RDP. The maximum likelihood tree in Figure 4 was generated using RAxML [52] under the GTRGAMMA model of sequence evolution. The clade consisting of the Hodgkinia sequences was moved to all other possible positions on the tree in Mesquite [53], and the log likelihood of each of these trees was estimated using the non-homogenous model implemented in nhPhyML [29] under a 4 category discrete gamma model using the shape parameter estimated from PUZZLE [54].
The protein sequence used in generating Figure 5 was DnaE (DNA polymerase III, α subunit), and the proteins used in generating Figure S1 were DnaE, InfB (translational initiation factor IF2), TufA (translational elongation factor Tu), RpoB (RNA polymerase, β subunit), and RpoC (RNA polymerase, β′ subunit). Individual alignments for each gene were generated using the linsi module of MAFFT [55] and (in the 5-protein alignment) concatenated. Columns containing gap characters were removed, leaving 861 columns in the DnaE alignment and 4152 columns in the 5-protein alignment. Parameters for a 1 invariant/4 Gamma distributed rate heterogeneity model were estimated using PUZZLE, and maximum likelihood trees were computed with PROML from the PHYLIP package [56] using the JTT model of sequence evolution. One hundred bootstrap datasets were generated using SEQBOOT from PHYLIP, trees were calculated as above, and bootstrap values for these trees were mapped back on the maximum likelihood tree calculated from PROML using RAxML. The family and order names and groupings on Figure 4, Figure 5, and Figure S1 were taken from [57] and the RDP website [20]. The genomes used in the phylogenetic analysis were (the accession numbers noted with asterisks were used in generating Figure 7): Zymomonas mobilis subsp. mobilis ZM4 (NC_006526), Erythrobacter litoralis HTCC2594 (NC_007722), Novosphingobium aromaticivorans DSM 12444 (NC_007794), Sphingopyxis alaskensis RB2256 (NC_008048), Candidatus Pelagibacter ubique HTCC1062 (NC_007205), Rickettsia rickettsii str. Iowa (NC_010263), Rickettsia typhi str. Wilmington (NC_006142*), Neorickettsia sennetsu str. Miyayama (NC_007798), Wolbachia pipientis (NC_010981), Wolbachia endosymbiont of Drosophila melanogaster (NC_002978), Anaplasma phagocytophilum HZ (NC_007797), Anaplasma marginale str. St. Maries (NC_004842), Ehrlichia ruminantium str. Gardel (NC_006831), Ehrlichia canis str. Jake (NC_007354), Rhodospirillum rubrum ATCC 11170 (NC_007643), Magnetospirillum magneticum AMB-1 (NC_007626), Acidiphilium cryptum JF-5 (NC_009484), Gluconobacter oxydans 621H (NC_006677), Gluconacetobacter diazotrophicus PAl 5 (NC_010125), Paracoccus denitrificans PD1222 (NC_008686/NC_008687), Rhodobacter sphaeroides ATCC 17025 (NC_009428*), Jannaschia sp. CCS1 (NC_007802), Silicibacter pomeroyi DSS-3 (NC_003911), Silicibacter sp. TM1040 (NC_008044), Roseobacter denitrificans OCh 114 (NC_008209), Caulobacter crescentus CB15 (NC_002696), Caulobacter sp. K31 (NC_010338), Maricaulis maris MCS10 (NC_008347), Brucella melitensis 16M (NC_003317/NC_003318), Brucella abortus S19 (NC_010742/NC_010740), Bartonella bacilliformis KC583 (NC_008783), Bartonella henselae str. Houston-1 (NC_005956), Mesorhizobium loti MAFF303099 (NC_002678), Mesorhizobium sp. BNC1 (NC_008254), Agrobacterium tumefaciens str. C58 (NC_003062/NC_003062), Rhizobium etli CFN 42 (NC_007761), Sinorhizobium medicae WSM419 (NC_009636), Sinorhizobium meliloti 1021 (NC_003047*), Rhodopseudomonas palustris BisA53 (NC_008435), Rhodopseudomonas palustris BisB18 (NC_007925), Bradyrhizobium japonicum USDA 110 (NC_004463), Bradyrhizobium sp. BTAi1 (NC_009485), Nitrobacter hamburgensis X14 (NC_007964), Nitrobacter winogradskyi Nb-255 (NC_007406), Xanthobacter autotrophicus Py2 (NC_009720), and Escherichia coli str. K12 substr. MG1655 (NC_000913).
Total protein was prepared from the bacteriomes of 10 female D. semicincta by homogenizing in 4 ml Buffer H (2% SDS, 100 mM Tris, 2% β-mercaptoethanol, pH 7.5) followed by centrifugation at 100,000×g for 30 min. The supernatant was recovered and precipitated in 12% TCA followed by 3 washes in cold acetone. The resulting protein pellet was resuspended in 150 µl sample loading buffer, and 30 µl (∼60 µg) of this sample was loaded onto a well of a 11 cm×8 cm×1.5 mm 10% acryl amide gel. Electrophoresis was performed in a mini cell (Bio-Rad) at 130 V. The entire lane was cut into 12 sections, and proteins in each section were identified by LC-MS/MS analysis.
The gel bands were washed, homogenized, reduced, alkylated and subjected to overnight in-gel tryptic digests. The peptide mixture was extracted, dried in speed-vac and dissolved in a 15 µl of 5% formic acid. The LC-MS/MS experiments were performed on a Q-TOF 2 mass spectrometer equipped with the CapLC system (Waters Corp., Milford, MA). The stream select module was configured with a 180 µm ID×50 mm trap column packed in-house with 10 µm R2 resin (Applied Biosystems, Foster City, CA) connected in series with a 100 µm ID×150 mm capillary column packed with 5 µm C18 particles (Michrom Bioresources, Auburn, CA) using a pressure cell. Peptide mixtures (10 µl) were injected onto the trap column at 9 µl/min and desalted for 6 min before being flushed to the capillary column. The peptides were then eluted from the column by the application of a series of mobile phase B gradients (5 to 10% B in 4 min, 10 to 30% B in 61 min, 30 to 85% B in 5 min, 85% B for 5 min). The final flow rate was 250 nl/min. Mobile phase A consisted of 0.1% formic acid, 3% acetonitrile and 0.01% TFA, whereas mobile phase B consisted of 0.075% formic acid, 0.0075% TFA in an 85/10/5 acetonitrile/isopropanol/water solution. The mass spectrometer was operated in a data dependent acquisition mode whereby, following the interrogation of MS data, ions were selected for MS/MS analysis based on their intensity and charge state +2, +3, and +4. Collision energies were chosen automatically based on the m/z and charge-state of the selected precursor ions. The MS survey was from m/z 400–1600 with an acquisition time of 1 sec whereas the trigged data-dependent MS/MS fragmentation scan was from m/z 100–2000 with an acquisition time of 2.4 sec.
The peak list was created using the Mascot distiller 2.2 software from Matrix Science (London, UK) using the default settings for Waters. The Mascot 2.2 search engine was used to assist in the search of the combined tandem mass spectra against a custom protein database. The custom protein database consisted of the Hodgkinia proteome, the nearly complete proteome of Sulcia muelleri from Diceroprocta semicincta (J.P.M., B.R.M., and N.A.M., unpublished), and the complete proteome from the pea aphid Acyrthosiphon pisum (build 1.1), the most closely related insect to D. semicincta for which a complete genome is available. The database contained 5,508,819 amino acids residues in 10,887 protein sequences. The parameters used for the searches were as follows: trypsin-specificity restriction with 2 missing cleavage site and variable modifications including oxidation (M), deamidation (N,Q), and alkylation (C). Both MS and MS/MS mass tolerance was set to 0.3 Da for the searching.
The Mascot significance threshold was set to 0.05, using MudPIT scoring, with a Mowse ion score cutoff of >31 (the cutoff for a peptide suggesting identity or extensive homology). The sequences in the custom proteome database were reversed to generate a decoy database for calculation of a false discovery rate, which was 2.6% (15 peptides found in the decoy database vs. 576 peptides found in the real database). For a peptide to be considered in the calculation of codon coverage (Figure 6), it had to originate from a protein with at least one other high-quality matching peptide. Eighty-seven (87) such peptides from 16 Hodgkinia proteins were found (Table S1). These peptides cover all 62 non-stop codons at least once; the peptides LIWPSAVLQAEEVWAGAR from HCDSEM_044 and VSCLIWTDINR from HisA span recoded UGA codons.
|
10.1371/journal.pcbi.1005443 | Geometry can provide long-range mechanical guidance for embryogenesis | Downstream of gene expression, effectors such as the actomyosin contractile machinery drive embryo morphogenesis. During Drosophila embryonic axis extension, actomyosin has a specific planar-polarised organisation, which is responsible for oriented cell intercalation. In addition to these cell rearrangements, cell shape changes also contribute to tissue deformation. While cell-autonomous dynamics are well described, understanding the tissue-scale behaviour challenges us to solve the corresponding mechanical problem at the scale of the whole embryo, since mechanical resistance of all neighbouring epithelia will feedback on individual cells. Here we propose a novel numerical approach to compute the whole-embryo dynamics of the actomyosin-rich apical epithelial surface. We input in the model specific patterns of actomyosin contractility, such as the planar-polarisation of actomyosin in defined ventro-lateral regions of the embryo. Tissue strain rates and displacements are then predicted over the whole embryo surface according to the global balance of stresses and the material behaviour of the epithelium. Epithelia are modelled using a rheological law that relates the rate of deformation to the local stresses and actomyosin anisotropic contractility. Predicted flow patterns are consistent with the cell flows observed when imaging Drosophila axis extension in toto, using light sheet microscopy. The agreement between model and experimental data indicates that the anisotropic contractility of planar-polarised actomyosin in the ventro-lateral germband tissue can directly cause the tissue-scale deformations of the whole embryo. The three-dimensional mechanical balance is dependent on the geometry of the embryo, whose curved surface is taken into account in the simulations. Importantly, we find that to reproduce experimental flows, the model requires the presence of the cephalic furrow, a fold located anteriorly of the extending tissues. The presence of this geometric feature, through the global mechanical balance, guides the flow and orients extension towards the posterior end.
| The morphogenesis of living organisms is a facinating process during which a genetic programme controls a sequence of molecular changes which will cause the original embryo to acquire a new shape. While we have a growing knowledge of the timing and spatial distribution of key molecules downstream of genetic programmes, there remain gaps of understanding on how these patterns can generate the appropriate mechanical force, so as to deform the tissues in the correct manner. In this paper, we show how a model of tissue mechanics can link the known pattern of actomyosin distribution in Drosophila tissues to the process of axis extension, which is a ubiquitous morphogenetic movement of developing animal embryos. We show in numerical simulations that the correct movement is obtained only if the geometry of the embryo presents some precise features. This means that prior morphogenetic movements responsible for these features need to have succeeded in order to carry on the next round of morphogenesis. This highlights the contribution of mechanical feedback on the morphogenetic programme and also how mechanical action integrates at the scale of the whole embryo.
| The morphogenesis of living organisms involves precise shape changes and displacements of the tissues that constitute the embryo under the control of gene expression [1]. These movements result from changes in the mechanical balance, which can be caused by local growth [2] or by local activation of the contractile machinery of actomyosin [3]. The action of these effectors can integrate at the scale of a whole tissue through the establishment of a new mechanical balance, which happens quasi-instantaneously in the absence of inertia [4]. This integration at a tissue-scale explains how a local process can effect a global deformation [5–9]. The mechanical coupling of different tissues and extra-cellular structures crucially changes the resulting mechanical balance globally, and in turn can lead to very different morphologies. For instance, uncoupling the Drosophila pupal wing blade from the cuticle at its margin prevents its elongation during hinge contraction [8].
Axis extension of the Drosophila embryo at gastrulation offers a good system to model convergence and extension morphogenetic flows, which are ubiquitous in the development of animals [10]. During axis extension, a region of the epithelial monolayer that makes up the embryo, the ventro-lateral part of the germ-band, narrows in the dorsoventral (DV) direction and simultaneously extends in the orthogonal direction (antero–posterior, AP) towards the posterior end, Fig 1a–1c. This movement is known to depend on the genetically specified organisation of actomyosin in a planar-polarised manner, with an enhanced Myosin II recruitment specifically along the cell–cell junctions aligned with the DV axis [11, 12], Fig 2a–2c. This planar-polarised actomyosin is responsible for active shortening of these DV junctions, which resolve in active cell intercalations [13]. The total tissue deformation of Drosophila axis extension is however not explained by the sole action of planar-polarised myosin [14]. In addition, the extending germ-band is subjected to an external pull originating from the invagination of the posterior midgut [6, 7], Fig 1a–1c. As a result, cells change shape, elongating along the AP axis, and the deformation of the tissue is the combination of both cell intercalation and cell shape strain contributions [14, 15].
In this system, the dynamics of cell intercalation and myosin planar-polarisation have been modelled using discrete element approaches, which explicitly account for each cell–cell junction [16–19]. However, so far, it has not been possible in these modelling approaches to avoid using arbitrary boundary conditions to represent the mechanical resistance of the surrounding epithelia. The several morphogenetic movements occurring at once during Drosophila gastrulation interact through the global mechanical balance, and thus germ-band extension and posterior midgut invagination are not independent of one another [6, 7]. In addition to the germ-band and posterior midgut, the neighbouring dorsal tissue, the future amnioserosa, is also being deformed during axis extension, although there is no active process reported in this tissue [5, 9]. In this paper, we choose to focus on the tissue scale dynamics and the importance of the global mechanical balance on morphogenetic flows. To understand this, we solve a mechanical problem set on a three-dimensional surface that has the shape of the embryo. We treat the apical surfaces of the epithelial cells as a single continuum. Indeed, in the course of a convergence–extension process for an isolated tissue, the dynamics governing the tissue-scale can be captured by a material law involving only the resistance to shear and the energetic cost of area variations, see Fig 2e–2h. These energetic costs sum per unit area the average cost of the deformations of each cell and the one of their rearrangement, which are the two components of tissue strain [15]. This coarse-grained view implies that the results of the simulation do not distinguish in what measure a given tissue deformation is achieved through cell intercalation or through cell shape change, Fig 2f and 2g.
At a continuum level, the mechanical effect of myosin contracting a meshwork of actin with a specific orientation can be modelled as a prestress with a specific orientation [20, 21]. This modelling of anisotropic actomyosin stress generation has already been succesfully applied to tissues undergoing morphogenesis [22, 23]. This allows us to ask the question whether the mechanism of intercalation is in itself the necessary step to convert the planar-polarised actomyosin activity into convergence and extension. Alternatively, the prestress generated by planar-polarised actomyosin may in itself cause tissue convergence and extension at a global scale, while local balances would govern how much of intercalation and how much of cell shape changes are incurred. In order to predict tissue-scale deformations, a mechanical model needs to have access to the reaction forces of any neighbouring tissue. The reaction forces of the other epithelia neighbouring the extending GB will depend on their own deformation. Therefore, it is necessary to include all of these epithelia in a mechanical approach. Here we propose a numerical technique that solves the flow generated by any patterning of actomyosin along the apical surface of the embryonic epithelia. Our modelling is based on some mechanical assumptions on how the apical actomyosin in the developing epithelium interacts with its environment, and on a rheological model of actomyosin itself [21]. The numerical simulation of flows on a curved surface is based on a novel finite element technique, which is not limited to potential flows as in previous literature [24], and for which we have proven accuracy properties [25].
Experimentally, it is now possible to image live embryos labelled with fluorescent proteins in toto, using light sheet imaging [26, 27]. In Drosophila, this allowed experimentalists to observe morphogenetic events happening in different regions of the embryo and investigate how these impact on each other [6, 9]. A current challenge is to quantify the corresponding deformations for the whole embryo as it has been done for smaller regions of tissue with limited curvature [6, 14, 15, 28, 29]. Another challenge is to develop numerical tools that permit the implementation of models over the whole embryo. Numerical methods taking into account the three-dimensional geometry of Drosophila embryo have been used to study the formation of the ventral furrow at gastrulation (reviewed in [30]) and germ-band extension [31]. However, the driving forces in the three-dimensional approaches were the observed cell-autonomous phenomenology rather than the patterning of actomyosin activation.
Here, without explicitly accounting for cell intercalations, we show that anisotropic myosin prestress can cause the global movements observed in Drosophila embryonic axis extension. We show that either a planar-polarisation of actomyosin in the germ-band or the pulling force due to the posterior midgut invagination are sufficient to generate a posterior-ward convergence and extension flow of the tissue, consistent with experimental evidence. Using existing movies of whole Drosophila embryos [6], we quantify cell flows and show that the numerical predictions are consistent with these. Finally, we show that a geometric feature of the embryo, the cephalic furrow, modifies the predicted flow and acts as a barrier for tissue deformation. This guides the convergence and extension flow towards the posterior of the embryo, therefore breaking the flow symmetry.
At the developmental stage of GB extension, the Drosophila embryo is made up of a single epithelial cell sheet that has an ovoid shape, with the cell apices facing outwards, see Fig 1. Gastrulation occurs immediately before GB extension, forming the ventral furrow, which begins to seal just at GB extension onset [6, 14]. For simplicity, we will consider in this paper only the times that follow the completion of mesoderm sealing, thus cells can be considered to be mechanically connected across the ventral midline, Fig 1d. The apical actomyosin cortex is mechanically coupled from one cell to the other by transmembrane adherens junctions [33], which, away from specific folds of the epithelium, are located within 1 μm of the apical surface [34]. From a mechanical point of view, the apical surface of the embryo can thus be seen as a thin layer of apical actomyosin seamed together by adherens junctions. Within this apical domain, the dynamics of active cell rearrangement and shape changes [13] and the causal planar-polarised recruitment of actomyosin [11, 12] mediate GB extension. In consequence, our mechanical approach is based on this thin active layer which we model as a thin shell, similarly to what has been done for the cortex of single cells, in combination either with an elastic material law [35] or a viscoelastic one [21]. For a given active force generation, the dynamics of deformation of this apical meshwork is determined by its material properties and by the global mechanical balance [4], including interactions with neighbouring fluids and structures further apically and basally. On the basal side of this meshwork, baso-lateral cell membranes, cell cytoplasms and nuclei (Fig 1d and 1e) are not known to actively deform, and have been shown to flow as a viscous medium during gastrulation [32], immediately prior to GB extension. This passive behaviour implies that they are felt only as a drag (viscous friction). Note that there is no report of the presence of extra-cellular matrix on the basal side of this epithelium. On the outer side of the apical meshwork is the perivitelline membrane, but no specific adhesions bind them together, and the perivitelline liquid can play the role of a lubricating fluid between the two. In a first approximation, these two effects result in a friction force per unit area equal to the product of the velocity v by a friction coefficient cf. In terms of forces, the balance of the forces tangentially to the surface is thus:
∇ Γ · σ = c f υ . (1)
The stress tensor σ is the tension in the apical shell of geometry Γ, and ∇Γ⋅ is the surface divergence operator [36].
In order to obtain a closed model (i.e., a self-sufficient one), we need to supplement mechanical balance with a material law which links the stress to the deformations of the apical shell, and, in the present case, to the myosin activity also. We have recently derived and validated such a material law [21] by quantitative comparison of predictions of forces exerted by the actomyosin cortex of single cells with experimental measurements. The main ingredients of this modelling are the elastic resistance of actomyosin to deformation, which gives its short-term response, the turnover of actin and actin-crosslinking proteins ranging from seconds to minutes [37], which leads to a long-term liquid-like behaviour, and the myosin activity, which endows the cytoskeleton with emergent material properties [21]. It is worthwhile to note that, while the linear theory of transiently crosslinked gels does not lead to anisotropic material properties [38], the myosin activity generates an anisotropic stress, see supplementary material in [21]. For a tissue, the material ingredients are similar, with the addition of cell rearrangements which provide an alternative cause of elastic stress relaxation. We have shown that indeed, the same material law can describe successfully the rheology of early epithelia [39], although the value of parameters can vary. In a linear approximation, the apical actomyosin of embryos can thus be expected to have a viscoelastic liquid behaviour which can be written in the general form:
τ α σ ∇ + σ - 2 η ε ˙ ( υ ) - η b ( ∇ Γ · υ ) P = σ a (2)
where τα is the relaxation time of actomyosin, σ ∇ the objective derivative of the stress tensor, η and ηb are effective shear and compression viscosities of actomyosin,
ε ˙ = 1 2 P ∇ Γ υ + ∇ Γ υ T P
is the rate-of-strain tensor, and P is the projection tensor onto Γ, which is also the identity tensor on the surface. The tensor σa describes the myosin contractility, and can be understood as a prestress: because of myosin action, the meshwork of actin is continuously being offset from its stress-free configuration. The pre-stress is proportional to the myosin concentration and rate of power strokes [21, 39]. The relaxation time τα was found to be around 15 minutes in our previous work [21], which means that for the 90-minute germ band extension process, we are interested in times longer than relaxation. We have also shown for similar equations in another biophysical context and in one dimension [40] that the term τ α σ ∇ did not introduce marked qualitative features to the flow for time scales as short as the relaxation time itself. For the sake of simplicity, we will thus neglect the term τ α σ ∇ in what follows. Fig 2h illustrates the terms that sum to the tissue stress.
With this hypothesis, the mechanical balance Eq (1) and the constitutive Eq (2) can be combined into a single equation:
∇ Γ · σ a = - 2 η ∇ Γ · ε ˙ ( υ ) - η b ∇ Γ ∇ Γ · υ + c f υ (3)
On the left-hand side of Eq (3) is the myosin term, which provides the energy for the motion. On the right-hand side are two terms corresponding to energy dissipation: forces of friction with the actomyosin’s environment, and the viscous forces, corresponding to the cost of deforming the apical shell. These do not distinguish between inter-cellular dissipation, i.e. the mechanical energy spent during cell rearrangements in breaking cell-cell adhesive bonds and the one gained when establishing new ones; and intra-cellular dissipation, such as the cost of deforming the actomyosin cortex. Rather, these are lumped together, and taking a linear approximation, represented by effective viscosities η and ηb.
The myosin contractility term σa is the source term that provides energy to the system and causes the deformations. We have shown theoretically and verified experimentally [21, 39] that to a first approximation it is proportional to the local myosin concentration. However, when actomyosin recruitment is anisotropic, the tensor σa is also anisotropic, see supplementary materials in [21], and can be decomposed into a spatially-dependent intensity σa and an orientation tensor A, σa = σa A. In the case of GB extension, it is observed experimentally [11, 19, 42] that myosin is activated in the GB, thus σa > 0 in the GB, and is recruited along DV-oriented cell junctions, thus A = eDV ⊗ eDV, see Fig 2.
Note that we impose mathematically that velocities are tangential with the embryo surface (see S1 Text). Indeed, this is what is observed experimentally, we have thus made the simplifying choice of assuming this rather than try to predict it. A model bypassing this hypothesis would need to be significantly more complicated, as the force balance in the normal direction would need to be calculated in addition to the one in the tangent plane. This force balance should include the pressure difference between the periviteline liquid and the embryo interior beyond the apical membranes (yolk and cytoplasms) and the viscous drag from these structures, and also the tension and bending forces in the cell apices. Although we do not model this, the force needed to keep the flow tangential in our simulations is calculated, and is the Lagrange multiplier associated with the tangentiality constraint.
In order to solve Eq (3), we use a finite-element approach. In [24], a related problem is addressed, and a finite-element method is described to solve surface-incompressible flow problems. This restriction to surface incompressibility allows the authors to reduce the problem to a PDE in terms of a scalar unknown, namely the stream function. Here we want a more generic approach allowing to address flows of finite surface-compressibility ηb. We have thus developed a Lagrange-multiplier approach which can approximate the solution of vectorial equations set on a curved surface of ℝ 3 [25], see S1 Text. The resulting problem is discretised with finite elements using a triangle mesh [43] of the embryo shape. The finite element free software rheolef is used to implement the method and its accuracy is verified on a test problem.
Using the above model and the numerical technique described in S1 Text, we obtain numerical approximations of the solution of this problem for a geometry Γ closely mimicking the shape of Drosophila embryo. Fig 3 shows an example of a simulation result. The green colour codes σa(x), the location where myosin is assumed to be activated along dorsoventral direction A = eDV ⊗ eDV, and the arrows are the predicted velocity of the surface displacement of the apical continuum. This predicted flow is strongly dominated by two laterally-located vortices, which have their centre slightly dorsal from the edge of the GB region. They are rotating such that the velocity in the GB is strongly towards the posterior. The vortices are situated in the tissues located dorsally of the germband (future amnioserosa), which will thus be extensively deformed by the flow. This is indeed the case in Drosophila development, although part of the deformations in vivo may be related to PMG invagination [6], which is not included in this simulation (see below).
In order to compare this flow qualitatively with the actual movements of epithelia during Drosophila germband extension, we have calculated the displacement of the centroid of cells in vivo. We have imaged a wildtype Drosophila embryo expressing plasma membrane associated GFP markers using light sheet microscopy (specifically mSPIM, multidirectional selective plane illumination microscopy) to observe the whole embryo volume throughout germband extension in four perpendicular views then reconstructed to four-dimensional movies [6]. Using custom built software (see S1 Text) we mapped apical cell trajectories on one lateral surface of the embryo. In Fig 1a, the displacement between two consecutive frames during the fast phase of GB extension is shown. Such in toto tracking permits an account of the tissue movement across the whole of the embryo simultaneously, compared to previous microscopy techniques which were restricted to the observation of a limited view field [9]. The global features of the flow predicted by our myosin-activity based simulations are also apparent in this tracking. There is an embryo-scale vortex on each lateral side of the trunk part of the embryo, with the same direction of rotation between simulations and observations. The morphogenetic flow is much reduced in the head region, in simulations this reduction is even stronger, and velocities in the head region are less than 5% of the maximum velocity. Discrepancies between the model prediction and the actual flow can be reduced when biophysically relevant mechanical parameters are adjusted, and when the PMG invagination effect is included: this is the objective of the next sections.
It has been shown [6, 7, 14] that GB extension is not solely due to the action of planar-polarised myosin within the GB, but also to the pulling force that another morphogenetic movement causes, namely the invagination of endoderm in the posterior region, also called post-midgut (PMG) invagination.
Our present numerical approach does not allow us to simulate deformations of the surface in the normal direction, which would be necessary to simulate the PMG invagination process. However, it is possible to mimick the effect of PMG invagination on neighbouring tissues by simulating an in-plane isotropic contraction of a posterior patch of tissue, Fig 2b. Fig 4e shows that this does generate an extension of the GB area, although the deformation on the dorsal side is greater in this case. Combined with planar-polarised myosin action in the GB, Fig 4b–4d shows that posterior contraction does modify the flow pattern substantially and provides a complementary cause of GB extension, consistently with the experimental studies cited above. In fact, because of the linearity of Eqs (1) and (2), the superposition principle applies and the flows shown in Fig 4b–4d can be written as the weighted sum of the flows with planar-polarised myosin only, Fig 4a, and posterior contraction only, Fig 4e. The location chosen for posterior contraction corresponds to the location of PMG in early GB extension, Fig 1b, rather than the stage for which we have tracked cell flow, Fig 1a. In spite of this, the agreement with the experimental observations, Fig 1a, is improved by the addition of posterior contraction. The flow within the posterior region itself is not relevant to compare, since the numerical method does not allow for the invagination in itself. In neighbouring tissues, posterior contractility modifies the predicted flow on the dorsal side, creating a posterior-ward flow in a small region close to the contractile patch, and reducing the flow rate in the rest of the dorsal side. In particular, the posterior contractility in Fig 4b leads to flow rates of similar relative magnitude to the ones observed experimentally in Fig 1a.
Although in toto imaging of Drosophila embryos is now possible using SPIM to track the morphogenetic movements which our model attemps to predict [6, 9, 26], see also Fig 1a, a global cartography at the embryo scale of myosin localisation and activation during GB extension is is not available yet. The patterning of gene expression upstream of Myosin II planar-polarised recruitment, on the other hand, is accurately described [13]. We tested several configurations of the extent of the region of GB where myosin is anisotropically localised, Fig 5. We find that the global flows present a common aspect and GB extension is consistently obtained. However, some features vary, and in particular an anterior-wards backflow develops along the ventral midline when the myosin activation zone extends more laterally, Fig 5b. These backflows are attenuated or disappear when the PMG contraction is large enough, by superposition with flows in Fig 4e.
In terms of the surface deformation of the epithelium, it is seen in Fig 5c and 5d, that the flow is qualitatively the same although the location of the boundary of the myosin activated region directly changes the location of the strong gradients of strain rate. Within each region, the simulation reveals nontrivial variations of the strain rate. We note in particular strong peaks of deformation rate close to the boundaries of the myosin activated region (Fig 5d), which are not reported in Drosophila. These peaks are strongly reduced if the myosin prestress is assumed to change smoothly rather than abruptly at the boundary of the region, as will be seen below. Overall, the similarity of the flows observed from zmax = −0.2R (Fig 3) and zmax = 0 (Fig 5a) and their qualitative agreement with experimental observations (Fig 1a) is indicative that the system is robust with respect to the exact pattern of myosin activation.
Four parameters appear in Eq (3): the magnitude of myosin prestress σa, the friction coefficient cf, and the two viscosities ηb and η. Two of these parameters, σa and η, set respectively the magnitude of stresses and velocities, leaving two free parameters: ηb/η, which is nondimensional; and cf/η, which is the inverse of the hydrodynamic length Λ.
The ratio ηb/η compares the bulk to the shear viscosity: if it is very large, then the flow will be nearly incompressible in surface, meaning that any surface element (and in particular, any cell) will conserve the same area through flow, and hence all deformations will be locally pure shear deformations. If the ratio is small, then the viscous cost of locally changing the apical area will be similar to that of pure shear, and, depending on the global force balance, area changes may dominate. Indeed, if ηb/η is ⅔ (such that the Poisson ratio is zero), then a unixial load (such as a perfectly planar-polarised myosin action could be supposed to produce) will result in an area change only and no pure shear at all, or, in developmental biology terms, in convergence only and no extension. This effect is illustrated in Fig 6, where ηb/η covers the range 10 to 103. In the latter case, the negative DV strain rate (convergence) exactly balances the positive AP strain rate (extension), whereas in the former case, convergence strongly dominates. The strain rates in all cases are not uniform across the ventral side depending on the AP position, with a marked decrease of the pure shear in the central part of the GB, however other factors are seen to affect this spatial distribution in the rest of the paper. Between ηb/η = 10 and ηb/η = 100, there is a switch from area-reduction dominated flow (area reduction rate 4 times AP strain rate) to a shear-dominated flow (area reduction rate of the same order as AP strain rate). Experimentally in Drosophila embryos, the deformations are not limited to pure shear but involve some area change [14]. By comparison of actual tissue flows in Fig 1a with our numerical result, the order of magnitude of ηb/η can be expected to be of the order of 10 to 100.
The hydrodynamic length Λ = η/cf is the characteristic length within which shear stress will be transmitted within the actomyosin. Beyond this length, friction with the exterior will balance the internal stress. It thus indicates the length over which the effects of a local force are felt. We tested the effect of a hydrodynamic length either much larger, much smaller or comparable to the size of the embryo (R, the radius of a transverse section, Fig 1). The results shown in Fig 7 show how the action of myosin gives rise to a more local flow pattern when the hydrodynamic length is small. This localisation is around the areas in which there is a gradient of myosin activity. In areas of uniform myosin activity, the resulting effect is a uniform tension (see a similar effect in models of cells plated on a substrate, e.g. [40, Fig 3b]). It can be seen on Fig 7e and 7f, that indeed the strain variations are more abrupt when hydrodynamic length is small, whereas the long range interactions allowed by a very large hydrodynamic length give rise to an embryo-scale flow. On the whole, flows which reproduce experimental observations better are obtained when Λ is of order 1 or more, that is, the hydrodynamic length is comparable to or larger than the GB width in DV. Indeed, there is no strong localisation of the flow features in our experimental results, Fig 1a. This is consistent with the order of magnitude of 10 to 100 μm found by laser ablation in other systems [44].
In simulations, Fig 3, it is seen that the flow follows the cephalic furrow in a parallel way. Thus we wondered whether the presence of the CF could be important for the flow pattern observed. To test this, we performed the same simulation on two different meshes, one featuring the CF and the other without it. In the absence of a CF, the flow at the anterior boundary of the GB does not deviate laterally but continues towards the anterior, Fig 8e. The flow field is thus much more symmetrical than in when the CF is included in the model, Fig 8b. From the mechanics, we expect that if both the geometry of the embryo and the myosin localisation patterns are symmetric, then the flow will be symmetric too (see Fig 8d for a verification of this). In real embryos, two sources of asymmetry arise: the localisation of the posterior boundary of the region of planar-polarised myosin recruitment pattern, and the invagination in the posterior midgut. Blocking it however does not completely suppress the posterior-ward extension of GB [7]. Our results suggest that another asymmetry could originate from the geometry of the embryo, with the CF acting as a barrier resisting flow towards the anterior. Indeed, if one introduces this geometric feature at one end of the otherwise perfectly symmetrical embryo, the flow is strongly asymmetric towards the posterior end, Fig 8a. In experimental embryos, the local influence of the CF on the flow is visible in tracked in toto images, Fig 1a, where a reduction of the magnitude of the velocity is seen in the row of cells posterior to the CF. The numerical simulations bring the additional understanding that, in addition to this local effect, this mechanical feature has a global influence on the morphogenetic flow and orients GB extension towards the posterior.
Using the qualitative study of the influence of each parameter (ηb/η, cf/η) and of the patterning of myosin activation, we propose a choice of simulation settings which can be expected to reproduce the features of GB extension flows. Indeed, the equations being linear, the superposition principle guarantees that the effect of each of these causes simply add up. In Fig 9a, the magnitude of PMG strain rate is as in Fig 4b, ηb/η is chosen as in Fig 6b, the hydrodynamic friction as in Fig 7b. The myosin prestress pattern is close to the one in Fig 5a, but with a graded decrease of prestress when reaching the boundary of myosin activated region. Fig 9b–9e show the rate of strain along lines of interest on the surface. Compared to similar plots where myosin prestress is not graded close to the boundary, they have the same global features but no peaks close to the myosin activation boundary: compare e.g. Figs 5d and 9e. This is indeed due to the gradation of myosin prestress, since a simulation where only this parameter is altered restores the peaks (S1 Fig).
In Fig 9b–9e, we also plot real data of tissue strain based on the velocity of cell centroids on the surface of an embryo at one time point of GB extension, shown in Fig 1a [15] (see also Methods). In order to do this comparison, we use the known value R = 90 μm, and adjust manually the time unit of the simulations (one scaling parameter). It is seen that the spatial dependence of each quantity of interest follow similar patterns of variations in the simulation and the real data.
Along the AP direction, Fig 9b and 9c, there is a systematic shift of the position of the first peak value, which is located immediately posterior of the CF (x = −100 μm) in simulations, and straddles it in real data. This may be due to the incompleteness of the mechanical model of the CF, which in simulations appears as an immoveable fold, while in reality this fold has some limited movement, and deepens over time during GB extension as cell apices enter it [45]. In spite of these differences, the magnitude and shape of the peaks of strain associated with the CF are accurately reproduced in the simulations.
Further into the GB, −25 μm ≲ x ≲ 175 μm, simulations and real data exhibit a plateau of negative DV strain (convergence), Fig 9c, and positive AP strain (extension), Fig 9b, with little area variations, Fig 9d. as already discussed in Fig 6d. At both the posterior and anterior (CF) end of the GB, there are regions in which the sign of the rate of strain invert. This is consistently observed in experiments and simulations. In these regions, there is thus a local convergence-extension process in the direction orthogonal to the one in the GB. In real data, part of the convergence in AP is due to the loss of tissue as it goes into the CF and deepens it [45], and there is a corresponding peak of area decrease. In simulations, the AP convergence at the two ends of the GB is driven by the push from the AP extending GB, and the DV extension to the resistance to area changes, see also Fig 6d.
Along DV, Fig 9e, the current tracking technique of 3D SPIM data offers a range limited to ca. 90 degrees in DV. Over this range, the agreement in terms of the magnitude of the DV strain is rather good, but its rate of variation along the DV direction is not accurately predicted. This may be linked with an inacurate hypothesis on the gradation of myosin prestress close to the boundary of the GB, since S1e Fig exhibits a very different slope from the simulation in Fig 9. Dorsally, simulations and experiments agree on the existence of an AP convergence and a DV extension. It is consistent with observations in [9] that cells in this dorsal region extend in the DV direction while not rearranging and incurring little area change.
Overall, there is a good agreement in qualitative features in strain rates using the parameters that have been chosen based on flow features in the previous sections. That the latter agreement would lead to the former is not trivial, and Figs 6d–6f, 7e and 7f show that similar flows can exhibit different features in terms of strain rate.
Our simulations are based on the knowledge of the embryo geometry, which includes the global shape of the continuum formed by epithelial cell apices and also the fold formed by the cephalic furrow (CF), and an assumption of the pattern of myosin expression over this surface and its planar polarisation. A mechanical model (Eq (3)) then allows us to predict the stress and flow field over the surface of the whole embryo, using a novel numerical technique that can solve the equations on the embryo surface in three-dimensions [25]. The mechanical model is based on a liquid-like constitutive relation for the actomyosin of the apical continuum, following [21, 39]. We only obtain a snapshot of the flow field that corresponds to a given myosin activation pattern. In the course of GB extension, this pattern is transported by the flow [13], and the flow field evolves in its details while preserving its global features [9]. Because there is no time evolution term in our model, Eq (3), there is a unique instantaneous flow field for a given pattern of myosin prestress σa. At a given time point, the results (Fig 9) compare well with corresponding observations of the morphogenetic movements of GB extension Fig 1a, which we track over the three-dimensional surface of the embryo using in toto imaging and tracking software [15]. This agreement suggests that the model captures the essential balance of stresses originating from the contractility of actomyosin. In order to go further into model validation, the next step will be to apply the model to measured embryo shape and myosin localisation patterns, which could in the future be obtained by in toto embryo imaging. This would allow the model to be validated by testing its ability to predict the flow that corresponds to each successive myosin localisation pattern and CF position in the course of GB extension.
We note that in order to obtain a good agreement of model predictions with observations in terms of strain, we have had to use a graded intensity of myosin prestress at the boundary of the planar-polarised actomyosin region, Fig 9, rather than use a sharp change, S1 Fig. It remains to be shown whether such a gradation of planar-polarised actomyosin can be observed experimentally. The DV patterning of myosin localisation, although important for the precise localisation of flow features, is not crucial to achieve a convergence-extension motion with the correct orientation, Fig 5. For a planar-polarised actomyosin region extending far laterally however, our simulations predict backflows close to the midline. Such backflows are not observed experimentally, although mutants such as torsolike that present planar-polarised myosin but not PMG invagination form ectopic folds in the germband [6], which could be due to a buckling phenomenon. The mechanical role of the invaginated mesoderm may have to be taken into account in order to improve our modelling of the midline region, Fig 1d.
In addition to the actomosin patterning, two mechanical parameters govern the flow we predict, the ratio of apical surface resistance to compression relative to its resistance to shear, and the hydrodynamic length. Both have a more general biophysical relevance (see e.g. [44]). The material parameters of the cortical actomyosin in Drosophila are not known, and quantifying them experimentally is challenging [39], since this early step of development requires the presence of the rigid vitelline membrane that surrounds the embryo and prevents direct mechanical measurements from the exterior. Magnetic tweezers have been used [46], but the magnetic particles were not directly associated with the actomyosin cortex and thus measured other mechanical properties than those required by our model. Laser cuts can provide valuable information on the evolution of tension [47] or its anisotropy [6]. By comparison with a model of subcellular actomyosin, its material parameters can be obtained (medial actomyosin relaxation time, shear viscosity and friction coefficient) [44]. These values correspond to a subcellular system while we focus on the tissue scale, which may lead to quantitative differences with our case. Indeed, although the model in [44] is formally identical to Eqs (1) and (2), their friction coefficient includes the friction cost of movements of the actomyosin cortex relative to the cell membrane, whereas both apical membranes and actomyosin cortices flow in GB extension. In spite of this, the hydrodynamic lengths of 14 to 80 μm found for actomyosin cortex of single-cell C. elegans embryos and actomyosin ring of gastrulating zebrafish respectively in [44]. The ratio of bulk to shear viscosity ηb/η that we retain is much larger than the ratio 3 that could be expected from a simple 3D-isotropic modelling of actomyosin [44]. This is important in the model in order for the DV stresses to lead to AP extension as well as DV convergence, since it is the bulk viscosity that couples these effects, Fig 2d. The origin of this large resistance to area variations could be due to the combination of the volume constraint of each cell and of a basolateral regulation of its height, as has been hypothesised in cell-based models [30]. Alternatively, active processes may regulate the apical actomyosin cortex thickness and density.
The method can account for the fine geometrical detail of the embryo apical surface, such as the CF. It is observed experimentally that during GB extension, the cell displacements are much less in the head region than in the trunk, Fig 1a. However, there is no report of a specificity of the cytoskeleton of cells in the head at this stage that could account for a locally enhanced stiffness. Our simulations lead us to propose that the presence of the CF may be the cause for these smaller displacements. Geometric features such as the CF are important for the mechanical equilibrium, since forces are transmitted directionally and any curvature will modify the equilibrium. We show that the CF modifies strongly the flow we predict and acts as a barrier for deformations, Fig 8. This geometrical feature thus guides the convergence–extension flow towards the posterior end of the embryo.
Our results confirm that either of the two mechanisms whose elimination was seen to correlate with a reduction of GB extension [6, 7, 11, 14, 42] can be the direct mechanical cause of a flow towards the posterior in the GB. The results in Fig 4 indicate that the GB extends under the effect of the anisotropic prestress of GB planar-polarised actomyosin, but also under the effect of the pull from the invaginating PMG. However, the precise flow patterns differ in these different cases. An important and obvious next step is therefore to obtain experimentally the distribution and polarisation of myosin on the entire embryo surface across time, for example using SPIM, for wildtype and different mutants, which should allow us to quantify the parameters and test the predictive power of the model. In the interval, our theoretical work already sheds light on the fundamental mechanisms at play and how they integrate in the complex 3D geometry of the embryo to yield the morphogenetic events that are observed.
Based on our simulations and the timings reported in the literature, we can indeed articulate a mechanical scenario for GB extension. The endoderm contraction that leads to PMG invagination, the first event correlated with GB extension ([6], and Fig 4B in reference), starts several minutes before the onset of GB extension can be detected. In Kruppel mutants, for which the planar polarisation of myosin in GB is deficient, GB still extends [14], this is mainly due to PMG invagination [6]. The corresponding simulation is shown in Fig 4e, PMG contraction generates a flow towards the posterior that does extend the posterior half of the germband but decays rapidly in space. The presence of the cephalic furrow for this extension does not have a strong influence there, see Fig 8f. Thereafter, from shortly before the onset and in the course of GB extension, myosin becomes increasingly planar-polarised [19]. The direct consequence of this is a lateral flow from dorsal to ventral, causing convergence, that is, a negative rate of strain of GB along the DV direction. Due to a rather large value of ηb, which corresponds to the in-plane compressibility viscosity of actomyosin, this causes GB extension, see Fig 6. Experimental evidence of such an in-plane low compressibility exists [7, 14], although it is not clear whether this is a passive mechanical property of the actomyosin cortex or an active one [7, 41]. In the absence of the cephalic furrow, Fig 8e, this extension occurs evenly in the anterior and posterior directions, in presence of the cephalic furrow, the viscous cost of flowing around the posterior end is much less than the cost of flowing into the furrow, and planar polarisation driven GB extension is biased towards posterior, even if the contribution of PMG invagination is not accounted for, Fig 3.
The two prominent features in which the existence of cell-cell junctions are important in GB extension in wildtype (WT) Drosophila are the planar-polarised recruitment of myosin, which preferentially enriches DV-oriented junctions, and the medio-lateral cell intercalations [11, 13, 42]. Because the DV-oriented junctions present both the characteristics of being enriched in myosin and of undergoing shrinkage to lead to intercalation, these two effects have so far been studied in association. However, some mutants such as eve that lack planar polarisation of myosin can still exhibit some cell intercalation, although to a much lesser proportion than cell shape changes [14]. Here, using our modelling approach we can envision the reverse case of studying convergence-extension due to planar-polarised myosin activity but without explicit cell intercalation. We show that the anisotropy of planar-polarised myosin activity is sufficient to explain convergence-extension, without the need for an intercalation mechanism.
We conclude that a mechanical model that does not involve individual cells but only a continuum standing for the apical actomyosin connected from cell to cell by apical junctions can produce a flow with strong similarities to GB extension. In vivo, cellularisation is of course important for the planar polarisation of myosin in the GB, since the DV-oriented cell-cell junctions are the location at which oriented actomyosin structures assemble. It is observed that acellular embryos do not exhibit myosin polarisation [6]. In the model however, it is possible to introduce planar polarisation independently of cellularisation. This is what we do when defining an anisotropic myosin prestress σa. The fact that no further account of the cellularisation of the embryo is necessary in the model suggests that at the tissue scale, one can address morphogenetic questions by considering ensemble displacements. In this approach, the effect of cell intercalation, which is governed by planar-polarised junctional myosin, is thus not directly taken into account, but rather encapsulated in a global tissue strain rate and its associate viscosities η and ηb, which also include the cell deformation [15], Fig 2f.
This tissue strain rate ε ˙ and the corresponding tissue-scale tension σ are related by the constitutive relation, Eq (2), which includes the contractility term σa resulting from planar-polarised myosin activity, and is thus the only term bearing a trace of the embryo’s cellularised organisation. The respective values taken by ε ˙ and σ locally depend on the mechanical balance, i.e. both the local myosin activity and the tension transmitted by neighbouring tissue, see Fig 2d. In the context of convergence-extension caused partly by invagination of the PMG, [14], it has been proposed that cell intercalation could relax the stress by allowing cell shape changes in the GB as it is extended by an extrinsic force. Here we propose that the crucial role of cellularisation in planar-polarised tissue convergence-extension lies in the mechanism through which oriented contractile actomyosin junctions are created, and not in the subsequent tissue deformation. This suggests that if planar-polarisation of actomyosin could be achieved through some other mechanism (as is the case in the Drosophila tracheal tubule [23]), convergence and extension could still be obtained in an acellular tissue, thus in the absence of cell intercalation. In cellularised tissue, an important role of cell intercalation could thus be the relaxation of the cell strain generated by the convergence and extension process, as was already suggested [14].
Planar-polarised myosin in the GB is known to generate a global flow at the surface of the embryo. Our simulations show that the global flow which is generated by such mechanical activity is dependent on the pre-existing geometry of the embryo, such as the presence or absence of the cephalic furrow. Thus, a prior morphogenetic movement such as cephalic furrow formation can affect further movements via mechanical interactions only.
This “messaging” proceeds via the establishment of a different mechanical balance depending on the geometry of the embryo, rather than the diffusion of a morphogen, e.g. [48]. In the early embryo, the distance over which these forces are transmitted is likely to be much larger than in later organisms, as there is no extra-cellular matrix structure that will relieve actomyosin from part of the stress. Indeed, we find that the hydrodynamic length is likely to be at least the width of the GB, consistent with laser ablation results [44], which implies direct mechanical interaction at this scale.
This mechanical messaging behaves differently from biochemical messaging. Its speed of propagation is the speed of sound in the force-bearing structure, here, the actomyosin. It does not propagate in an isotropic way but in a more complex directional one, and contains directional information. Regions of interest within the embryo should thus not be treated as isolated systems, since a distant geometric property of the embryo can have a direct impact on the mechanical stress felt locally when intrinsic forces are being generated.
This prompts further development of computational tools such as the one we present. Tangential flows on curved surfaces are also observed in other epithelia (such as follicular epithelium of Drosophila ovaries), but is also relevant to cortical flows in single cells, prior to mitotic cleavage for example. Mechanical approaches of flat epithelia have shed light on many aspects of tissue growth and dynamics [19, 49–51], in particular at the scale of a few cells, which is the relevant one for cell rearrangements. At the other end of the spectrum of tissue dynamics, 3D phenomenological models of shape changes during ventral furrow formation have been proposed [30]. Here we propose a first step in bridging the gap between these approaches, with the objective to be able to address complex morphogenetic events in their actual geometry, and thus to fully account for the influence of current morphology on the mechanical balance that leads to further morphogenetic movements.
|
10.1371/journal.pntd.0000838 | Mass Treatment with Azithromycin for Trachoma Control: Participation Clusters in Households | Mass treatment to trachoma endemic communities is a critical part of the World Health Organization SAFE strategy. However, non-participation may not be at random, affecting coverage surveys and effectiveness if infection is differential.
As part of the Partnership for Rapid Elimination of Trachoma (PRET), 32 communities in Tanzania, and 48 in The Gambia had a detailed census taken followed by mass treatment with azithromycin. The target coverage in each community was >80% of children ages <10 years. Community treatment assistants observed treatment and recorded compliance, thus coverage at the community, household, and individual level could be determined. Within each community, we determined the actual proportions of households where all, some, or none of the children were treated. Assuming the coverage in children <10 years of the community was as observed and non-participation was at random, we did 500 simulations to derive expected proportions of households where all, some, or none of the children were treated. Clustering of household treatment was detected comparing greater-than-expected proportions of households where none or all of children were treated, and the intraclass correlation (ICC) was calculated. Tanzanian and Gambian mass treatment coverages for children <10 years of age ranged from 82–100% and 62–99%, respectively. Clustering of households where all children were treated or no children were treated was greater than expected. Compared to model simulations, all Tanzanian communities and 44 of 48 (91.7%) Gambian communities had significantly higher proportions of households where all children were treated. Furthermore, 30 of 32 (93.8%) Tanzanian communities and 34 of 48 (70.8%) Gambian communities had a significantly elevated proportion of households compared to the expected proportion where no children were treated. The ICC for Tanzania was 0.77 (95% CI 0.74–0.81) and for The Gambia was 0.55 (95% CI 0.51–0.59).
In programs aiming for high coverage, complete compliance or non-compliance with mass treatment clusters within households. Non-compliance cannot be assumed to be at random.
| Trachoma, an infectious disease, continues to cause blindness. A great deal of the trachoma burden is concentrated in developing countries. The World Health Organization recommends mass treatment for entire communities in trachoma-endemic regions. In 32 Tanzanian and 48 Gambian communities with trachoma, mass treatment was directly observed following a census. Community coverage was mostly greater than 80%. Larger-than-expected proportions of households where all children were treated and where none of the children were treated were found in each country. Household clustering of treatment was higher in Tanzania compared to The Gambia. However, children who were not treated were not more likely to be infected compared to children who were treated. We found that treatment and non-treatment within communities does not occur at random but rather clusters within households. These findings impact the design of future coverage surveys and suggest that further research evaluate factors that are associated with familial non-compliance.
| Trachoma is the leading infectious cause of blindness [1]. As the most common ocular neglected tropical disease, active trachoma is estimated to affect 40.6 million people worldwide, and another 8.2 million experience visual impairment or blindness [2]. Trachoma is largely confined to regions of extreme poverty [3].The World Health Organization (WHO) African region contains more than two thirds of all active trachoma cases and approximately 47% of all trichiasis cases [2].
The WHO recommends azithromycin mass drug administration as a key part of the Surgery, Antibiotics, Face-washing, Environmental change (SAFE) strategy for eliminating trachoma. The WHO advocates a treatment coverage goal of at least 80% to be effective [4]. Although evidence is needed to determine the impact of coverage at different thresholds, national trachoma control programs need to be able to measure non-participation in order to meet distribution targets. Because active trachoma and infection largely reside in preschool age children within communities, antibiotic treatment should particularly target this group for maximal effectiveness [5].
Treatment coverage surveys often carry an implicit assumption that missing treatment occurs at random. There are no data to support this conjecture, and this is reason for concern. There are ample data that trachoma clusters in families and in neighborhoods [6]–[10], and that transmission within households and across households does occur [10], [11]. If treatment also tends to cluster, and is differential by infection status, then even the effect of high coverage may be compromised. Moreover, any treatment clustering will affect the precision of coverage estimates, and how one designs coverage surveys.
We examined the clustering of treatment at the household level using data at baseline from 32 communities in central Tanzania and 48 communities in The Gambia who are enrolled in the three-year Partnership for Rapid Elimination of Trachoma (PRET) project.
The study was approved by the Johns Hopkins Medical Institutional Review Board, the Tanzanian National Institute for Medical Research, the London School of Hygiene and Tropical Medicine Ethical Review Board, and The Gambia Government/Medical Research Council Joint Ethics Committee.
All individual participants for this study provided consent. All adults provided informed written consent in both The Gambia and Tanzania.
The study was conducted in 32, geographically distinct, communities within the Kongwa district of Tanzania. Communities were eligible if they were located in the Kongwa district, the community leadership gave consent, and the estimated community prevalence of trachoma was greater than or equal to 20% and less than 50% in preschool age children. We excluded communities where the estimated population was greater than 5,000 persons. In Tanzania, a household was defined as persons who used a unique doorway to sleeping quarters.
In The Gambia, Census Enumeration Areas (EAs) were used as communities because villages (collection of households with a distinct name) varied so much in size. The estimated prevalence of trachoma for most EAs was less than 20% and no EA is larger than 5,000 persons. The EAs are hereafter referred to as communities.
Forty-eight Gambian communities participated, located within four districts: Lower Baddibu, Central Baddibu, Foni Bintang Karanai, and Foni Kansala. Communities were eligible if they were located in the target district, community leadership gave consent to participate in the trial, and they had a prevalence of active trachoma greater than 5% in preschool age children, based on the best available data (no community had more than 50% trachoma). In The Gambia, a household is defined as persons who all eat together.
These communities varied in size between countries. In Tanzania, communities averaged around 1500 population. In The Gambia, communities' average population size was 700 persons.
Details on the PRET project methods, enrollment, and study procedures are described elsewhere [12] and summarized below.
In Tanzania, prior to mass treatment, trained research staff conducted a census of every household in the 32 communities. Baseline census information, the names, age, and gender of all persons residing in the household, was obtained from the head of household. Education completed by the head of household, distance to water, and presence of latrine were also collected for each household. These data were used to generate treatment books for mass distribution. In four randomly selected communities, all children less than ten years were screened at baseline for active trachoma (trachomatous follicular (TF) and/or trachomatous inflammation-intense (TI)) and ocular infection with Chlamydia trachomatis, using a commercially available test, Amplicor, which tests for the chlamydial plasmid (Roche Diagnostics, Indianapolis, IN). Data from this group were used to determine if non-compliance with mass treatment was associated with infection status at the baseline census.
In The Gambia, a census was completed in each of the 48 communities by trained research staff interviewing the head of the household. The data were used to generate treatment books to monitor mass treatment.
Mass treatment was a single dose of azithromycin, 20 mg/kg up to one gram. Azithromycin was offered to all members of the community aged over six months. For children younger than six months, tetracycline eye ointment was offered, and in The Gambia, pregnant women were also offered tetracycline eye ointment as an alternative. The regimen consisted of six weeks of tetracycline ointment, twice a day. Guardians of children less than six months were provided with tetracycline, and were responsible for complying with the regimen. In both countries, the minimum target for coverage of mass treatment was 80% of children aged less than ten years.
In Tanzania, a team of Community Treatment Assistants (CTAs) was trained in each community and assigned specific neighborhoods. They were given treatment books based on the census lists. The CTAs were responsible for providing treatment to all community residents, on pre-announced days. Azithromycin was offered at a central location in the neighborhood and, if necessary, at the household, was directly observed and noted in the treatment books. The CTAs were each provided a small incentive of 1000 TsH ($0.90) at the conclusion of mass treatment, if treatment verification showed that recorded treatment agreed with findings from the household treatment verification. Two to six CTAs were responsible for each community, and treatment days varied from two to five. Procedures identical to the National Program were carried out in Kongwa except that more supervision was provided and some communities were allowed additional days.
In The Gambia, a meeting was held with the community leaders to plan mass treatment for the community. There were typically one to two National Eye Care Programme (NECP) community ophthalmic nurses (CONs) in the community, and a variable number of ‘friends of the eye’ (nyateros) who were enlisted to help with distribution. Residents came to a central location within each community and treatment was directly observed. If any family member was missing, the family was asked to make sure the member came before the end of the day for treatment. The CON was notified in advance of persons who missed mass treatment, and asked to personally go and inform them of the subsequent treatment day in their community, and to advise treatment. One or two treatment days were offered. The drug distribution itself was provided by the NECP team, who received per diems of 200 dalasi/day ($7.88). This program was the NECP treatment for The Gambia.
Coverage data was derived from treatment registers completed by the CTAs. For each community, we calculated the total coverage as the proportion of the censused population who had received either a tube of tetracycline or an observed dose of azithromycin. We calculated coverage of children as the proportion of children aged less than ten years who had received a dose of azithromycin or topical tetracycline; the denominator was all children aged less than ten years resident in the community during census. For each community, we then took households with children aged less than ten years and calculated the proportion of households where all children were treated, where no children were treated, and where some of the children were treated. This constitutes the observed proportions for each community.
Next, data were stratified according to community and 500 model simulations of household treatment distributions were run. These simulations assumed that children were treated and not treated at random, and for each community, the coverage of children less than ten years for the simulated data was set equal to the observed coverage. For each community, the observed proportions of households where all/none/some of the children were treated were compared with the simulated data. The intraclass correlation coefficient (ICC) of treatment status for children of the same household is reported, as a measure of clustering.
Because of the large variability of household size in The Gambia, we compared the ICC of households with fewer than four children to the ICC of households with four or more children.
To address the question if non-treatment was differential by infection status prior to treatment in Tanzania, logistic regression models were employed using treatment as the dependent variable and infection status prior to treatment as the predictor. We corrected standard errors to account for the clustering at household level using the generalized estimating equation approach.
Approximately 46,634 individuals, including 17,332 children under ten years, resided in the 32 Tanzanian communities at baseline. The mean community population in Tanzania was approximately 1,457 people. Community size ranged from 703–2496. The mean number of persons per household was similar across Tanzanian communities, and the average number of children aged under ten per household was 1.7 (Table 1).
In The Gambia, 33,695 individuals, with 11,321 children below ten years, lived in the 48 Gambian communities (Table 1). The Gambia had an average community population of 702 people, with the size ranging from 327–1621. On average, there were approximately 11 individuals living in a Gambian household, and an average of four children aged less than ten years.
Coverage for both settings was high, reflecting the target of greater than 80% (Table 2). In Tanzania, the average coverage was 94% for children under age ten, ranging from 81%–100% in children 0–9 years and 65%–100% for all ages. The Gambian average coverage for children below age ten was lower, 89%, ranging from 62 to 99%. The range for total coverage of the population was 62 to 98%, with a mean coverage of 86%, similar to Tanzania.
In both Tanzania and The Gambia, most households had all children treated (Table 3). The percentage of households where all children were treated, and the percentage of households where no child was treated, were both greater than would be predicted if non-treatment occurred at random, although the evidence for clustering was less strong in The Gambia (Figure 1).
In Tanzania, the ICC for treatment status of children within households was 0.77 (95% CI 0.74–0.81). The mean percentage of households where no children under ten were treated was 4.6% (95% CI 3.2%–6.1%) compared to the estimate based on simulation with the presumption of non-treatment at random of 2.2% (95% CI 1.4%–2.9%).The mean percentage of households where all children were treated was 90.3% (95% CI 87.8%–92.8%), compared to the estimate based on simulation of 87.7% (95% CI 84.4%–90.1%). Thirty of 32 communities (94%) had greater than the predicted proportion of households with none of the children treated, if non-treatment was at random.
In The Gambia, the ICC was 0.55 (95%CI 0.51–0.59). On average, the percentage of households with no children treated was 4.8% (95% CI 3.0%–6.7%) compared to the estimate from simulations of 2.1% (95% CI1.5%–2.7%). The mean percentage of households with all children treated was 74.2% (95% CI 70.1%–78.3%) versus the estimate from simulations of 67.0% (95% CI 62.0%–72.1%). Of the 48 Gambian communities, 34 (70.8%) had significantly higher numbers of households with no children treated, compared to prediction if non–treatment was at random.
Because the range of household size was much greater in The Gambia, we compared the ICC for households with less than four children aged under ten years, to the ICC for households with four or more children in the Gambian data. For smaller household size, the ICC was 0.60 (95% CI 0.54–0.65), compared to Tanzania which was 0.77. The ICC was significantly lower for Gambian households with four or more children aged less than ten years, 0.50 (95% CI 0.45–0.56).
In the four Tanzanian communities where infection status was available for all children under ten years of age, households where none of the children were treated were no more likely to have children with infection than households where at least one child was treated (odds ratio 0.82 (95% CI 0.54–1.27). The PCR swab positive infection rate prior to treatment in children who were not treated was not statistically significantly different than that for children who were treated, 16.8% versus 23.9%, p = 0.09.
Our study in two different settings shows that in communities that carry out azithromycin mass treatment with high coverage, non-treatment (as well as all-treatment) of children clusters in households. There was strong evidence that treatment (and non-treatment) did not occur at random. Compared to assumptions of random non-treatment, we demonstrated significant levels of treatment and non-treatment clustering within households in Tanzania and The Gambia. A number of community programs for other diseases have also observed household clustering of treatment [13]–[15], and we have now found a similar trend for trachoma mass treatment.
Tanzanian households were more homogenous with respect to treatment compared to The Gambia. In part, this appears to be driven by the larger household size in The Gambia, where we observed higher ICC for households with less than four children compared to households with four or more children. With increasing numbers of children in a household, the bar is clearly higher for reaching “all children treated.” However, the ICC for Tanzania with four or fewer children was still greater than for The Gambia. This may also reflect differences in the approach to obtaining high coverage, and the higher coverage achieved in Tanzania. For children aged 0–9 years, Tanzanian communities had a 94% coverage compared to The Gambia's coverage of 89%. As coverage levels among children drop, it appears that the likelihood of more partially covered households increases.
Our data demonstrate that non-participation clusters, and that national programs should develop strategies to identify and treat households that do not participate, with examples of possible reasons for non-participation supported by the literature [16]–[18]. Our results support the argument that participation in public health programs is dependent on a number of social factors, which differ among households.
Our findings provide evidence that non-participation in mass drug administration is a non-random event; therefore, coverage surveys using cluster-sampling design should include a design effect to ensure that they have an appropriate sample size. If surveys estimate the number of persons treated, then an assumption that non-treatment is at random will result in sample sizes that are underpowered, and have the potential for erroneous estimates whose directionality is unpredictable. The design effect for Tanzania, for example, is 1.6 for an average household size of 1.8 children; because of the larger cluster size in The Gambia, the design effect is 2.7. A greater design effect means more household clusters must be included to achieve a minimum appropriate sample size.
In Tanzania, we also showed that prior to treatment, households with no treated children were no more likely to have at least one child with ocular C. trachomatis infection as households with some/all children treated. Consequently, households opting out of the treatment are not more likely to be infected, and thus do not represent a disproportionate threat to re-emergence within the community. If infection had been associated with households where none of the children was treated, even programs with high target coverage would have a more challenging job maintaining any reduction in trachoma prevalence. We also showed that infection rates prior to treatment were no different in children who were untreated compared to children who were treated, so while infected children still represent a threat of re-emergence, they are not differentially located in the untreated group. There was insufficient infection at baseline in The Gambia to address this question.
Although our study did not measure risk factors associated with participation, other community-based studies have shown that household factors such as absence of family notification [19], lack of family support for treatment [20], and increasing distance from the distribution site [21] were associated with non-participation. However, there is a need to investigate further the household factors related to clustering of non-treatment of children. Such research would help trachoma control programs by deepening their understanding of factors that might be altered to improve familial mass treatment participation.
We observed that where coverage in children was higher, the households tended to be more homogenous with respect to treatment. It may be easier for CTAs to encourage a guardian who has had some of their own children participate in mass treatment to treat the remaining family members, than to convince a guardian who has not brought any children for treatment to participate. In addition, with increasing number of days of mass treatment, it likely becomes easier to return to households a sufficient number of times to reach persons who may have been away. Consequently, as efforts persist to improve coverage, it may be that most households will have all members treated, or households will cluster with all non-treated members, and the community will consist of relatively fewer numbers of partially treated households.
There are some limitations to our analyses. In both countries, target coverage was very high, above 80%, thus limiting the generalizability of findings to communities where coverage rates are much lower; however, this is the coverage target that is recommended by WHO for trachoma control programs. We also constrained, for the simulations, the overall coverage in the community to match the observed coverage. Other approaches could have been used, such as taking the overall coverage for all communities combined. However, we wanted to assess household clustering, so using the overall community level value as the target for each community let us simulate treatment of children at random within the community. Finally, CTAs may have been tempted to report better than average coverage. However, treatment verification procedures reported excellent correspondence between CTA records and verification (fewer than 1% discrepancies).
In summary, we measured treatment clustering within households of countries with moderate and low trachoma infection, Tanzania and The Gambia, in the context of achieving high coverage rates. Participation in azithromycin mass treatment was not at random. Most communities had higher numbers of household with either all-treated or none-treated children. As treatment clustering is at the household level, an evaluation of household risk factors related to clustering may assist in understanding this phenomenon and contribute to the development of trachoma control programs with higher coverage. Regardless, national trachoma control programs need a plan to capture children who may be missed in mass treatment campaigns, and ensure continued participation through multiple rounds.
|
10.1371/journal.pbio.1000172 | Persistent cAMP-Signals Triggered by Internalized G-Protein–Coupled Receptors | G-protein–coupled receptors (GPCRs) are generally thought to signal to second messengers like cyclic AMP (cAMP) from the cell surface and to become internalized upon repeated or prolonged stimulation. Once internalized, they are supposed to stop signaling to second messengers but may trigger nonclassical signals such as mitogen-activated protein kinase (MAPK) activation. Here, we show that a GPCR continues to stimulate cAMP production in a sustained manner after internalization. We generated transgenic mice with ubiquitous expression of a fluorescent sensor for cAMP and studied cAMP responses to thyroid-stimulating hormone (TSH) in native, 3-D thyroid follicles isolated from these mice. TSH stimulation caused internalization of the TSH receptors into a pre-Golgi compartment in close association with G-protein αs-subunits and adenylyl cyclase III. Receptors internalized together with TSH and produced downstream cellular responses that were distinct from those triggered by cell surface receptors. These data suggest that classical paradigms of GPCR signaling may need revision, as they indicate that cAMP signaling by GPCRs may occur both at the cell surface and from intracellular sites, but with different consequences for the cell.
| Cells respond to many environmental cues through the activity of cell surface receptor proteins, which sense these cues and convey that information to signaling molecules inside the cell. G-protein–coupled receptors (GPCRs) form the largest eukaryotic family of plasma membrane receptors. They convert the information provided by extracellular stimuli into intracellular second messengers, like cyclic AMP (cAMP). After prolonged stimulation, they are internalized inside cells, an event that to date has been thought to terminate the production of second messengers. Though many of the key steps of GPCR signaling are known in detail, precisely how signaling and termination actually occur in time and space (i.e., in subcellular compartments or microdomains) is still largely unexplored. To observe GPCR signaling in living cells, we generated mice expressing a fluorescent sensor that allows monitoring the intracellular levels of cAMP with a microscope. We utilized this system to study, directly in native thyroid follicles, the signal sent by the receptor for thyroid-stimulating hormone (TSH). Our findings indicate that TSH receptors are internalized rapidly after activation but continue to stimulate cAMP production inside cells and that this sustained, cAMP production is apparently required for localized activation of downstream components. These data challenge the current model of the GPCR-cAMP pathway by suggesting the existence of previously unrecognized intracellular site(s) for cAMP generation and of differential signaling outcomes as a result of intracellular GPCR signaling. Such intracellular site(s) may provide specialized signaling platforms, thus contributing to the spatiotemporal regulation of cAMP production and to signaling specificity within the GPCR family.
| G-protein–coupled receptor (GPCR) signaling is thought to involve a series of steps occurring at the cell surface: coupling of receptors to G-proteins, activation of G-proteins, and ultimately, triggering of G-protein-regulated effectors (i.e., adenylyl cyclase, phospholipase C, calcium channels, GIRK channels, etc.) [1]. Soon after activation, many GPCRs desensitize in a process that involves phosphorylation by G-protein–coupled receptor kinases (GRKs) and binding of β-arrestins [1]. Subsequently, most GPCRs are internalized via clathrin-coated pits or other less characterized pathways, and are either dephosphorylated and recycled back to the cell surface or targeted to lysosomes for degradation [1].
Although receptor internalization was originally considered to contribute to desensitization by reducing the number of receptors present on the cell plasma membrane, endocytosis has been subsequently and unexpectedly found to promote or even be required for receptor resensitization [1],[2]. Furthermore, novel data suggest that receptor internalization does not always lead to signal termination. This possibility has been clearly demonstrated for tyrosine kinase receptors, such as the epidermal growth factor receptor (EGFR), that were shown to continue signaling after being internalized [3]–[6]. In the case of GPCRs, instead, internalized receptors are thought capable of switching to a “nonconventional” signaling pathway, i.e., a β-arrestin-mediated activation of the mitogen-activated protein kinase (MAPK) cascade [7]. A very recent study has revealed yet another type of intracellular GPCR signaling in yeast: Gpa1, the yeast homolog of Gα can be activated by pheromone receptors on endosomes, where it stimulates phosphatidylinositol 3-phosphate production [8]. Despite such recent data, there is a current consensus that activation of canonical G-protein effectors, such as adenylyl cyclase, by GPCRs occurs exclusively at the cell surface.
Describing the spatiotemporal dynamics of signaling cascades is a major goal of cell biology. In the case of GPCR signaling, this would imply answering fundamental questions such as: Where in the cell and for how long are GPCRs active after interacting with their ligands? Are there subcellular microdomains specialized for different types of GPCR-mediated signals? What are the functional consequences of GPCR internalization on signaling? Does signaling to cyclic AMP (cAMP) or other second messengers occur only at the plasma membrane, or are there additional sites of GPCR activity inside the cell? Most of these questions are still awaiting answers. The major reason for this relies on the fact that biochemical techniques, until recently the only ones available for such analyses, require cell disruption and therefore have limited temporal and, generally, no spatial resolution. To tackle those limitations, we and others have developed a series of genetically encoded fluorescent reporters that allow the direct visualization of key steps of GPCR [9]–[11] and cyclic nucleotide signaling [12]–[19], by means of microscopy techniques based on fluorescence resonance energy transfer (FRET). The introduction of these techniques has led to new insights into the mechanisms of GPCR activation and the biology of cAMP.
Despite the important advance represented by the introduction of FRET-based techniques, most studies required transfection of genetically encoded fluorescent reporters into primary cells or cell lines—thus, quite far from the physiological context. This is a fundamental issue, as the type, location, and concentration of each component, as well as the size and shape of cells, are expected to greatly influence the spatiotemporal features of signaling networks [20]–[22]. To be able to study GPCR-cAMP signaling in a highly physiological context, here we generated reporter mice with ubiquitous expression of an inert fluorescent sensor for cAMP. These mice were utilized to study the dynamics of a GPCR-cAMP signaling cascade, i.e., that activated by thyroid-stimulating hormone (TSH), within the intact multicellular functional unit that constitutes thyroid tissue.
To monitor cAMP levels in living cells and tissues, we generated transgenic mice (CAG-Epac1-camps) with ubiquitous expression of our previously described cAMP sensor (Epac1-camps) [14]. We followed the same strategy used to create green fluorescent protein (GFP) mice [23]. Instead of GFP, we cloned the Epac1-camps sequence (encompassing the yellow fluorescent protein [YFP], the cAMP binding domain of Epac1, and the cyan fluorescent protein [CFP]) under the control of the hybrid CMV enhancer/chicken β-actin (CAG) promoter (Figure 1A), and performed pronuclear injections of one-cell–stage mouse embryos with this construct. After several rounds of injections, we obtained ten PCR-positive pups, three of which showed a high level of body fluorescence and gave rise to transgenic offspring according to the Mendelian ratio. Careful analysis of isolated cells and tissues from these three lines demonstrated that two of them had a heterogeneous expression of the sensor, which was present in only 60%–80% of the cells. By contrast, the third line expressed Epac1-camps in virtually all cells and was therefore chosen for subsequent experiments. Figure 1 reproduces fluorescent images of the head and a series of organs isolated from the transgenic mice. Compared to wild-type littermates, these mice had high levels of sensor expression in almost all tissues and cells, excluding erythrocytes and hair. For example, we found high fluorescence in the eye and the skin (Figure 1B), as well as in the brain, heart, kidney, and ileum (Figure 1C). Interestingly, transgenic mice did not show any abnormalities and had a normal life expectancy, demonstrating that the presence of the sensor in most cells of the organism did not interfere with its proper development and physiological functions.
Next, we isolated several types of embryonic and adult primary cells from CAG-Epac1-camps mice to measure cAMP levels in real time. cAMP levels were monitored by FRET microscopy on live cells as previously described [14]. From E14.5 embryos, we obtained murine embryonic fibroblasts (MEFs) and cortical neurons, which showed high fluorescence. Stimulation of MEFs with the β-adrenergic agonist isoproterenol resulted in a robust decrease of the YFP/CFP ratio (Figure 2A), indicative of an increase of cAMP levels [14]. As already reported in other cell types [24],[25], this response was transient, due to the protein kinase A (PKA)-dependent activation of phosphodiesterase 4 (PDE4), which could be counteracted by addition of a specific inhibitor (rolipram). In cortical neurons, isoproterenol induced a similar type of reaction, though of smaller amplitude. Like in MEFs, the response was transient and could be enhanced by addition of rolipram (Figure 2B). From adult mice, we isolated cardiac myocytes and peritoneal macrophages. In line with our previous observations [26], β-adrenergic stimulation of cardiac cells led to an increase of cAMP levels, which was further enhanced by rolipram (Figure 2C). The cAMP response to isoproterenol in macrophages was more sustained and only minimally affected by rolipram (Figure 2D). Finally, the β-adrenergic antagonist propranolol completely inhibited the effect of isoproterenol on cAMP production in cardiomyocytes (Figure 2E) and macrophages (Figure 2F), thus showing that the observed FRET variations were specific.
Then, we utilized CAG-Epac1-camps mice to monitor GPCR-cAMP signaling in an intact physiological system. We chose thyroid cells for several reasons. First, these cells are strictly dependent for all their functions, e.g., thyroid hormone production and growth, on the activation of a GPCR, the TSH receptor, which is expressed on their basolateral membrane and the effects of which are largely mediated by cAMP [27],[28]. Second, thyroid cells form supracellular structures, known as thyroid follicles, which constitute both the anatomical and the functional units of thyroid tissue. Importantly, it is possible to isolate and culture thyroid follicles so to maintain their original 3-D structure and polarization, which is reflected by a high spatial organization of the TSH receptor-cAMP signaling cascade [27],[29],[30] (Figure 3A). Thus, they represent a unique model to study the spatiotemporal dynamics of GPCR-cAMP signaling. For this purpose, we established a method that allowed us to isolate, maintain, visualize, and manipulate mouse thyroid follicles under the microscope. Among various protocols, the best results were obtained by enzymatic dissociation followed by deposition on a thin layer of collagen gel (Figure 3B). This protocol preserved good 3-D morphology for at least 24–48 h of culture, without hampering microscopic observation or manipulation (Figure 3B and Video S1).
Next, we utilized thyroid follicles isolated from CAG-Epac1-camps mice to monitor in real time the cAMP response to TSH stimulation. Saturating concentrations of TSH resulted in a fast and robust decrease of the YFP/CFP ratio (Figure 4A and 4B and Video S2), indicative of an increase of cAMP levels [14]. Nevertheless, the Epac1-camps sensor was not saturated, as shown by the further decrease of FRET values obtained by fully activating adenylyl cyclase with forskolin. In contrast to what was observed in other cell types, like for example, embryonic fibroblasts and cortical neurons (for comparison see Figure 2A and 2B), no reduction of cAMP levels was seen after prolonged exposure to various concentrations of TSH (up to 30 min), consistent with limited or no PDE activation under our experimental conditions (Figure 4C). This type of sustained cAMP response further drove our attention to TSH receptor signaling in thyroid cells.
To rule out the possibility that the presence of the Epac1-camps sensor might interfere with cAMP signaling, we compared the cAMP-response to TSH stimulation in thyroid cells isolated from wild-type and CAG-Epac1-camps mice. The intracellular levels of cAMP, measured by an immunoenzymatic assay, were indistinguishable between wild-type and transgenic cells (Figure S1).
Importantly, the lack of appreciable desensitization of the TSH receptor-cAMP signal allowed us to evaluate the kinetics of the return of cAMP to baseline after transient stimulation with TSH. To ensure controlled and fast stimulation, thyroid follicles were kept under laminar-flow perfusion with an apparatus that allowed the rapid exchange between different extracellular solutions. First, we applied very short (10 s) and repeated stimuli with TSH (30 U/l), which were followed by a complete return of cAMP to basal values (Figure 5A). Then, we stimulated thyroid follicles with saturating (30 U/l) or supra-saturating (300 U/l) concentrations of TSH for 30 s (Figure 5B), 2 min (Figure 5C), or 10 min (Figure 5D). Note that 300 U/l were required to reach maximum activation within 30 s, whereas 30 U/l were sufficient for 2-min and 10-min applications. Stimulation with these fully activating concentrations of TSH produced comparable changes of cAMP levels, independently of the duration of the application (Figure 5E). Unexpectedly, stimuli of duration equal or longer than 30 s were associated with an incomplete return of cAMP to basal levels even after extensive washout (Figure 5B–5D). Interestingly, the extent of signal irreversibility increased with the duration of the TSH application, but was not dependent on the cumulative dose of TSH (for example, compare the effect of 300 U/l for 30 s to that of 30 U/l for 2 min), occurred already after 30 s, and was nearly maximal after 10-min stimulation (Figure 5F and 5G). In contrast to TSH receptor activation, stimuli of comparable intensity and duration with a forskolin analog, which directly activates adenylyl cyclase, yielded completely reversible cAMP signals (Figure S2).
In theory, the incomplete restoration of cAMP levels observed after prolonged TSH stimulation might have been due to an inactivation of PDEs. To investigate this possibility, we added a nonselective PDE inhibitor (IBMX) at the end of the washout phase. In contrast to control follicles that were not previously stimulated with TSH, on which IBMX had only a marginal effect, IBMX treatment caused a robust increase in cAMP levels when applied after TSH stimulation for 10 min and subsequent washout (Figure 6). These results demonstrated that, after extensive washout from a prolonged TSH stimulus, PDEs were still highly functional. Thus, it is unlikely that an inactivation of PDEs was the cause of the incomplete recovery of cAMP levels. On the contrary, the fast and robust increase of cAMP levels after IBMX addition suggested that the receptor-adenylyl cyclase system was indeed continuing to signal.
The extent and kinetics of endogenous TSH receptor internalization in primary thyroid cells have been difficult to evaluate due to its very low expression levels. However, the available information suggests that the TSH receptor is internalized and recycled back to the plasma membrane, without being targeted to lysosomes [31],[32]. To monitor the internalization of the endogenous TSH receptor, we utilized the well-established method of following the endocytosis of a fluorescent ligand [3]. For this purpose, we labeled bovine TSH with a fluorescent dye (Alexafluor594), which resulted in no major alterations of its biological activity (Figure S3A); in addition, TSH-Alexa594 bound specifically to HEK293 cells transfected with TSH receptor cDNA (Figure S3B).
Initially, to simultaneously monitor TSH and TSH receptor internalization, we cotransfected HEK293 cells with a YFP-tagged TSH receptor construct and β-arrestin 2, which in this context is required for efficient TSH receptor internalization [33], and stimulated them with TSH-Alexa594 for different periods of time. Both ligand and receptor were found to cointernalize rapidly, reaching a maximum after approximately 20 min (unpublished data). Interestingly, 40 min after stimulation, the ligand and the receptor still appeared to be present in the same intracellular compartments, suggesting that they were not sorted apart during this period of time (Figure 7).
Then, we utilized the fluorescent ligand to follow TSH receptor internalization in primary thyroid cells. When we attempted to visualize whole thyroid follicles loaded with TSH-Alexa594, we found them to have a relatively high autofluorescence in the whole visible spectrum. Thus, visualization of TSH-Alexa594 was largely hampered, and possible only to a limited extent through spectral unmixing. The resulting images, though of poor quality, were suggestive of TSH-Alexa594 being efficiently internalized in thyroid follicles (Figure S4). To better evaluate this phenomenon, we analyzed the binding and internalization of TSH-Alexa594 in single primary mouse thyroid cells, which showed much lower autofluorescence. Interestingly, we found that TSH internalized quickly, with TSH-Alexa594–positive vesicles being detectable 5 min after TSH stimulation and maximal internalization reached approximately after 20 min (Figure 8). Importantly, at this point, a major fraction of labeled TSH was found inside the cells, in vesicles prevalently concentrated around the nucleus. To further investigate the structure and the dynamics of TSH-containing vesicles, we performed a series of time-lapse experiments on cells stimulated with TSH-Alexa594 that were visualized with a total internal reflection fluorescence (TIRF) microscope set to have a high penetration depth. This approach allows us to visualize cytoplasmic structures close to the plasma membrane with a high signal/background ratio [34] and appeared particularly suited for thyroid cells that assume a very thin shape in culture, having a maximal thickness of approximately 1–2 µm only (unpublished data). Interestingly, we found that TSH-Alexa594 was contained in tubulovesicular organelles, forming a highly dynamic and interconnected network (Figure 9 and Video S3), the structure of which could be appreciated only partially by confocal microscopy.
So far, our results suggested that the TSH receptor internalized quickly without evidence of cAMP desensitization and that, after prolonged stimulation, a significant proportion of the TSH receptor-cAMP signal became irreversible. Of note, both phenomena, i.e., receptor internalization and signal irreversibility, had roughly similar kinetics. We therefore hypothesized that, over time, a fraction of the receptors was brought to a state or a compartment where they could no longer be freed of their ligands, but were still able to signal. Trimeric G-proteins are frequently found at non-plasma membrane compartments, such as endosomes, the endoplasmic reticulum, and the Golgi, where they are thought to play roles in vesicle trafficking [8],[35],[36]. Similarly, adenylyl cyclase activity and immunoreactivity have been found on intracellular membranes, though such studies were frequently limited by the very low expression of adenylyl cyclases and the relatively low affinity of the available antibodies [37],[38]. Therefore, it appeared possible that the other components required for TSH receptor signaling might also be present inside the cell. Based on these considerations, we attempted to localize Gαs as well as adenylyl cyclase III, V, and VI, the major isoforms expressed in thyroid cells [39], by immunofluorescence. Our results indicated that Gαs was present on intracellular vesicles and tubulovesicular structures that were typically concentrated around the nucleus (Figures S5, 10, and 11A and 11B). Adenylyl cyclase staining was of much lower intensity. Nonetheless, adenylyl cyclase immunoreactivity was found occasionally on the plasma membrane and largely on intracellular vesicles (Figures S5 and 11C). TSH stimulation did not cause apparent modifications of Gαs or adenylyl cyclase subcellular localization (unpublished data). The specificity of Gαs, adenylyl cyclase III, and adenylyl cyclase V/VI staining was checked by competition with the peptides used to raise the primary antibodies (Figure S6). In addition, the specificity of Gαs immunofluorescence was evaluated by comparing its subcellular pattern to the localization of YFP-tagged Gαs transfected in HEK293 cells (unpublished data).
To identify the structures that were positive for Gαs, we performed a series of colocalization experiments with markers of intracellular compartments. First, we used a fluorescent transferrin conjugate to label early endosomes and the recycling endosomal compartment. This is a frequently used method that takes advantage of the endocytosis and recycling of transferrin together with its receptor [40]. To this end, primary mouse thyroid cells were stimulated with fluorescent transferrin for various periods of time (2–60 min), followed by fixation and immunofluorescent staining of Gαs. Early time points (2–5 min) were used to visualize early endosomes, whereas later time points were employed to identify recycling endosomes. Gαs and transferrin were not found simultaneously on early endosomes (unpublished data). However, at later time points (20–60 min) transferrin appeared to be contained in vesicles, presumably a recycling compartment, associated with the perinuclear tubulovesicular structure positive for Gαs (Figure 10A). Of note, some of these vesicles were also positive for Gαs. In addition, primary mouse thyroid cells were simultaneously treated with fluorescent transferrin and TSH to evaluate whether they followed similar or distinct endocytic pathways (Figure 10B and unpublished data). A partial colocalization between internalized TSH and transferrin was observed at all time points considered (2–60 min)—notice that in Figure 10B, several TSH-positive vesicles contain also transferrin—suggesting that they followed to some extent the same trafficking pathway. An antibody against Rab7 was used to label late endosomes. No colocalization was observed between Gαs and Rab7 (Figure 10C). Finally, an antibody against a Golgi-resident protein, Golgi 58K, was used to mark the Golgi compartment [41]. A high degree of colocalization was present between Gαs and Golgi 58K (Figure 10D), suggesting that a relevant fraction of Gαs was located on membranes of the Golgi complex.
Next, we treated primary thyroid cells with TSH-Alexa594, as a marker of TSH receptor, and looked at its colocalization with Gαs and adenylyl cyclases. Interestingly, internalized TSH and Gαs were frequently found in close association, with TSH being present on vesicles adjacent to the Gαs-positive tubulovesicular structure (Figure 11A and 11B and Video S4). As observed in the case of transferrin, some of these vesicles were also positive for Gαs. TSH and adenylyl cyclase III were found to colocalize occasionally on the plasma membrane and more frequently on intracellular vesicles or tubulovesicular structures (Figure 11C and Video S5). Little colocalization was observed between internalized TSH and adenylyl cyclase V/VI (unpublished data).
To further investigate the subcellular localization of adenylyl cyclases, we labeled them with BODIPY-forskolin, a fluorescent forskolin analog [42]. The specificity of BODIPY-forskolin staining was initially evaluated in HEK293 cells transfected with adenylyl cyclase VI cDNA, which showed a stronger labeling compared to control mock-transfected cells (Figure 12A). Then, we utilized BODIPY-forskolin to stain primary thyroid cells that were visualized with a TIRF microscope as described above. BODIPY-forskolin staining was largely present on intracellular vesicles and tubulovesicular structures (Figure 12B). Utilizing this approach, we could also visualize simultaneously the internalized TSH-Alexa594 and BODIPY-forskolin in cells that were stimulated with the fluorescent ligand. A consistent colocalization was observed at the level of several intracellular vesicles and tubulovesicular structures, in agreement with the previous immunofluorescence data (Figure 12C). The dynamic nature of these intracellular structures can be appreciated in Video S6. Finally, we utilized the fluorescent forskolin analog to perform triple stainings in which we attempted to simultaneously visualize the internalized TSH-Alexa594, Gαs (by immunofluorescence), and adenylyl cyclases. The resulting images were indicative of the three components being simultaneously present on vesicles that were often adjacent to the perinuclear tubulovesicular structure positive for Gαs (Figure 12D). Of note, the latter structure was also labeled by BODIPY-forskolin (Figure 12D).
Finally, we utilized immunogold electron microscopy to evaluate the subcellular localization of the internalized TSH, Gαs, and adenylyl cyclase III at the ultrastructural level. None of the available antibodies against Gαs worked in immunogold stainings. To visualize the internalized TSH, we labeled it with Alexafluor488, which was recognized by a specific antibody. Similarly to TSH-Alexa594, TSH-Alexa488 had a conserved biological activity (unpublished data). Consistent with the previous immunofluorescence results, both TSH-Alexa488 and adenylyl cyclase III were found to be present in endosomes (Figure S7).
Although our data were so far compatible with TSH receptor signaling from intracellular sites, direct functional evidence was still missing. We therefore reasoned that if our hypothesis was correct, treatments capable of inhibiting TSH receptor endocytosis should hamper TSH receptor sequestration and, as a result, increase the reversibility of cAMP signals after TSH washout. We first tried several treatments that have been reported to inhibit GPCR internalization, such as concanavalin A, phenylarsine oxide and hypertonic sucrose [7]. Concanavalin A and phenylarsine oxide were found to partially inhibit forskolin-dependent cAMP production and, therefore, were not further evaluated (unpublished data). By contrast, pretreatment with 0.43 M sucrose for 10 min did not hamper TSH-dependent cAMP accumulation (Figure 13A), whereas it almost completely inhibited TSH-Alexa594 internalization (Figure 13B, Video S7 and Video S8). Importantly, pretreatment of thyroid follicles with hypertonic sucrose was associated with a complete reversibility of the cAMP signal produced by TSH stimulation for 2 min (Figure 13C). The mechanism of action of hypertonic sucrose is not completely understood, though this treatment is known to alter the polymerization of clathrin, thus hampering the formation of coated pits at a very initial phase [7]. Another frequently used method to inhibit GPCR internalization is to interfere with dynamin function. Dynamin is in fact required for later phases of clathrin-dependent endocytosis, such as pinching and release of completed coated pits from the plasma membrane. Recently, Macia et al. developed a membrane permeable dynamin inhibitor, dynasore [43], which could be applied directly to intact thyroid follicles. We therefore repeated the same experiments in the presence of dynasore, which had a consistent inhibitory effect on TSH internalization (Figure 13D and Video S9). Analogously to what was observed with hypertonic sucrose, pretreatment for 20 min with 80 µM dynasore, a concentration that produced a maximal effect on TSH internalization, resulted in a major increase of cAMP signal reversibility after TSH washout (Figure 13E). Taken together, these results strongly suggested that TSH receptor internalization was responsible for the observed irreversibility of cAMP signaling.
A possible concern when considering the occurrence of TSH receptor signaling on endocytic membranes is the effect of acidic pH on TSH–TSH receptor interactions. In fact, the low pH of early (pH 5.9–6.2) and late endosomes (pH.5.0–6.0) [44] is known to promote the dissociation of several receptor–ligand complexes. To evaluate the effect of acidic pH on TSH receptor signaling, we incubated primary thyroid follicles at different pH levels and evaluated the cAMP response to TSH stimulation by FRET microscopy. In agreement with previous observations [45], our results suggested that efficient TSH binding and activation of the TSH receptor was still possible at pH 5.0 (Figure S8).
To further verify the hypothesis that the TSH receptor may continue to signal to Gαs/adenylyl cyclase after internalization, we conducted a series of experiments based on cell fractionation. Since in initial tests we found that these studies required a higher amount of starting material than could be obtained from primary thyroid cells, these experiments were performed on the FRTL5 thyroid cell line. FRTL5 cells are a widely used model of well-differentiated thyroid cells, which, among other features of normal thyrocytes, express thyroid-specific genes, conserve a functional TSH receptor-cAMP signaling pathway and retain the capability to synthesize thyroid hormones [28],[46].
First, we established a method to separate the intracellular content from the plasma membrane. In order to allow a better characterization of the supposed intracellular TSH receptor signaling compartment, we looked for a method that was fast, as gentle as possible, and produced with high yield an intracellular fraction with the lowest possible contamination from the plasma membrane. The best results were obtained utilizing a protocol based on the separation of the plasma membrane with magnetic beads coated with concanavalin A, which binds selectively to the glycoproteins present on the cell surface [47]. Figure 14A shows the results of a typical fractionation experiment. Please notice the virtual absence of the Na+/K+ ATPase, used as a plasma membrane marker, in the intracellular fraction after the second round of purification (lane 5).
Next, we evaluated by Western blot analysis the presence of Gαs and adenlyl cyclase III in the different subcellular fractions (Figure 14B). As expected, bands corresponding to Gαs and adenylyl cyclase III were present in the total homogenate as well as in the plasma membrane fraction. In addition, the same bands were also present in the intracellular fraction, thus confirming the evidence based on the results of the immunofluorescence experiments that both Gαs and adenylyl cyclase III were present in significant amounts also in the intracellular compartment.
Once the fractionation method was established, we attempted to directly measure the activity of adenylyl cyclase in the plasma membrane and in the intracellular fraction of cells that were previously stimulated with TSH. The prediction was that if the TSH receptor continued to signal to Gαs/adenylyl cyclase after internalization, this should result in an increase of the adenylyl cyclase activity in the intracellular fraction. This appeared to be indeed the case, as we found that pretreatment of FRTL5 cells for 30 min with TSH was associated with an increased adenylyl cyclase activity in the intracellular fraction (Figure 14C). By contrast, pretreatment with TSH was not associated with any modifications of the adenylyl cyclase activity in the plasma membrane fraction. This is not surprising, as TSH is expected to dissociate from the TSH receptors present on the plasma membrane during the time required for the cell fractionation. Nevertheless, a certain amount of functional receptors was left on the plasma membrane, as adenylyl cyclase activity in this fraction could be increased by adding TSH in the reaction mixture (Figure 14C).
Altogether, these results further supported the hypothesis that the TSH receptor continued to signal to Gαs/adenylyl cyclase after internalization.
To better understand the possible consequences of intracellular GPCR signaling, we generated a mathematical model of the GPCR-cAMP pathway. Our model is based on the recent work done by Neves et al. on β2-adrenergic receptor signaling [21]. In this model, the receptor, G-proteins and adenylyl cyclase are placed on the plasma membrane, whereas ATP, cAMP, PDE4, and PKA are cytosolic (Figure 15A). In addition, to mimic an intracellular signaling compartment (ICSC), we placed G-proteins and adenylyl cyclase also on an intracellular membrane and simulated the internalization of both GPCR and ligand to this compartment (Figure 15B). Simulations were performed with the Virtual Cell software [48]–[50]. Initial non-spatially resolved simulations, aimed at validating the model and assessing the effects of different geometries and parameters as well as of receptor recycling, are described in Text S1 and Figures S9, S10, S11. Subsequently, a system of partial differential equations was used to obtain a spatiotemporal description of the GPCR-cAMP pathway. The results of the simulations performed with this model were in good agreement with our experimental data. In Figure 15C are reported the results of a spatial simulation in which a cell was given a transient stimulus, by mimicking a perfusion system that provided and subsequently removed the ligand from the extracellular space. Notice that in the absence of the ICSC, the cAMP response is fully reversible. On the contrary, in the presence of the ICSC, the signal is no longer reversible, in agreement with the results of the FRET experiments. Interestingly, the different types of cAMP gradients generated in the presence or absence of the ICSC are also reflected by different degrees and spatial patterns of PKA activation. Thus, our mathematical model predicted that GPCR signaling to cAMP from an ICSC should have important consequences also on the spatiotemporal dynamics of downstream signaling events.
Based on the results of the mathematical simulations, we investigated whether the inhibition of TSH receptor internalization had some effects on the signaling events downstream of cAMP production. One of the earliest effects of TSH in thyroid cells is the reorganization of the actin cytoskeleton. This is a well-known phenomenon, mainly consisting in the depolymerization of stress fibers, that is implicated in the reuptake of thyroglobulin and in the induction of thyroid-specific genes [51]–[53]. Therefore, we evaluated whether blocking receptor internalization might affect the depolymerization of actin in response to TSH (Figure 16A and 16B). Primary mouse thyroid cells were stimulated with TSH in the presence or absence of endocytosis inhibitors, as described above, then fixed and stained for actin with fluorescent phalloidin. Hypertonic sucrose itself was found to induce a modification of actin architecture, and was not further evaluated (unpublished data). Instead, dynasore treatment did not cause any appreciable modifications of cellular shape or actin polymerization (Figure 16A). As expected, TSH alone caused a generalized depolymerization of actin, which was observed both in the central cellular compartment (Figure 16A) and in lamellipodia (Figure 16B). By contrast, in the presence of dynasore, TSH induced only a partial depolymerization in the central cellular compartment (Figure 16A) and failed to induce depolymerization in lamellipodia, which showed an even thicker actin mesh compared to that of control cells (Figure 16B).
The PKA-substrate vasodilator-stimulated phosphoprotein (VASP) is a key effector of cAMP in the reorganization of actin cytoskeleton [54],[55]. VASP is concentrated at actin hot spots (lamellipodia, filopodia, cell–substrate, and cell–cell contacts) where it regulates the polymerization and branching of actin filaments. Importantly, its function is regulated by PKA via phosphorylation at Ser 157 [54],[55]. We therefore investigated whether VASP was phosphorylated in response to TSH stimulation in thyroid cells and the possible consequences of TSH receptor internalization on this pathway. VASP phosphorylation was evaluated by Western blot analysis, both by monitoring the intensity of the slower migrating band detected by a VASP antibody and by utilizing an antibody specific for VASP phosphorylated at Ser 157. In the absence of endocytosis inhibitors, TSH caused a 2-fold induction of VASP phosphorylation. However, pretreatment with endocytosis inhibitors caused a substantial reduction of VASP phosphorylation in response to TSH (Figures 16C and S12).
To confirm the possible involvement of VASP in the control of actin polymerization in thyroid cells, we examined its subcellular localization by immunofluorescence. As observed in other cells in which VASP regulates actin polymerization, VASP was concentrated at the ends of actin filaments (Figure 16D).
Finally, we evaluated whether blocking TSH receptor internalization also altered the spatial pattern of VASP phosphorylation by PKA. To this end, we stimulated thyroid cells with TSH alone or in the presence of dynasore, and visualized VASP phosphorylated at Ser 157 by immunofluorescence (Figure 16E). In the absence of dynasore, TSH caused a robust increase of VASP phosphorylation throughout the cell. In particular, phosphorylated VASP colocalized with spots of depolymerized actin in the central cellular compartment. By contrast, only a minor induction of VASP phosphorylation and no spots of phosphorylated VASP in the central cellular compartment were observed in the presence of dynasore (Figure 16E). As a control, the generalized depolymerization of actin and the pattern of VASP phosphorylation induced by forskolin were not modified by dynasore (Figure 16F).
Taken together, these results suggest that TSH receptor internalization not only modifies the temporal dynamics of cAMP signaling by leading to persistent cAMP production, but may also affect the intensity and the spatial patterning of downstream signals.
The conventional model of GPCR signaling is based on the central concept that signaling to second messengers such as cAMP is taking place only at the cell plasma membrane [1]. The role assigned to receptor internalization is essentially to reduce the number of GPCRs present on the cell surface, thus contributing to signal desensitization, or to bring the receptors to an intracellular site for dephosphorylation and resensitization [1],[2],[56]. Whatever the fate of the internalized GPCRs, they are thought to stop signaling to second messengers once inside the cell. Here, we provide evidence that a GPCR continues to stimulate cAMP production after internalization. cAMP signaling by internalized receptors appears to be different from that occurring at the plasma membrane, as it is of a sustained nature and leads to a different pattern of downstream signals.
Although the biochemical steps involved in GPCR signaling are known in detail, their location in space and time in living cells is poorly understood. This represents a major drawback, as there is emerging evidence that signaling cascades are highly organized in space and time, and intrinsically dynamic [20]. On such a basis, there is an urgent need to develop new tools and techniques to monitor signaling events with submicrometer resolution and fast temporal dynamics. Recently, the application of FRET to color variants of Aequorea victoria GFP has allowed the development of a toolbox of genetically encoded sensors to observe intracellular signaling events in real time. Several sensors have been described that exhibit FRET changes on exposure to cAMP [57]. A sensor based on the dissociation of the PKA subunits was first described by Adams et al. (using rhodamine and fluorescein as fluorophores) [58] and later modified by Zaccolo and colleagues to be genetically encoded by substituting rhodamine and fluorescein with GFP variants [12],[13]. This sensor has led to major insights into the biology of cAMP [13],[25],[59]–[61]. We have recently described another type of sensors, including Epac1-camps, that are based on a single cAMP-binding domain derived from cAMP-binding proteins [14],[26]. These sensors, being based on a cAMP-binding domain alone and not on the entire protein, are devoid of any signaling activity. Therefore, they are extremely well tolerated and are expected not to alter the functions of the cell [57]. This is probably why the CAG-Epac1-camps mice are viable and healthy, whereas attempts to create genetically modified organisms with ubiquitous expression of PKA-based sensors have so far been unsuccessful [62],[63].
The newly generated cAMP reporter mice were used to monitor GPCR signaling in living cells. A major strength of our study was the use of a highly physiological system, i.e., 3-D thyroid follicles. Thyroid follicles represent a unique model to study the spatiotemporal dynamics of cAMP signaling because they maintain the supracellular organization, size, and polarization possessed by thyroid cells in vivo, and constitute a rare example of cells that are under the strict control of a GPCR (the TSH receptor) and of cAMP for virtually all their functions (e.g., thyroid hormone production, cell proliferation) [27],[28]. The conservation of the original cellular architecture of thyroid tissue is reflected by the high degree of spatial organization of the TSH receptor-cAMP signaling cascade: the TSH receptor is expressed on the basolateral membrane; upon binding of TSH, cAMP produced at the basolateral membrane diffuses through the cytosol to activate cytosolic PKA I and PKA II, mainly located in the Golgi complex; PKAs in turn phosphorylate targets localized in different cellular compartments (e.g., cytosol, nucleus, as well as apical and basolateral membranes) [27],[28],[64]–[66]. The maintenance of the original cellular size and shape is of fundamental importance, as modifications of these parameters are expected to alter the properties of signaling cascades, including the shape of the gradients of cAMP and other soluble messengers [20]–[22]. Another advantage of thyroid follicles is that they already express all the required signaling machinery at endogenous levels. Here, we describe a method to directly visualize cAMP signaling in thyroid follicles, thus allowing, for the first time to our knowledge, a precise monitoring of the kinetics of a GPCR-cAMP signaling cascade, at endogenous levels of expression and in its native multicellular functional unit. We believe that this represents a major step towards depicting signaling pathways in their natural context.
By monitoring cAMP signaling and receptor internalization, we obtained a series of unexpected findings. First, the TSH receptor internalizes rapidly and consistently into a pre-Golgi compartment in close association with Gαs and adenylyl cyclase III. Second, the robust internalization of the TSH receptor is not associated with any appreciable desensitization of the cAMP signal. Third, prolonged TSH receptor stimulation leads to a sustained production of cAMP; by contrast, short stimuli are completely reversible. Fourth, blocking receptor internalization can prevent the irreversibility of cAMP signals. Fifth, TSH receptor internalization is required to ensure a normal pattern of actin rearrangement and VASP phosphorylation downstream of cAMP. On the basis of these findings, we suggest that the TSH receptor continues to signal to adenylyl cyclase after internalization, and that the location of TSH receptor-cAMP signaling affects the spatial patterns of the downstream signals.
The consequences of GPCR signaling from inside the cell appear to be multiple, as anticipated by Miaczynska et al. [4]. First, sustained cAMP production from internalized receptors may provide a memory mechanism, allowing thyroid cells to maintain constant thyroid hormone production in the presence of fluctuations in plasma TSH concentration. Indeed, TSH release from the pituitary follows a circadian rhythm, with a zenith in the first hours of the morning, but those fluctuations are not associated with a change in the production of thyroid hormones [67]. Additionally, intracellular membranes may provide specialized platforms for signal compartmentalization. This is the case of MAPK activation by GPCRs, which appears to happen selectively on endosomes [7], and of tyrosine kinase receptors that activate a distinct subset of substrates once inside the cell [3]–[6]. In the case of cAMP signaling, internalized receptors appear to be more efficiently coupled to PKA, as suggested by the effect of endocytosis inhibitors on VASP phosphorylation. This could be explained if the receptors need to be brought close to an intracellular pool of PKA for efficient kinase activation and/or VASP phosphorylation.
The existence of spatial microdomains of cAMP signaling inside the cell has been debated for years. Unlike Ca++, whose apparent diffusion is limited by the high buffering capacity of cytosolic proteins [68],[69], measurement of cAMP diffusion gave values in the range of 270–780 µm2/s, i.e., as fast as would be expected in a simple electrolyte solution [70],[71]. For this reason, cAMP has been traditionally considered a far-reaching messenger, capable of crossing the whole cell to convey the information generated at the plasma membrane. Although this might be the case in certain circumstances, as for example, in the case of long-term facilitation in Aplysia sensory neurons [72], it is difficult to reconcile the free diffusion of cAMP with the high spatial organization of the cAMP signaling cascade (adenylyl cyclases, PKAs, AKAPs, PDEs, etc.) [11],[57],[59],[72]–[78]. This paradox has been partially resolved by recent reports of restricted cAMP diffusion. A striated pattern of cAMP signaling has been observed in cardiomyocytes with a genetically encoded PKA-based sensor [13]. In addition, a study by our group has revealed that in these cells, β1-adrenergic receptors produce generalized cAMP responses, whereas β2-adrenergic receptors generate locally confined signals [26]. Furthermore, the existence of cAMP microdomains is supported by a body of indirect evidence [15],[59],[75],[79]–[86], and the formation of cAMP gradients is predicted on the basis of the spatial segregation of adenylyl cyclases on the membrane and PDEs in the cytosol [20],[22],[87]. Recent calculations suggest that such gradients may have a length of 2.5–4 µm [22]. If this is indeed the case, signals originating near the plasma membrane could hardly reach deep inside the cell. In this perspective, GPCR-cAMP signaling from intracellular sites might provide a new basis to explain the activation of distant targets and the specific effects observed with different types of activation.
In an attempt to identify the intracellular compartment(s) where sustained TSH receptor-cAMP signaling was occurring, we analyzed the subcellular localization of the internalized TSH, Gαs, and adenylyl cyclases by a combination of experimental approaches. In agreement with previous observations [31], our results indicate that TSH and its receptor are internalized rapidly in endosomes. Later on, at least part of the internalized TSH is found together with transferrin in perinuclear vesicles that probably represent a recycling endosomal compartment. Interestingly, some of these vesicles appear to contain also Gαs and adenylyl cyclases. Although we cannot exclude the possibility that subdetectable levels of Gαs may be present together with adenylyl cyclases in early endosomes and that TSH receptor-cAMP signaling might be taking place also in this compartment, those perinuclear vesicles that simultaneously contain TSH, Gαs, and adenylyl cyclases are the most likely candidates to be the intracellular sites of TSH receptor-cAMP signaling. An interesting aspect that emerges from our experiments and that should be taken into consideration is the highly dynamic nature of the intracellular structures where the internalized TSH, adenylyl cyclases, and Gαs are present. These compartments appear in fact highly complex, sometimes having specialized subdomains that prevalently contain one component and rapidly exchange their contents by fusing to each other or through budding of new vesicles. Additional studies, probably requiring the generation of better antibodies and new tools to simultaneously monitor the localization of the receptor, Gαs and adenylyl cyclases in real time, are needed to further investigate the microscopic anatomy and the dynamic nature of the intracellular cAMP signaling compartment(s).
In summary, our data show that GPCR signaling in a physiological setting may be more complex than current knowledge, based mostly on studies with transfected cells, suggests. In particular, the subcellular localization of GPCRs, either on the plasma membrane or intracellular, appears to be an important parameter affecting the duration and the spatial pattern of downstream signals. These findings may lead to reconsidering the current model of GPCR signaling and suggest new and intriguing scenarios on the function of GPCRs inside the cell.
All animal work was done according to the regulations and with the permission of the government of Lower Franconia.
Cell culture media and reagents were from Pan Biotech. Glass-bottom Petri dishes were from World Precision Instruments. Collagen and dispase were from Roche Diagnostics. Bovine TSH (bTSH), dynasore, the mouse monoclonal antibody against Golgi 58K protein, the cAMP enzyme immunoassay kit (CA200), and alumina WN-6 columns were from Sigma-Aldrich; 7-deacetyl-7-[O-(N-methylpiperazino)-γ-butyryl)]-forskolin (DMPB-forskolin) was from Merck. Rabbit polyclonal antibodies against Gαs, adenylyl cyclase III, and adenylyl cyclase V/VI were from Santa Cruz Biotechnology. Antibodies against Rab7 (mouse monoclonal), Na+/K+ ATPase (mouse monoclonal), and EEA1 (rabbit polyclonal) were from Abcam. Antibodies against VASP (rabbit monoclonal) and P-VASP (Ser 157, rabbit polyclonal) were from Cell Signaling Technology. Cy2-conjugated anti-rabbit and anti-mouse polyclonal antibodies were from Jackson ImmunoResearch. Collagenase I and II, Alexafluor488 and Alexafluor594 succinimidyl esters, the Alexafluor594-conjugated goat anti-rabbit polyclonal antibody, Alexafluor488-conjugated transferrin, Alexafluor488-conjugated phalloidin, the rabbit anti-Alexafluor488 antibody, and Dynabeads Biotin Binder magnetic beads were from Invitrogen. The goat anti-rabbit and anti-mouse antibodies conjugated with horseradish peroxidase were from Amersham Pharmacia Biotech and Millipore. The ECL detection kit was from Amersham Pharmacia Biotech. Effectene transfection reagent was from Qiagen. Biotinylated concanavalin A was from Vector Laboratories. [α-32P]ATP was from PerkinElmer Life Sciences. All other reagents were from Sigma-Aldrich.
To generate transgenic cAMP-sensor mice, we followed the strategy used for GFP mice [23]. FVB one-cell embryos were injected by standard procedures with a genetic construct in which the sensor sequence was inserted between the CAG promoter and the rabbit β-globin polyadenylation signal (See Figure 1A). The original pCAGGS expression vector has been described and provided by J. Miyazaki [88]. The screening of pups for transgene insertion was performed by PCR analysis as previously described [26]. Three transgenic lines were obtained that showed different levels of fluorescence. The line with the highest expression was used for further experiments. Mice and freshly isolated organs were imaged with a Leica macroFluo (Z6APO-A) microscope, using the YFP emission filter.
Mouse thyroid follicles were isolated according to a previously published protocol [30], with minor modifications. Thyroid lobes were dissected from 1–2 mo-old mice. The lobes were collected in a 1.5-ml microcentrifuge tube, containing 1 ml of digestion medium, which consisted of 100 U/ml collagenase I, 100 U/ml collagenase II, and 1 U/ml dispase, dissolved in Dulbecco's modified Eagle's medium (DMEM)/F-12. Enzymatic digestion was carried out for 1 h in a 37°C water bath, with manual shaking every 15 min. After digestion, isolated individual follicles were washed three times with culture medium and plated on glass-bottom 35-mm Petri dishes, coated with a thin layer of collagen gel. The collagen gel was prepared by spreading 8 µl of a collagen solution (3 mg/ml in 0.2% acetic acid) onto the glass surface, followed by addition of a neutralizing solution (0.4 M NaHCO3, 0.2 M HEPES [pH 7.4]). For the culture of individual thyroid cells, follicles isolated from two to three mice were seeded in a 100-mm culture dish and grown to confluence in a monolayer for 3–4 d. They were then completely dissociated to single cells with trypsin (0.25%)-EDTA (0.02%) and plated on 24-mm glass coverslips. Follicles and isolated thyroid cells were maintained in DMEM/F-12+20% FCS (37°C, 5% CO2).
FRTL5 cells were cultured in Coon's modified Ham's F12 medium supplemented with 5% FCS, 1% penicillin, 1% streptomycin, and a mixture of five hormones, and bTSH (6H) as previously described [89]. Twenty-four hours before the cell fractionation and adenylyl cyclase assay experiments, FRTL5 cells were switched to complete medium without bTSH (5H).
Primary mouse embryonic fibroblasts (MEFs) and cortical neurons were isolated from embryonic day (E) 14.5 transgenic embryos as previously described [14],[90]. Cortices were dissociated with trypsin and cells were plated onto poly-d-lysine-coated glass coverslips in the serum-free Neurobasal-A medium containing B-27 supplement, 0.5 mM l-glutamine, antibiotics, and 25 µM glutamate. Twenty-four hours later, the medium was changed to glutamate-free. Experiments were performed 2 d after plating. MEF isolation was performed using standard trypsin digestion of the embryos with subsequent plating on glass coverslips in DMEM medium containing 10% FCS, 2 mM l-glutamine, and antibiotics. Imaging experiments were performed 24 h after the isolation. Adult cardiac myocytes were isolated and measured 2–5 h after isolation as described [26]. Peritoneal macrophages were isolated as described [14] and maintained for 24 h in DMEM medium containing 10% FCS, 2 mM l-glutamine, and antibiotics. Experiments were performed 24–48 h after the isolation. Cos-7 and HEK293 cells were cultured in DMEM+10% FCS.
TSH labeling was performed with Alexafluor488 or Alexafluor594 succinimidyl esters, following the manufacturer's protocol. Briefly, 5 mg of bTSH were dissolved in 0.5 ml of 0.1 M NaHCO3 buffer (pH 8.3). Then, 0.5 mg of either reactive dye dissolved in DMSO was added to the tube while vortexing, and the reaction was incubated for 1 h at room temperature with continuous stirring. The protein conjugates were separated from the unreacted dyes by gel filtration on Sephadex G25 columns. The concentration of TSH-Alexafluor488 and TSH-Alexafluor594 in the collected fractions was about 1 mg/ml. The degree of labeling was typically of one to two fluorescent moieties per molecule of TSH. After labeling, the TSH preparations were immediately aliquoted and stored at −20°C.
Cos-7 cells were transfected with human TSH receptor cDNA or control empty vector by the diethylaminoethy-dextran method followed by a dimethylsulfoxide shock. Two days after transfection, the cells were used for cAMP determinations and flow immunocytofluorimetry to evaluate the transfection efficiency. For cAMP determinations, culture medium was replaced with Krebs-Ringer-HEPES buffer (KRH) for 30 min. Thereafter, the cells were incubated for 60 min in fresh KRH supplemented with 25 µM rolipram and various concentrations of labeled or unlabeled bTSH. At the end of the 1-h incubation, the medium was discarded and samples were extracted with 0.1 M HCl. The cell extracts were dried in a vacuum concentrator, resuspended in water, and diluted appropriately for cAMP evaluation by radioimmunoassay, utilizing standard procedures. cAMP levels in primary thyroid cells were measured with a commercial ELISA (CA200; Sigma-Aldrich), following the manufacturer's protocol.
HEK293 cells were plated on 24-mm round glass coverslips and transfected with human TSH receptor cDNA or control empty vector by Effectene, following the manufacturer's protocol. After 48 h, the medium was replaced with a buffer containing 144 mM NaCl, 5.4 mM KCl, 2 mM CaCl2, 1 mM MgCl2, 20 mM HEPES, 1% BSA (pH 7.3) and treated with 3 µg/ml TSH-Alexa594. In addition, the specific binding of TSH-Alexa594 was confirmed in primary mouse thyroid cells by competition with a 100-fold molar excess of unlabeled TSH (unpublished data). TSH-Alexa594 bound to cells was visualized by TIRF microscopy.
For fluorescent microscopy, glass-bottom Petri dishes or glass coverslips mounted in an experimental chamber were placed on a Zeiss Axiovert 200 inverted microscope equipped with an oil-immersion 63× objective, a polychrome IV light source (Till Photonics), a 505 DCXR beam splitter, and a CoolSNAP-HQ CCD-camera (Visitron Systems). FRET was monitored using MetaFluor 5.0 software (Molecular Devices) as the ratio between emission at 535±20 nm (YFP) and emission at 480±15 nm (CFP), upon excitation at 436±10 nm. The imaging data were analyzed utilizing MetaMorph 5.0 (Molecular Devices) and Prism (GraphPad Software) software, by correcting for spillover of CFP into the 535-nm channel and direct YFP-excitation, to give corrected YFP/CFP ratio data. Images were acquired every 5 s, with 5-ms illumination time, which resulted in negligible photobleaching for over 30-min observation. To study agonist-induced changes in FRET, cells and thyroid follicles were continuously superfused with phenol red–free medium containing 1% BSA or the same plus agonists and/or endocytosis inhibitors, with a custom apparatus. All experiments were performed at 37°C. Confocal images were acquired with a Leica SP5 confocal microscope (Leica). To visualize TSH-Alexa594, fixed cells were excited with a 594-nm laser line, and images were acquired with a high-sensitivity APD detector. TIRF images were acquired with a Leica AM TIRF microscope, equipped with 488- and 561-nm laser lines and a fast high-sensitivity EM-CCD camera (Cascade 512B). Confocal and TIRF images were analyzed with ImageJ (U. S. National Institutes of Health, http://rsb.info.nih.gov/ij/, 1997–2007). AVI Videos were compressed with the Cinepak codec (Radius).
Cells were plated on 24-mm glass coverslips, washed with PBS and fixed with 4% paraformaldehyde for 15 min at room temperature. Cells were then permeabilized with PBS+0.1% Triton X-100 for 3 min at room temperature, blocked with PBS+5% goat serum for 1 h at room temperature, incubated with the indicated primary antibodies overnight at 4°C and incubated with the appropriate secondary antibodies for 2 h at room temperature. Antibodies against Gαs, adenylyl cyclase III, adenylyl cyclase V/VI, Golgi 58K, VASP, and P-VASP were used at 1∶50 dilution. The Rab7 antibody was used at 1∶200 dilution. All antibody solutions were prepared in PBS+5% goat serum. Secondary antibodies were used at 1∶400 (Cy2-conjugated goat anti-rabbit and anti-mouse polyclonal antibodies) or 1∶2,000 (Alexafluor594-conjugated goat anti-rabbit polyclonal antibody) dilutions. The specificity of the immunofluorescent stainings was evaluated by omitting the primary antibodies and by competition with the peptides used to raise the primary antibodies (for Gαs, adenylyl cyclase III, and adenylyl cyclase V/VI). In addition, the specificity of Gαs immunofluorescence was evaluated by comparing the localization of transfected Gαs-YFP and Gαs immunostaining in HEK293 cells (unpublished data). BODIPY-forskolin labeling was performed by incubating the cells for 10 min at room temperature with 100 nM BODIPY-forskolin dissolved in PBS+1% BSA. Then, the cells were washed twice with PBS and imaged immediately.
Cells were plated on 24-mm glass coverslips, washed with PBS, and then fixed with 4% paraformaldehyde for 10 min at room temperature. Cells were then permeabilized with PBS+0.1% Triton X-100 for 3 min at room temperature, blocked with PBS+1% BSA for 1 h at room temperature, and incubated with 1 U of Alexafluor488-conjugated phalloidin for 20 min at room temperature.
Primary mouse thyroid cells were fixed with 2% paraformaldehyde and 0.2% glutaraldehyde in PBS for 1 h, embedded in 12% gelatin, and infiltrated with 2.3 M sucrose. Ultrathin cryosections were obtained with a Reichert-Jung Ultracut E with a FC4E cryoattachment and collected on copper-formvar-carbon-coated grids. Immunogold labeling on ultrathin cryosections was performed as previously described [91]. Briefly, the sections were incubated with the rabbit anti-adenylyl cyclase III antibody or the rabbit anti-Alexafluor488 antibody, followed by incubation with 15-nm protein A-gold. Control sections were incubated with an unrelated antibody or without primary antibodies. Very low levels of labeling were detected in all control sections (unpublished data). All samples were examined with a Philips CM10 or a Fei TECNAI G12 electron microscope.
Streptavidin magnetic beads (Dynabeads Biotin Binder) were washed using a modification of the manufacturer's protocol. Briefly, 600 µl of magnetic beads were placed in a 1.5-ml Eppendorf test tube and were separated from the solution by placing the vial in contact with a magnet. The supernatant was discarded by pipetting, and the beads were washed with 1 ml of TBS buffer (150 mM NaCl, 50 mM Tris-HCl [pH 7.4]) supplemented with 2 mM EDTA and 1 mg/ml BSA, and resuspended in 600 µl of the same buffer. Then, 60 µl of biotinylated concanavalin A (5 mg/ml) were added, and the suspension was mixed in a rotor for 60 min at room temperature to allow binding of biotinylated concanavalin A to streptavidin. Finally, the beads were washed thrice with TBS buffer supplemented with 2 mM EDTA and 1 mg/ml BSA, and resuspended in 600 µl of homogenization buffer (HB: 250 mM sucrose, 25 mM KCl, 2.5 mM Mg(OAc)2, 25 mM Hepes/KOH [pH 7.4]) and stored at 4°C until use.
The protocol for plasma membrane separation with concanavalin A immobilized on magnetic beads was based on the method described by Lee et al. [47]. FRTL5 cells (20 100-mm Petri dishes per condition) were stimulated with 30 U/l bTSH for 30 min at 37°C where indicated, and harvested by trypsinization. The subsequent steps were performed at 4°C. First, FRTL5 cells were washed once with 5H medium and twice with TBS, and resuspended in HB. Then, the equivalent of 100 µl of concanavalin A beads for each Petri dish was added to the cell suspension, and the samples were incubated for 30 min at 37°C under continuous rotation. At the end of the first incubation, the cells were lysed by gently passing them eight times through a 1-ml syringe with a 26G needle, and the plasma membrane fraction bound to the beads was separated with the help of a magnet. Thereafter, the nuclei were sedimented by centrifuging the samples at 800×g for 10 min at 4°C. The remaining postnuclear supernatant was further purified by adding 100 µl of concanavalin A beads for each Petri dish and repeating the incubation and the magnetic separation procedure to give rise to a second bead-bound fraction and to the final intracellular fraction. The beads with the attached fractions were washed twice with HB supplemented with 1 mg/ml BSA and twice with HB. For Western blot experiments, the proteins bound to the magnetic beads were eluted by resuspending the beads in SDS sample buffer (2% SDS, 10% glycerol, 50 mM dithiothreitol, 0.01% bromophenol blue, 62.5 mM Tris/HCl [pH 6.8]), incubating them at 60°C for 10 min and removing the beads with the help of a magnet. For adenylyl cyclase activity determinations, the beads with the attached membrane fraction were resuspended in HB buffer and used directly in the adenylyl cyclase assay.
The determination of adenylyl cyclase activity was based on the method originally described by Jakobs et al. [92],[93]. Briefly, 300 µg of proteins were added to an incubation medium in a total volume of 100 µl with final concentrations of 50 mM Tris/HCl (pH 7.4), 300 mM sucrose, 100 µM cAMP, 10 µM GTP, 100 µM ATP, 1.25 mM Mg(Ac)2, 100 µM IBMX, 0.2% BSA, 15 mM creatine phosphate, and 0.4 mg/ml creatine kinase. Samples were incubated with about 300,000 cpm of [α-32P]-ATP for 60 min in the incubation medium. The reaction was stopped by addition of 400 µl of a 125 mM ZnAc solution and 500 µl of a 144 mM Na2CO3 solution. Samples were centrifuged for 5 min at 14,000 rpm in a laboratory microcentrifuge. Finally, 800 µl of the resulting supernatant were applied to alumina WN-6 columns that were eluted twice with 2 ml of 100 mM Tris/HCl (pH 7.4). The eluates were counted in a β-counter.
Primary mouse thyroid cells were seeded in six-well plates and starved in serum-free medium for 4 h. Thereafter, the cells were preincubated in serum-free medium plus/minus endocytosis inhibitors and incubated at 37°C in the presence of 1 U/l bTSH for 30 min. This concentration of TSH was chosen because it elicited a robust phosphorylation of VASP, without causing the VASP signal to be completely saturated. At the end of the incubation, the cells were washed with PBS, lysed with SDS sample buffer, and immediately heated for 5 min at 95°C. The levels of VASP phosphorylation were evaluated by Western blot analysis. Afterwards, the membranes were stripped and reprobed with an antibody against total VASP.
Protein concentration was determined by BCA assay. Protein extracts were electrophoresed on a 10% SDS polyacrylamide gel and electro-transferred to a nitrocellulose membrane. Membranes were blocked with TBS-T+3% milk, probed with the indicated primary antibody overnight at 4°C, and incubated with the appropriate horseradish peroxidase-conjugated secondary antibody for 1 h at room temperature. The following dilutions of primary antibodies were used: anti-Na+/K+ ATPase 1∶10,000, anti-EEA1 1∶400, anti-Golgi 58K 1∶5,000, anti-Gαs 1∶10,000, and anti-adenylyl cyclase III 1∶10,000. Detection was performed utilizing the ECL detection kit.
Simulations were performed in the Virtual Cell modeling environment [48]–[50]. The receptor was placed on the plasma membrane. G-proteins and adenylyl cyclase were both on the plasma membrane and on the membrane of the ICSC. ATP, cAMP, PDE4, and PKA were cytosolic. In some instances, we simulated the internalization of the receptor and its ligand to the ICSC. We used the initial concentrations and diffusion coefficients utilized by Neves et al. [21], which are mostly based on experimentally determined values. Initial concentrations, displayed in units of molecules/µm2 for membrane components and µM for cytosolic components, are provided in Table S1. For those components not shown, the initial concentration was set at zero. Reactions and kinetic parameters are shown in Table S2. Spatial simulations were run using the regular grid, finite volume solver. The geometric parameters used in the simulations are provided in Table S3. The mean steady-state concentration of cAMP, obtained by running the model in the absence of ligand until all the components reached steady state, was used as the initial concentration for subsequent simulations. A detailed description of the mathematical model and the results of additional simulations can be found in Text S1. The entire model, parameters and geometries are available at http://vcell.org/.
Values are expressed as mean±standard error of the mean (SEM). Data normality was checked with the Kolmogorov-Smirnov test. Differences between means were assessed by two-tailed t-test (for two groups) or one-way ANOVA followed by Bonferroni post hoc test (for three or more groups). For analysis of signal normalization after TSH washout, signal reversibility was calculated from YFP/CFP ratio values by setting the minimum value equal to zero and the value before TSH stimulation equal to 100%. The values obtained from different replicates were globally fitted to a first-order exponential function. Fits were compared by using F test, having a null hypothesis that Ymax values were the same for all datasets. Alternatively, data from each replicate were individually fit to a first-order exponential function, and the obtained Ymax values were compared by one-way ANOVA.
|
10.1371/journal.pcbi.1002731 | How Recent History Affects Perception: The Normative Approach and Its Heuristic Approximation | There is accumulating evidence that prior knowledge about expectations plays an important role in perception. The Bayesian framework is the standard computational approach to explain how prior knowledge about the distribution of expected stimuli is incorporated with noisy observations in order to improve performance. However, it is unclear what information about the prior distribution is acquired by the perceptual system over short periods of time and how this information is utilized in the process of perceptual decision making. Here we address this question using a simple two-tone discrimination task. We find that the “contraction bias”, in which small magnitudes are overestimated and large magnitudes are underestimated, dominates the pattern of responses of human participants. This contraction bias is consistent with the Bayesian hypothesis in which the true prior information is available to the decision-maker. However, a trial-by-trial analysis of the pattern of responses reveals that the contribution of most recent trials to performance is overweighted compared with the predictions of a standard Bayesian model. Moreover, we study participants' performance in a-typical distributions of stimuli and demonstrate substantial deviations from the ideal Bayesian detector, suggesting that the brain utilizes a heuristic approximation of the Bayesian inference. We propose a biologically plausible model, in which decision in the two-tone discrimination task is based on a comparison between the second tone and an exponentially-decaying average of the first tone and past tones. We show that this model accounts for both the contraction bias and the deviations from the ideal Bayesian detector hypothesis. These findings demonstrate the power of Bayesian-like heuristics in the brain, as well as their limitations in their failure to fully adapt to novel environments.
| In this paper we study how history affects perception using an auditory delayed comparison task, in which human participants repeatedly compare the frequencies of two, temporally-separated pure tones. We demonstrate that the history of the experiment has a substantial effect on participants' performance: when both tones are high relative to past stimuli, people tend to report that the 2nd tone was higher, and when they are relatively low, they tend to report that the 1st tone was higher. Interestingly, only the most recent trials bias performance, which can be interpreted as if the participants assume that the statistics of stimuli in the experiment is highly volatile. Moreover, this bias persists even in settings, in which it is detrimental to performance. These results demonstrate the abilities, as well as limitations, of the cognitive system when incorporating expectations in perception.
| Perception is a complex cognitive process, in which noisy signals are extracted from the environment and interpreted. It is generally believed that perceptual resolution is limited by internal noise that constrains our ability to differentiate physically similar stimuli. The magnitude of this internal noise is typically estimated using the 2-alternative forced choice (2AFC) paradigm, which was introduced to eliminate participants' perceptual and response biases [1], [2]. In this paradigm, a participant is presented with two temporally-separated stimuli that differ along a physical dimension and is instructed to compare them. The common assumption is that the probability of a correct response is determined by the physical difference between the two stimuli, relative to the level of internal noise. Performance is typically characterized by the threshold of discrimination, referred to as the Just Noticeable Difference (JND). Thus, the JND is a measure of the level of internal noise such that the higher the JND, the higher the inferred internal noise.
However, the idea that there is a one-to-one correspondence between the JND and the internal noise is inconsistent with theoretical considerations which postulate that participants' performance can be improved by taking into account expectations about the stimuli in the process of perception or decision-making. If the internal representation of a stimulus was uncertain, the prior expectations should bias the participant against unlikely stimuli. The larger the uncertainty, the larger the contribution of these prior expectations should be. The Bayesian theory of inference describes how expectations regarding the probability distribution of stimuli should be combined with the noisy representations of these stimuli in order to optimize performance [3].
In fact, expectations, formalized as prior distribution of stimuli used in the experiment, have been shown to bias participants' responses in a way that is consistent with the Bayesian framework (reviewed in [4]). In particular, responses in the 2AFC paradigm have been shown to be biased by prior expectations: when the magnitudes of the two stimuli are small relative to the distribution of stimuli used in the experiment, participants tend to respond that the 1st stimulus was larger, whereas they tend to respond that the 2nd stimulus was larger when the magnitudes of the two stimuli are relatively large [5]–[7]. In a previous study we have shown that this bias, known as the “contraction bias”, can be understood in the Bayesian framework: following the presentation of the two stimuli, the participant combines her noisy representations of the two stimuli with the prior distribution of the stimuli to form two posterior distributions. Rather than comparing the two noisy representations of the stimuli, the participant is assumed to compare the two posteriors in order to maximize the probability of a correct response. The contribution of the prior distribution to the two posteriors is not equal. The larger the level of noise in the representation of the stimulus, the larger is the contribution of the prior distribution to the posterior. The level of noise in the representation of the magnitude of the 1st stimulus is larger than the level of noise in the representation of the magnitude of the 2nd stimulus because of the noise associated with the encoding and maintenance of the 1st stimulus in memory [8], [9]. As a result, the posterior distribution of the 1st stimulus is biased more by the prior distribution than the posterior distribution of the 2nd stimulus. If the prior distribution is unimodal, both posteriors are contracted towards the median of the prior distribution. Because the posterior of the 1st stimulus is contracted more than the posterior of the 2nd stimulus, participants' responses are biased towards overestimating the 1st stimulus when it is relatively small and underestimating it when it is relatively large [7].
One limitation of the Bayesian model is that it relies heavily on the assumption that the prior distribution of stimuli is known to the observer. While this assumption may be plausible in very long experiments comprising a large number of trials (e.g. thousands in [10]) or in experiments utilizing natural tasks (e.g., reading, [11]), it is unclear how Bayesian inference can take place if participants have less experience in the task.
In this paper we study participants' pattern of responses in a 2AFC tone discrimination task in relatively short experiments consisting of tens of trials. We report a substantial contraction bias that persists even when it hampers performance due to a-typical statistics. We show that participants' pattern of behavior is consistent with an “implicit memory” model, in which the representation of previous stimuli is a single scalar that continuously updates with examples. Thus, this model can be viewed as a simple implementation of the Bayesian model that provides a better account of participants' perceptual decision making.
We measured the performance of our participants in the random 2AFC paradigm (Materials and Methods, Fig. 1), in which subjects compared the frequencies of two sequentially presented tones drawn from a broad frequency range. Averaged across the population of participants, the JND was 13.6%±0.7% (SEM), which is higher than typically reported in the literature ([12], [13]). The relatively high value of the JND, which is likely to result from the lack of experience of the participants and the fact that no reference was used, is comparable with previous studies using the random frequency paradigm, with short stimuli and untrained participants [14], [15].
As predicted by the contraction bias, the JND did not capture the full pattern of participants' responses. This is depicted in Fig. 2A. The coordinates of each dot in Fig. 2A correspond to the frequencies of the 1st and 2nd tones in a trial, referred to as and . Blue and red dots denote trials, in which the participant's response was correct and incorrect, respectively. The closer the dots are to the diagonal, the smaller is the difference in the frequencies of the two tones. Therefore naively, one would expect that the probability of a trial to be incorrect (red dot) would be highest near the diagonal. Moreover, if the probability of a correct response as a function of is symmetrical around 0, as implicitly assumed when measuring the JND, then the pattern of red and blue dots is expected to be symmetrical around the diagonal. In contrast, we found that the pattern of incorrect responses is highly non-symmetrical. Participants tended to err more when both frequencies were high and and when both frequencies were low and . To quantify this asymmetry, we considered separately two regions: the Bias+ region corresponds to trials in two sections of this plane (yellow in Fig. 2A): in the first section are trials in which the frequencies of both stimuli are above the median (1000 Hz) and the frequency of the 1st tone is lower than that of the 2nd tone. In the second section are trials in which the frequencies of both stimuli are below the median frequency and the frequency of the 1st tone is higher than that of the 2nd tone. Similarly, The Bias− region (gray in Fig. 2A) corresponded to trials in which the frequencies of both stimuli are above the median (1000 Hz) and the frequency of the 1st tone is higher than that of the 2nd tone and trials in which the frequencies of both stimuli are below the median frequency and the frequency of the 1st tone is lower than that of the 2nd tone. Participants' rate of success differed greatly between the Bias+ and Bias− regions. Participants were typically successful when either the two tones were low (<1000 Hz) and the 2nd tone was lower (lower left yellow region, 88.2%±0.5% correct responses, mean ± SEM) or when the two tones were high (>1000 Hz) and the 2nd tone was higher (upper yellow region, 88.4%±0.6% correct responses). On the other hand, performance was relatively poor either when the two tones were low and the 1st tone was lower (lower left gray region, 63.2%±0.8% correct responses) or when the two tones were high and the 1st tone was higher (upper gray region, 61.8%±0.8% correct responses). These effects were highly significant in each of the two quadrants (p<10−6, Monte Carlo Permutation test). The differential level of proficiency in the yellow and gray regions indicates a substantial contraction bias, in line with that bias described in previous studies [6], [7]: when the frequency of the 1st tone was relatively low, participants tended to overestimate it (leading to successful performance when the 1st tone was higher). The opposite was true when the frequency of the 1st tone was relatively high (leading to successful performance when the 1st tone was lower). The differential level of proficiency in the yellow and gray regions is evident not only in the response pattern of the population of participants but also in the response pattern in individual blocks (Fig. S1A–C). Moreover, it was evident for all levels of proficiency in the task (Fig. S1D).
To further illustrate the contraction bias, we constructed a two-dimensional histogram of participants' performance by binning the space of Fig. 2A and computing the fraction of correct responses in each bin (Fig. 2B, grayscale). The non-symmetrical distribution of the shades of gray of the squares around the diagonal reflects the contraction bias. Note in particular the two squares denoted by arrows. Despite the fact that they were of equal ‘objective’ difficulty (the absolute difference in frequencies was the same), the performance in the bottom right square region was almost perfect (92.2% correct responses; n = 324), whereas it was almost at chance level in the top left square region (50.8% correct responses; n = 323; p<10−33, Fisher's exact test). It should be noted that the bias in participants' response cannot be accounted for by a general preference in favor of one of the alternative answers, because the bias is opposite in the low and high frequencies.
The non-symmetrical performance around the diagonal (Fig. 2) is not captured by a single performance measure, the JND. This has motivated us to consider a measure of performance that captures some of this asymmetry. To that goal, we computed two separate JNDs for each participant (see Materials and Methods): one for the trials in the regions in which the contraction bias augments behavior (Bias+, yellow) and the other for the regions in which the contraction bias impairs behavior (Bias−, gray). These JNDs differed by more than 6 fold (the medians of JNDs across the population were 4.1%, and 27.0% for the Bias+ and Bias− regions, respectively; p<10−5, Monte Carlo Permutation test). In fact, as depicted in Fig. S2 in the Supporting Information section, a participant's proficiency on a trial depended more on the contraction bias (i.e. Bias+ versus Bias− regions) than on the participant's overall proficiency (overall low versus high JND). These results demonstrate the substantial contribution of this bias to behavior.
In a previous study we have shown that the contraction bias in a visual discrimination task is consistent with a model of an ideal detector that utilizes Bayes' rule to incorporate the prior distribution with the sensed stimuli in order to optimize performance [7]. In agreement with that study, such a Bayesian model, with 2 free parameters that correspond to the noise in the internal representation of each of the two stimuli, can qualitatively account for the observed contraction in the two-tone discrimination task (see Fig. S3 in the Supporting Information section).
However, it should be noted that the Bayesian model relies on the assumption that the prior distribution of stimuli is known to the observer, which seems unreasonable in our experiment, which consisted of merely tens of trials. Therefore, it is not clear how the history of trials experienced by the participants in the experiment contributes to the bias. To address this question, we considered the contribution of individual trials to the bias. Because the statistics of stimuli in our experiment are stationary, all past trials are equally informative about the prior distribution. Therefore, normative considerations that incorporate an assumption of stationarity imply that the effect of past trials on participants' choices will be independent of the number of trials elapsed between these trials and the choice. By contrast, previous studies have reported that participants' responses are influenced to a greater degree by recent stimuli, which is known as the recency effect [16]–[21]. In addition, the activity of neurons in the primary auditory cortex has been shown to contain information about both current and previous stimuli [22]. To test for recency in our dataset, we fitted a linear non-linear model that relates the response in each trial to a linear combination of present and past stimuli according to the following equation:(1)where is the probability that the model would report that the frequency of the 1st tone was higher than that of the 2nd tone in trial ; is the normal cumulative distribution function such that ; and are parameters, and are the frequencies of the 1st and 2nd tone, respectively, in trial and is the geometric mean of the frequencies of all stimuli in the experiment until trial .
To gain insights into the behavior of the model (Eq. (1)) we consider the simple case in which and . In this case, Eq. (1) becomes , which corresponds to a model participant that is indifferent to the history of the experiment and its choices depend solely on the ratio of the frequencies of the two tones and the internal noise. The value of denotes the level of internal noise of the model participant. If is very small, then independently of the frequencies of the stimuli, and , , and the model participant responds at random. In contrast, if is very large, then where is the Heaviside step function such that for and for . In other words, if is very large the model participant always answers correctly. The larger the value of , the smaller the JND of the model participant. The values of the parameters determine the contribution of past stimuli to perception, where the value of determines the contribution of the stimulus presented trials ago and the value of determines the contribution of the average frequency of past stimuli to perception. If all past stimuli contribute equally to perception, as expected from normative participants who assume that the distribution of stimuli is stationary then we expect and . In contrast, if the participant assumes that the statistics of the experiment is non-stationary then we expect the most recent trials to have a stronger effect on behavior, resulting in whose magnitude decreases as the value of increases.
Assuming that , we analyzed the sequence of frequencies and decisions of our participants. We found the values of the parameters (Fig. 3, green), (dark blue) and (black) that minimize the mean square error (MSE), the mean square distance of the vector of probabilities, from the vector of choices, such that if the participant responded that the frequency of the 1st tone was higher than the frequency of the 2nd tone in trial and otherwise. Note that the values of and in Fig. 3 are larger than the values of all other coefficients, . This result reflects the simple fact that the tones presented in a trial influence the decision in that trial more than tones presented in previous trials. The recency effect is manifested in the non-zero coefficients of (see Materials and Methods). As depicted in Fig. 3, the contribution of past trials to choice diminishes within several trials. This result is consistent with other findings of rapid perceptual learning [23], [24] (but see also [25]) and demonstrates that at least some aspects of the prior distribution are estimated using a small number of the most recent trials. It should also be noted that the contribution of past stimuli to decision is dominated by past values of and not past values of (Fig. 3. See also Materials and Methods).
The recency effect described in the previous section is difficult to reconcile with a Bayesian inference model that takes into account the stationary statistics of the experiment. This finding has motivated us to consider the possibility that the contraction bias described in Fig. 2 emerges from simpler cognitive processes that do not require an explicit representation of the prior distribution. In this section we present a simple model that accounts for the contraction bias and the recency effect, which does not explicitly keep track of the prior distribution of stimuli presented in the experiment.
In our model, the memory trace of past stimuli is a single scalar (rather than the full prior distribution). In response to the presentation of , the participant updates the value of such that is a linear combination of the past value of with the present stimulus, corrupted by sensory and encoding noise. Formally, the value of in trial , is given by(2)where is the weight given to the memory and is the noise associated with the encoding of . We assume that this noise is Gaussian with variance and is uncorrelated across trials: , where is the Kronecker delta function, if and if .
A decision in a trial in this model depends on the relative values of and . If , the model responds that “”. Otherwise it responds that “”. In this model we assume that the noise is restricted to the representation of . The reason for ignoring noise in the representation of is that noise in is mathematically equivalent to a larger magnitude noise in when considering decision in a given trial.
It is easy to show that in this model, is an exponentially weighted sum of the current and past stimuli and their respective encoding noises:(3)Note that in this model past values of do not contribute to behavior. This reflects the dominance of past values of in Fig. 3 (see also Materials and Methods). It should also be noted that in this model, the contribution of past stimuli to decision (which plays the role of the prior distribution in the Bayesian model) is encoded using the same variable as the encoding of . Therefore, the model does not require any form of separate representation of the long term memory of past trials.
The implicit-memory model is characterized by two parameters that denote the level of noise, and the extent to which the history of the experiment affects perception, . Fig. 4 depicts the results of a simulation of a population of implicit memory models, each with the parameters and best fitting a single block in our dataset (see Materials and Methods). As shown in Figs. 4A and 4B, the model results in a contraction bias, which is comparable to the experimentally observed (compare Figs. 4A and 4B to Figs 2A and 2B, respectively). A quantitative analysis reveals that the goodness-of-fit of the Implicit memory model is comparable to that of the Bayesian model (Fig. S4). However, in contrast to be Bayesian model that assumes a constant prior, the contribution of very recent trials to performance (Eq. (1)) in the Implicit memory model is similar to that of our participants (compare Fig. 4C to Fig. 3).
The contraction bias in Fig. 2 can be justified using optimality considerations, in which prior knowledge is incorporated with the observations in order to maximize performance (Fig. S3). Would contraction bias persist in an experiment in which it impairs performance due to the dependencies between the frequency distribution of the two tones?
In order to address this question, we conducted a second experiment (Experiment 2 in the Materials and Methods), in which we manipulated the correlations between the frequencies of the two tones such that in some blocks the contraction bias is beneficial to performance whereas in others it is detrimental. Contraction bias is beneficial in the Bias+ region (yellow in Figs. 2A and 4A) and is detrimental in the Bias− region (gray in Figs. 2A and 4A). Therefore, in this experiment we manipulated the fraction of trials in the Bias+ and Bias− regions in different blocks. In one condition, the two tones were chosen such that the 2nd tone was typically higher than the 1st when the two frequencies were relatively high, and the 2nd tone was typically lower than the 1st when the two frequencies were relatively low. We refer to this condition as the ‘Bias+ condition’, because there were many more trials in the Bias+ region than in the Bias− region (11,233 vs. 1172). In the other condition, the two tones were chosen such that the 1st tone was typically higher than the 2nd when the frequencies of the two tones were relatively high and the 1st tone was typically lower than the 2nd when the frequencies of the two tones were relatively low. This ‘Bias− condition‘ was comprised of substantially more trials in the Bias− region than in the Bias+ region (8111 vs. 952). Figs. 5A and 5B depict the distribution of trials and correct and incorrect responses in the Bias+ and Bias− conditions, respectively. Similar to the pattern of responses in the first experiment (Fig. 2A), participants were more likely to be correct in the Bias+ regions, compared to the Bias− regions. This was true both for the Bias+ condition (82.0%±0.4% correct responses vs. 44.5%±1.6% correct responses, p<10−126 Fisher exact test) and the Bias− condition (88.0%±1.2% correct responses vs. 72.6%±0.6% correct responses, p<10−21 Fisher exact test). The JNDs were significantly different in the two conditions: the mean JND in the Bias+ condition was only 4.3%±0.6%, compared to 14.1%±1.1% in Bias− condition (Fig. 5C, black, p<10−25, Wilcoxon rank sum test).
In the framework of the Bayesian model, the difference in proficiency between the two conditions is surprising because given the joint distribution, the detection problem in the two conditions is symmetric. However, our results indicate that our participants did not utilize these probabilities when making a decision about the relative frequencies in this task.
To test the ability of the implicit-memory model to account for the results of the second experiment, we fitted the parameters of the model ( and ) to the experimental data of the Bias+ condition. We then simulated each of the model participants in both the Bias+ and Bias− conditions. The resulting JNDs (mean ± SEM 3.7%±0.5% and 13.3%±0.9% for the Bias+ and Bias− conditions, respectively, purple in Fig. 5C) are not statistically different from to the experimentally measured JNDs (4.3%±0.6% and 14.1%±1.6%; p = 0.78 and p = 0.85, respectively, Wilcoxon rank sum test), suggesting that the participants did not utilize the differential statistics of the two tones in the two conditions. For example, they did not decrease the weights of recent trials even when their performance was consequently hampered. In fact, adapting to the Bias− condition simply by setting the weight of the history-dependence parameter to 0 (effectively eliminating the contribution of past stimuli to decision in the model) would have improved their performance. To demonstrate this, we simulated the model participants in the Bias− condition while assuming that . The resultant JND was only 9.1%±0.7%, lower than the JND of the model participants when assuming the history-dependence parameter measured in the Bias+ condition.
In this work we showed that the contraction bias is a dominant determinant of participants' behavior in a 2AFC tone frequency discrimination task. Some aspects of this bias are consistent with the behavior of an ideal detector that utilizes the prior distribution to maximize performance. However, a substantial recency effect combined with a failure of the participants to utilize the joint distribution of the stimuli implies that this Bayesian-like computation is approximated using a much simpler algorithm, in which the prior distribution is not fully represented.
What information does the cognitive system store about the prior distribution? The full Bayesian model represents one extreme approach, in which it is assumed that the participant has full information about the joint distribution of the two stimuli. The standard way in which signal detection theory is applied to psychophysics represents the other extreme, in which the participant does not have (or does not utilize) any prior information about the identity of the stimuli (but only about the probability of each response being correct [1]). The contraction bias in Fig. 2 demonstrates that participants have some information about the marginal probabilities. However, the strong recency effect (Fig. 3) indicates that this marginal probability is constantly updated using a small number of most recent observations, even in stationary environments. In a normative framework, the recency effect, observed previously in various tasks [26], [27], implies that participants believe that the environment is highly volatile and as a result only the very recent history is informative about future stimuli.
The results of experiment 2 (Fig. 5) indicate that participants are either unable to compute the joint distribution or unable to utilize it, at least within a single experimental block of 80 trials. The implicit memory model can be viewed as a minimal modification of the standard approach of applying signal detection theory to perception in the direction of the full Bayesian model. Here, participants represent the prior distribution of the stimuli with a single scalar, which is an estimate of the mean of the marginal of the prior distribution. Nevertheless this implicit model captures many aspects of the behavioral results. Further studies are needed to determine whether, and to what extent other moments of the prior distributions are learned and utilized in the 2AFC discrimination task, especially under longer exposure to distribution statistics.
Several studies have shown that the magnitude of the contribution of the prior distribution to perception on a given trial depends on the level of internal noise [10], [28]. In particular in the framework of the 2AFC task, increasing the delay between the 1st and 2nd stimuli [29], [30] or introducing a distracting task between them [7] enhances the contraction bias. These results are consistent with the Bayesian approach. How can these results be accounted for in the framework of the implicit memory model? One possibility is to assume that the relative contribution of the prior in the simplified online rule of Eq. (2) is affected by perceptual noise. However, it should be noted that at least in one case, the level of noise was determined after the encoding of the 1st stimulus [7]. The dependence of on the level of noise can be accounted for in the framework of the implicit memory model if we assume that the computation of , which incorporates the prior knowledge with the response to the 1st stimulus, is carried out simultaneously by several neurons, or populations of neurons, which are characterized by different values of [22], [31], [32]. At the time of the decision, the magnitude of the noise determines which populations of neurons will be the most informative with respect to the 1st stimulus. If the level of noise is high, the populations of neurons that are more affected by past trials (for whom the value of is large) will dominate perception, resulting in a substantial contraction bias. Otherwise the populations that are less affected by past trials will dominate perception, resulting in a small contraction bias.
Almost 40 years ago, Tversky and Kahneman characterized irrational decision making and reasoning and concluded that “people rely on a limited number of heuristic principles which reduce the complex tasks … to simpler judgmental operations. In general, these heuristics are quite useful, but sometimes they lead to severe and systematic errors” [33]. Our study extends these results to the domain of implicit perceptual judgments.
The research was approved by the department ethics committee, and all participants signed consent forms.
150 participants (mean age 24±3.1 years) engaged in a 2AFC high/low pure tone frequency discrimination task, after signing consent forms. 18 participants were excluded due to poor performance on a hearing test or because they did not complete the full schedule. Each participant performed 2 blocks of 80 trials. Each trial consisted of two 50 ms pure tones, with 10 ms linear rise time, and 10 ms linear fall time, separated by 950 ms. Immediately after the 2nd stimulus was played, the text ‘Which tone was higher?’ appeared on screen, and the participant responded by clicking one of two on-screen buttons using a computer mouse, with no time constraint. Visual feedback of a smiling face or a sad face was presented for 300 ms after correct and incorrect responses, respectively. After a pause of 700 ms the next trial began (Fig. 1). All stimuli were presented binaurally through Sennheiser HD-265 linear headphones using a TDT System III signal generator (Tucker Davis Technologies) controlled by in-house software in a sound attenuated room in the laboratory. Tone intensity was 65 dB SPL. Both the 1st and the 2nd frequencies in each trial were drawn from a wide distribution according to the following procedure: a frequency was drawn from a uniform distribution between 800 Hz and 1200 Hz. Another frequency, either or was drawn with a probability 0.5, where was controlled by an adaptive 3-down 1-up staircase, in which the initial difference between the stimuli in each block was 20% and was bounded from below by 0.1%. The step size decreased every four reversals, from 4.5% to 2% to 1% to 0.5% to 0.1%. One of the two frequencies was randomly selected as and the other frequency was selected as . This schedule is expected to converge to a for which the participant answers correctly in 79.4% of the trials ([34]; Fig. 2A, dots). Blocks that did not converge to at least 65% correct responses in the last 40 trials were excluded from the analysis (12 of 264 blocks). The JND was calculated as the average difference between the stimuli frequencies in the last 6 reversals. As a result of the adaptive staircase schedule, the ratios between the frequencies of the two stimuli tended to decrease in the first trials of the block. On average, after 15 trials this ratio stabilized and therefore the first 15 trials of each block were excluded from the analysis.
To estimate the JND in a Bias+ or Bias− region of a block, we fitted a cumulative normal distribution function psychometric curve that relates the response in each trial to the difference in the logarithm of the 1st and 2nd frequencies: where is the normal cumulative distribution function, such that . The value of the parameter was chosen as to minimize the square difference between the vector predictions and the vector of choices such that on trials in which the participant responded “” and otherwise. Assuming that the cumulative normal distribution function reflects the probability of responding “”, the corresponding value of the JND is the difference in the natural logarithms of and such that the probability of a correct response is the asymptotic performance level in our staircase paradigm, 0.794. Therefore, .
To test for differences in performance between different regions, we used a Monte Carlo permutation test in which the identities of and in a trial were randomly shuffled. We used 106 permutations, and in all cases the experimentally observed differences were larger than the differences observed in all permutations, resulting in p<10−6.
To test for differences in the JNDs between different regions, we used a Monte Carlo permutation test in which the identities of and in a trial were randomly shuffled. We estimated the JND of these simulated results using the same process as described for the data, and estimated the median JND+ and median JND- for the whole population. We used 105 permutations and the experimentally observed difference was larger than the difference observed in all permutations, resulting in p<10−5.
In order to verify the contribution of the parameters for to the linear-non-linear model (Eq. 1), we compared several models using a cross validation test: the parameters of the different models were estimated using all blocks but one, and these parameters were used in order to compute the MSE for that block. The MSE of the model was computed by repeating this procedure for all blocks in the experiment and averaging the resultant MSE.
We considered three models: (1) a naïve model with no history dependence: ; (2) a model with a global history term, ; (3) the full model with an explicit history dependence of three previous trials, and a global term, . The resultant MSEs are ; ; . We found that is significantly smaller than and ( and respectively, Wilcoxon signed rank test).
In order to verify that the contribution of past trials is dominated by values of , we compared two additional models, using the same analysis as above: (4) a model in which the recent history is represented by only: ; (5) a model in which the recent history is represented by only: . The resultant MSEs are and . While is not statistically different from (), is significantly higher () indicating that the model with only coefficients corresponding to the contribution of is as predictive as the full model.
Experiment 2 was similar to experiment 1, except for the joint distribution of and : in each trial, a frequency was chosen such that the natural logarithm of , measured in Hz, was drawn from a normal distribution with mean 6.908 (corresponding to 1000 Hz), and standard deviation 0.115. In all trials, the mean of and (in the logarithmic domain) was . Another frequency, either or (in the logarithmic domain) was drawn with a probability 0.5, where was controlled by an adaptive 3-down 1-up staircase schedule. In contrast to Experiment 1, the order of frequencies was biased and depended on . In trials in which , was chosen to be larger than with a probability . In contrast, in trials in which , was chosen to be larger than with a probability . We studied two conditions: in one condition, which we refer to as “Bias+”, . In the second condition, referred to as “Bias−”, . 60 participants (mean age 23.8±3.3 years) that did not participate in experiment 1 performed 6 interleaved blocks of Bias+ and Bias− conditions, with the order counterbalanced across participants. Similar to experiment 1, each block consisted of 80 trials.
Rewriting Eq. (3), where is a “signal” term that depends on previous trials and is a “noise” term. The probability of responding “” response is thus given by , where is the normal cumulative distribution function, and is the standard deviation of . Because we excluded the first 15 trials from our analysis, we assumed that . We fitted the pair to the remaining 65 trials of each block to minimize the square error between the predictions of the model and the actual responses, .
|
10.1371/journal.pbio.2001461 | Neuronal synchrony and the relation between the blood-oxygen-level dependent response and the local field potential | The most widespread measures of human brain activity are the blood-oxygen-level dependent (BOLD) signal and surface field potential. Prior studies report a variety of relationships between these signals. To develop an understanding of how to interpret these signals and the relationship between them, we developed a model of (a) neuronal population responses and (b) transformations from neuronal responses into the functional magnetic resonance imaging (fMRI) BOLD signal and electrocorticographic (ECoG) field potential. Rather than seeking a transformation between the two measures directly, this approach interprets each measure with respect to the underlying neuronal population responses. This model accounts for the relationship between BOLD and ECoG data from human visual cortex in V1, V2, and V3, with the model predictions and data matching in three ways: across stimuli, the BOLD amplitude and ECoG broadband power were positively correlated, the BOLD amplitude and alpha power (8–13 Hz) were negatively correlated, and the BOLD amplitude and narrowband gamma power (30–80 Hz) were uncorrelated. The two measures provide complementary information about human brain activity, and we infer that features of the field potential that are uncorrelated with BOLD arise largely from changes in synchrony, rather than level, of neuronal activity.
| There are several methods for measuring activity in the living human brain. Here, we studied functional magnetic resonance imaging (fMRI), which depends on the vascular response to neuronal activity, and surface field potentials, which measure electrical activity from many neurons. These two widely used measurements of human brain activity often provide different and potentially conflicting results. We propose a quantitative model for how these two measurements integrate activity from neuronal populations. The fMRI signal is highly sensitive to the average level of local neuronal activity but not the degree of synchrony between neurons. In contrast, the field potential is most sensitive to synchronous neuronal signals. Our model accounts for several observations seen in fMRI and field potential data: some very large features of field potential recordings, such as gamma oscillations, can occur with little to no associated fMRI signal. The model predicts this because the gamma oscillations result more from increased neuronal synchrony than increased neuronal activity. Other field potential signals, such as broadband changes, which are likely driven by the level of neuronal activity rather than a change in synchrony, are highly correlated with fMRI. The two measures thus provide complementary information about human brain activity.
| Most measurements of activity in the living human brain arise from the responses of large populations of neurons, spanning the millimeter scale of functional magnetic resonance imaging (fMRI) and electrocorticography (ECoG) to the centimeter scale of electro- and magneto-encephalography (EEG and MEG). Integrating results across methods is challenging because the signals measured by these instruments differ in spatial and temporal sensitivity, as well as in the manner by which they combine the underlying neuronal population activity [1–3]. Differences in scale can be partially bridged by bringing the measurements into register. For example, EEG and MEG sensor data can be projected to cortical sources subject to constraints from simultaneously recorded fMRI data [4] or from independent fMRI localizers [5], and ECoG electrodes can be aligned to a high-resolution anatomical MRI image [6] and compared to the local fMRI signal.
Yet, even when electrophysiological and fMRI data are spatially registered, striking differences in the sensitivity to stimulus and task are often observed, indicating differences in how neuronal responses contribute to the measured physiological signals. For example, the fMRI blood-oxygen-level dependent (BOLD) signal and EEG evoked potentials differ in which brain areas are most sensitive to visual motion (area MT+ with fMRI [7] versus V1 and V3A with EEG [8]). Within the same visual area, fMRI and source-localized EEG evoked potentials can show different effects of task in similar experimental paradigms, such as the effect of spatial attention on the contrast response function (additive in fMRI [9], multiplicative in EEG [10]). Even when the spatial scale of the two signals is approximately matched at acquisition, such as ECoG electrodes and fMRI voxels (both at approximately 2 mm), systematically different patterns of responses can be obtained, such as compressive spatial summation in fMRI versus nearly linear summation in ECoG steady-state potentials (but not ECoG broadband signals) [11]. Such fundamental functional differences cannot be explained by numerical measurement-to-measurement transformations. Rather, these differences must reflect the fact that the measurements are based on different aspects of the neural population response. To explain the differences in measurement modalities requires a computational framework that derives each of these signals from the neuronal responses.
One approach toward developing such a framework has been to measure the BOLD signal and electrophysiological signals simultaneously, or separately but using the same stimulus and task conditions, and to ask how features of the electrophysiological response compare to the BOLD signal. This approach has revealed important patterns, yet after several decades of careful study, some apparent discrepancies remain. A number of studies comparing band-limited power in field potential recordings to the BOLD signal have shown that increases in power between 30 Hz and 100 Hz (gamma band) are more highly correlated with BOLD amplitude than power changes in other bands [12–17]. Yet, power changes in this band do not fully account for the BOLD signal: very large power changes can occur in the gamma band without a measurable BOLD signal change [18, 19], and power changes in lower frequency bands can be correlated with the BOLD signal independently of power changes in the gamma band [20–23]. It therefore cannot be the case that field potential power in the gamma band is a general predictor of BOLD, even if the two measures are often correlated. Another source of disagreement is that within the gamma band, some reports claim that BOLD is best predicted by synchronous (narrowband) signals [13], and others claim that BOLD is best predicted by asynchronous (broadband) neural signals [11]. Moreover, in some cases, it has been reported that no feature of the local field potential (LFP) predicts the intrinsic optical imaging signal (closely related to BOLD) as accurately as multiunit spiking activity [24]. Consistent with this claim, a comparison of both motion and contrast response functions measured with single units and with BOLD suggested a tight coupling between BOLD and single-unit responses [25–27]. To our knowledge, there is currently no single model linking the electrophysiological and BOLD signals that accounts for the wide range of empirical results.
The numerous studies correlating features of electrophysiological signals with BOLD provide constraints in interpreting the relationship between the two types of signals, yet the approach has not led to a general, computational solution. We argue that one reason that correlation studies have not led to computational solutions is that any particular feature of the field potential could be caused by many possible neuronal population responses. For example, a flat field potential (minimal signal) could arise because there is little activity in the local neuronal population or it could arise from a pair of neuronal subpopulations responding vigorously but in counterphase, resulting in cancellation in the field potential. The same field potential in the two situations would be accompanied by different levels of metabolic demand and presumably different levels of BOLD signal. Similarly, any particular BOLD measurement could be due to many different patterns of neural activity. For example, stimulation of a neuronal population that inhibits local spiking can cause an elevation in the BOLD signal [28], as can stimulation of an excitatory population that increases the local spike rate [29]. In short, there can be no single transfer function that predicts the BOLD signal from the field potential because the field potential does not cause the BOLD signal; rather, the neuronal activity gives rise to both the field potential and the BOLD signal.
We propose that many of the different claims pertaining to the relationship between BOLD amplitude and features of the field potential can be accounted for by a modeling framework in which BOLD and field potential measurements are predicted from simulated neuronal population activity, rather than by predicting the BOLD signal directly from the field potential. In this paper, we model fMRI and ECoG responses in two stages, one stage in which we simulate activity in a population of neurons, and a second stage in which we model the transformation from the population activity to the instrument measures. By design, the model employs a minimal set of principles governing how the instruments pool neuronal activity, rather than a biophysically detailed description of neuronal and hemodynamic events. This approach enables us to ask whether this minimal set of principles is sufficient to guide simulations of neuronal population activity, such that the parameters of the simulations are fit to ECoG measurements from human visual cortex, and the output of the simulations predicts fMRI BOLD responses in the same regions for the same stimuli.
We first present an analytic framework to capture basic principles of how the BOLD signal and the field potential pool neuronal signals across a population (2.1). Using this framework, we derive equations for the relationship between each instrument measure (BOLD and LFP) and the underlying neuronal activity, as well as the relationship between the instrument measures. This section shows that synchrony is expected to have a large effect on the LFP signal but not on the BOLD signal. The analytic framework provides a way to derive the instrument measures from neuronal population activity, but it does not specify the neuronal population activity itself. In the next section (2.2), we develop a method for simulating neuronal population time series from a small number of parameterized inputs, and we show how the simulated neuronal activity can be converted to (simulated) LFP and BOLD by applying the equations derived in 2.1. Next, we fit parameters for simulating population neuronal activity using ECoG data from human V1, V2, and V3 and compare the BOLD responses derived from these simulations to measured BOLD responses from V1, V2, and V3 (2.3). Finally, we quantify the relationship between simulated BOLD and LFP, and between measured BOLD and ECoG, and show that the same patterns hold for simulation and data (2.4).
The fMRI BOLD signal and the LFP measure neuronal population activity in a fundamentally different manner. The goal of this analytic framework is to capture these differences in simple mathematical expressions and from these expressions, derive the relationship between the two instrument measurements. We purposely omit a large number of biophysical details, such as cell types, neuronal compartments, the dynamics of blood flow, and so forth, both for tractability and in order to emphasize the basic principles of how different measures integrate neuronal activity. In the sections that follow, we then show that, when coupled to simulated neural responses, the model can account for many important patterns observed in fMRI and ECoG data from human visual cortex.
For this analytic framework, we consider how a population of n neurons responds to a stimulus or task during a brief epoch (time 0 to T), assumed to be on the order of a second. Each neuron will produce a time-varying dendritic current, denoted as Ii(t) for the ith neuron, resulting from the trans-membrane potential. We would like to know how these currents, I(t), relate to the fMRI BOLD signal and to the LFP signal measured by an ECoG electrode.
We assume that the LFP arises primarily from dendritic membrane currents [2]. We ignore output spikes. (Although spikes can influence the LFP [30], it is generally thought that the influence is smaller than synaptic and dendritic currents [2], and including spikes would not change the logic of our arguments.) For the ith neuron, the contribution to the LFP is then αi × Ii(t). The constant αi depends on the distance and orientation of the neuron with respect to the electrode, as well as the electrode’s impedance. For simplicity, we assume that each neuron is equidistant from the electrode and has the same orientation, like pyramidal neurons perpendicular to the cortical surface, and therefore its contribution to the electrode measurement is scaled by the same constant, α. These neurons act together like a single, equivalent circuit, and hence the LFP time series will sum the contribution from each neuron,
LFP(t)=α∙∑inIi(t)
(Eq 1)
Field potential recordings are often summarized as the power (or band-limited power) in the time series [31]. Here we summarize the LFP response within a short time window as the power in the signal summed over the time window T:
LFPpower=α∙∫0T([∑inIi(t)]2)dt]powerofsum
(Eq 2)
Importantly, Eq 2 is a linear/nonlinear (L/N) computation, since the LFP power is computed by first summing the signals (L), and then computing the power (N).
The BOLD signal pools neural activity in a fundamentally different manner because it depends on metabolic demand [e.g., for reviews, see 1, 32]. (Recent work on the neurobiology of neurovascular coupling indicates that much of the BOLD signal is not caused directly by changes in the level of metabolic products such as glucose, but rather by signaling molecules that tend to correlate with metabolic demand [33]). For simplicity, we discuss the BOLD signal throughout in terms of metabolic demand and return to this issue explicitly in the Discussion, (3.6 The role of a simple model in understanding the relation between BOLD and LFP). The metabolic demand of each neuron will increase if the cell depolarizes (excitation) or hyperpolarizes (inhibition) [28, 34, 35]. Hence, the metabolic demand of a neuron is a nonlinear function of its membrane potential: either a positive or negative change in voltage relative to resting potential causes a current, thereby resulting in a positive metabolic demand. There are many possible nonlinear functions one could assume to summarize the metabolic demand from the dendritic time series, such as the rectified signal (absolute value) or the power (squared signal). For tractability, we assume the metabolic demand of the ith neuron is proportional to the power in the time-varying trans-membrane current integrated over time: βi × (POWER(Ii(t)) or βi×∫0T(Ii(t)2)dt, with βi a scaling constant for the ith neuron. (Similar results were obtained if we used the absolute value rather than the power). For the entire population of n neurons, we then assume the BOLD signal will sum the metabolic demand of each neuron. For simplicity, we use the same scaling constant for each neuron:
BOLD=β∙∑in(∫0T(Ii(t)2)dt)sumofpower
(Eq 3)
Importantly, Eq 3 is an N/L computation, since the power is computed first (N) and then the signals are summed (L), opposite to the order of operations for the LFP in Eq 3 (Fig 1) (personal communication from David J Heeger). In other words, we approximate the BOLD signal as the sum of the power, and LFP as the power of the sum, of the separate neuronal time series. The difference in the order of operations can have a profound effect on the predicted signals, as in the simple example with two neurons depicted in Fig 1C and 1D. The BOLD signal pooled over the two neurons is the same whether the time series from the two neurons are in phase or out of phase, whereas the LFP power is large when the time series are in phase and small when they are out of phase.
These approximations allow us to make predictions about the relation between LFP and BOLD. By theorem, we know that the power of the sum of several time series is exactly equal to the sum of the power of each time series plus the sum of the cross-power between the different time series (Eq 4):
∫0T([∑inXi(t)]2)dt=∑in(∫0T(Xi(t)2)dt)+∑i≠jn(∫0T(Xi(t)∙Xj(t))dt)PowerofsumSumofpowerSumofcross-power
(Eq 4)
Applying this theorem to Eqs 2 and 3 shows the relationship between our models of BOLD and LFP power:
LFPpower=αβ∙BOLD+α∙∑i≠jn(∫0T(Ii(t)∙Ij(t))dt)
(Eq 5)
We can now see that the LFP power depends on two quantities, one of which is related to the BOLD signal, and one of which is unrelated to the BOLD signal (Eq 5). The first quantity summarizes the total level of neural activity (summed across neurons), and the second quantity summarizes the relationship between neural time series (the cross-power, similar to covariance). If and when the second term tends to be large compared to the first, then the LFP power will not be closely related to the BOLD signal.
One cannot deduce from first principles whether the first term in Eq 4 (summed power) or the second term (summed cross-power) will dominate. However, the number of elements contributing to the two terms is quite different: For n neurons, the first term has n numbers (the power in each neuron’s time series), whereas the second term has nearly n2 numbers (all the pairwise cross-powers). Hence, if there is any appreciable covariance, then the LFP power will be dominated by the second term, and the correlation with BOLD will be weak (except in cases where the cross-power and power are highly correlated).
To see how these equations translate to quantitative measures of BOLD and LFP, we consider a small neuronal population whose time series conform to a multivariate Gaussian distribution. We assume that each neuron’s time series has the same mean, m; the same variance, σ2; and all of the pairwise correlations have the same value, ρ:
X∼N(μ,Σ)μ=(m⋮m)Σ=[σ2⋯σ2ρ⋮⋱⋮σ2ρ⋯σ2]
(Eq 6)
X is the population time series, μ is the mean of each time series, and ∑ is the covariance matrix. We can now rewrite the simulated BOLD signal (the sum of the power) and the LFP (power of the sum) in terms of the parameters of the multivariate Gaussian (and arbitrary scaling factors α, β),
BOLD=β∙[n∙(m2+σ2)]LFPpower=α∙[n∙(m2+σ2)+(n2−n)(m2+σ2ρ)]
(Eq 7)
where n is the number of neurons. This enables us to visualize how the BOLD signal and the LFP power depend on just three values: the variance, correlation, and mean in the neural time series, rather than on all the individual time series (Fig 2). For these neuronal time series, the LFP, modeled as the power of the sum of neuronal time series (panel A), is dominated by the neuronal cross-power (panel C). The BOLD signal, modeled as the sum of the power in the neuronal time series (panel B), makes little contribution to the LFP, except when the correlation between neurons is low (ρ is close to 0); in this case, there is no cross-power, and BOLD and LFP power are correlated.
In section 2.1, we proposed formulae to derive instrument measures from neuronal population activity. Here, we ask how we might simulate neuronal activity with a small number of parameters. A low-dimensional characterization of the population activity is useful since we normally do not have access to the time series of an entire population of neurons. Moreover, a low-dimensional representation can lead to better understanding and generalization even when high-dimensional data are available [36, 37]. After simulating the population activity, we then use the analytic framework from section 2.1 to compute the BOLD and LFP signals. The parameters for the simulations were fit to ECoG recordings from human V1, V2, and V3 [38]. Because there were recordings from multiple electrodes and multiple stimuli, we ran multiple simulations fit to the different ECoG responses. We then used these simulations to predict the BOLD signal and compared these predictions to the measured BOLD signal for the same stimuli and same cortical locations (but in different observers). The steps for simulating the neuronal population data and the derived LFP and BOLD, and for comparing the simulations to empirical data, are summarized in Table 1.
In the ECoG experiments, there were four grating stimuli of different spatial frequencies, three noise patterns with different power spectra, and one blank stimulus (mean luminance). For each of the 8 stimuli and each of 22 electrodes in V1, V2 and V3, we decomposed the measured ECoG responses into three spectral components: broadband, narrowband gamma, and alpha (Fig 3). An important feature of this dataset is that the three components of the ECoG responses showed different patterns across stimuli [38]: stimuli comprised of noise patterns caused large broadband increases but little to no measureable narrowband gamma response, whereas grating stimuli elicited both broadband increases and narrowband gamma increases. Gratings and noise stimuli both resulted in decreases in alpha power compared to baseline (also see S1 Fig). Had all three responses been tightly correlated with each other, it would not be possible to infer how each relates separately to the BOLD signal.
The simulations were structured to approximate the experimental design and the results of our ECoG experiments. To match the design of our ECoG experiments, a simulated experiment consisted of 240 trials, each of which were 1 second long (30 repeats of 8 conditions). The LFP time series were transformed to power spectra, which were averaged across the 30 repeated trials of the same condition. The simulation parameters—i.e., the level of the three inputs, C1, C2, and C3—were fit to the measured ECoG summary metrics (broadband, gamma, and alpha) for each of the 8 conditions for a particular electrode (Fig 6). To verify the validity of this procedure, we asked whether the simulations using the fitted parameters produce simulated spectra, which, when analyzed like the ECoG spectra, reproduce the original values of broadband, gamma, and alpha. In other words, do we close the loop from measured spectral components (broadband, gamma, and alpha) to inferred input parameters (C1, C2, C3) to simulated population activity, to simulated spectral components (broadband, gamma and alpha)? The original values are not reproduced exactly because the simulations are stochastic, but overall, the original broadband, gamma, and alpha values are recovered with high accuracy (S9 Fig).
As described above, the fitting of the parameters for C1, C2, and C3 was constrained by the assumptions that for C1, the correlation between neurons was 0 (and the amplitude was varied for fitting); for C2, the amplitude was fixed at a nonzero value (and the correlation was varied for fitting); and for C3, the correlation was fixed at a nonzero value (and the amplitude was varied for fitting). Results from alternative models with different constraints show poorer fits and are described briefly below and more extensively in S7 Fig.
Importantly, the parameter fits did not take into account the measured BOLD responses. Hence the simulations provided a test: if the input parameters were chosen to produce outputs that match the measured ECoG responses (training data), does the simulated BOLD signal accurately predict the measured BOLD signal (test data)? We measured BOLD responses in four healthy subjects to the same visual stimuli as used in ECoG (subjects are different from the ECoG subjects) and extracted the signal from regions of interest in visual cortex matched to the previously recorded ECoG electrode locations (S2 Fig and S3 Fig). For an example V1 site, the predicted BOLD signal accurately matched the measured BOLD signal, with 89% of the variance in the measured BOLD signal explained by the prediction, as quantified by R2—the coefficient of determination (Fig 7A). Across V1 sites, the predicted BOLD signal from the simulations accounted for a median of 80% of the variance in the measured data (Fig 7C). For an example V2 site, the predicted BOLD signal also matched the measured BOLD signal (R2 = 0.74, Fig 7B). Across V2/V3 sites the simulations explained a median of 40% of the variance in the data. The explained variance in V2/V3 is substantial but lower than in V1. One likely reason for the higher variance explained in V1 is that for the particular stimuli used in these experiments (gratings and noise patterns), the BOLD response reliability was higher in V1. For example, the median R2 computed by using half the BOLD data as a predictor for the other half (split half by subjects) was 86% for V1 and 63% for V2/V3. Similarly, the stimulus-evoked BOLD responses in V1 were larger than in V2 and V3, with more stimulus-related variance to explain: a mean of 1.8% signal change in V1 versus 1.2% in V2 and 0.8% in V3 (S4 Fig). It is possible that a stimulus set more tailored to extrastriate areas, such as textures or more naturalistic scenes, would have evoked more reliable responses in extrastriate cortex.
For each of the 22 simulations, the three input parameters C1, C2, and C3 defining each of the 8 stimulus conditions were fit to produce the LFP data from the corresponding ECoG electrode. By design, the C1 (broadband) and C3 (alpha) inputs were fit to ECoG data by varying the level per neuron, whereas C2 was fit to data by varying the correlation across neurons. In principle, for any of the three inputs, the ECoG data could have been fit by varying either the level per neuron or correlation across neurons. For completeness, we tested all 8 combinations of models (S7 Fig). The most accurate model, quantified as the R2 between the measured BOLD and the simulated BOLD (median across the sites in V1 or in V2/V3), was the simulation type used in the main text, in which C1 and C3 varied in the level per neuron and C2 varied in the correlation across neurons. Models in which the broadband correlation rather than level was used to fit the ECoG broadband power were much less accurate. The models in which the gamma LFP power was fit by modulating the level rather than the correlation in the simulated population caused a small drop in R2.
The previous analysis showed that when simulations were fit to ECoG data, the simulated BOLD response predicted the measured BOLD response. Here, we used regression analyses to assess how the simulated LFP predicted the simulated BOLD response and how the measured LFP predicted the measured BOLD response.
This study investigated the relationship between electrophysiological and BOLD measurements in human visual cortex. Our modeling framework decomposed the signals into two stages, a first stage in which we simulated the neuronal population responses (dendritic time series), and a second stage in which we modeled the transfer of the neuronal time series to the BOLD signal and the field potential. This approach differs from the direct comparison of electrophysiological signals and BOLD. The explicit separation into stages clarified both a similarity and a difference between the BOLD amplitude and the field potential power: the two can be approximated as the same operations on the neuronal population activity, but applied in a different order. Specifically, within a brief window, we modeled the BOLD amplitude as the sum of the power in the neuronal time series and the field potential as the power of the sum of the neuronal time series. Because the order of operations differs, the two signals differ, and each is blind to particular distinctions in the neuronal activity. For example, the BOLD signal (according to our model) does not distinguish between synchronous and asynchronous neural signals with the same total level of activity. In contrast, the field potential does not distinguish counterphase responses from no response. Even if one knew the exact mechanism of neurovascular coupling and the precise antenna function of an electrode, one still could not predict the relationship between the BOLD signal and the field potential without specifying the neuronal population activity that caused both. Hence, the relationship between the two types of signals is not fixed but rather depends on the structure of the underlying responses of the neuronal population.
Although we do not have access to the complete set of individual neuronal responses in any of our experiments in visual cortex, we approximated the responses by specifying the type of signals common to visual cortex. We therefore limited the space of neuronal population responses by modeling the activity as arising from three types of signals, enabling us to compute the complete set of field potentials and BOLD responses to a variety of conditions. Finally, we compared the simulated patterns of BOLD and field potential responses to the actual responses we observed in data from human subjects. These patterns are discussed and interpreted below.
Many studies have reported correlations between BOLD and power in the gamma band LFP (30–130 Hz) (review for human studies: [52]). Yet, changes in gamma band power do not reflect a single biological mechanism. For example, several recent studies have emphasized that LFP power changes in the gamma band reflect multiple distinct neural sources, including narrowband oscillations and broadband power shifts, with very different stimulus selectivity and biological origins [38, 53, 54].
Broadband changes have been proposed to reflect, approximately, the total level of Poisson-distributed spiking (or spike arrivals) in a local patch of cortex [40]. In contrast, the narrowband gamma response is caused by neural activity with a high level of cell-to-cell synchrony [55] and likely depends on specialized circuitry [56]. While the two responses are sometimes distinguished as “high gamma” (referring to broadband signals) and “low gamma” (referring to oscillatory signals), this distinction is not general. Broadband signals can extend into low frequencies [11, 57] so that the two signals can overlap in frequency bands. Hence, separating gamma band field potentials into an oscillatory component and a broadband (nonoscillatory) component is not reliably accomplished by binning the signals into two temporal frequency bands, one low and one high, but rather requires a model-based analysis, such as fitting the spectrum as the sum of a baseline power law (to capture the broadband component) and a log-Gaussian bump (to capture the oscillatory component) [38].
There is strong experimental support for the idea that increases in broadband LFP power primarily reflect increases in asynchronous neural activity rather than increases in coherence. First, experiments have shown that broadband power is correlated with multiunit spiking activity [54, 58]. Second, unlike the case of narrowband gamma LFP, changes in broadband LFP are not accompanied by changes in broadband spike-field coupling (Figure 1A-B in [43]). The possibility that neuronal synchrony sometimes affects broadband signals cannot be ruled out, for example, as shown in cases of pharmacological manipulations in nonhuman primates [59]. In such cases, there would not be a simple relationship between broadband power and BOLD.
The prior literature has not shown definitively whether broadband LFP, narrowband gamma, or both predict the BOLD signal. The first study that directly compared simultaneously recorded BOLD and electrophysiology showed that both LFP power in the gamma frequency range (40–130 Hz) and multiunit spiking activity (MUA) predicted the BOLD signal [16] and further, that when the LFP power diverged from MUA, the gamma band LFP predicted the BOLD signal more accurately than did spiking. This study however did not separately test whether a narrowband (oscillatory) or a broadband (nonoscillatory) component of the LFP better predicted the BOLD response.
Other studies, too, have shown a variety of patterns when correlating LFP power changes in the gamma band with BOLD. Some reported that BOLD amplitude correlates with narrowband gamma activity [13], while others showed that BOLD correlates with broadband changes [11], and many did not distinguish narrowband from broadband power in the gamma band [60]. Simultaneous recordings of hemodynamic and neuronal activity in macaque V1 showed that BOLD signals from intrinsic optical images can occur in the absence of gamma band LFP changes [61] and that, in some circumstances, multiunit activity predicts the BOLD signal more accurately than gamma band LFP [24, 62].
Here, we separately quantified the broadband power (spanning at least 50–150 Hz) and narrowband gamma power. We found that the amplitude of broadband changes accurately predicted the BOLD signal in V1. The empirical results and the models help resolve the question of why “high gamma” has been shown to correlate with BOLD, and “low gamma” sometimes does not [24]. The likely reason is unrelated to the difference in frequency range, nor to the size of the spectral perturbation in the LFP. In fact, the elevation in broadband power is relatively small (2- or 3-fold) compared to the elevation in power often observed in narrowband gamma oscillations (10x or more)[38]. Instead, “high gamma” is predictive of the BOLD signal in many cases not because of the specific frequency range, but because this signal captures the level of asynchronous neuronal response; this signal happens to be most clearly visible in the high-frequency range (>100 Hz) in which it is not masked by rhythmic lower-frequency responses. Hence, the distinction in predicting the BOLD response is not about “high” versus “low” gamma but rather synchronous versus asynchronous responses, and the broadband signal, sometimes labeled high gamma, maps onto the first term on the right-hand side of Eq 4, the portion of the field potential measurement that sums the energy demand of each neuron.
Our model fits and data support this view. When we captured the stimulus-related broadband response by simulating a change in broadband coherence across neurons rather than a change in the level of response in each neuron, our predicted BOLD response was highly inaccurate (S7 Fig).
In contrast, we propose that “low gamma” often does not predict the BOLD response because “low gamma” reflects narrowband oscillatory processes, which largely arise from a change in neuronal synchrony across the population rather than a change in the response level per neuron. This corresponds to the second term in the right-hand side of Eq 4, the portion of the field potential measurement that reflects the cross-power arising from currents in different neurons, which in our model, is independent of the signals giving rise to the BOLD signal.
Our results and model do not argue that narrowband gamma oscillations will never be predictive of the BOLD signal. If, in a particular experiment, narrowband gamma oscillations were to covary with broadband increases, we would expect both signals to correlate with BOLD. This might occur in an experiment with gratings of different contrast; with increasing contrast, narrowband gamma responses, broadband responses, and BOLD responses all increase [21, 63], and all three measures would correlate across stimuli. In such an experiment, if narrowband gamma oscillations had a higher signal-to-noise ratio than the broadband response, then the oscillatory signal would likely show a higher correlation with BOLD. In contrast, when the choice of stimulus or task can independently modulate broadband power and gamma oscillations so that the two LFP measures are not correlated, as in the experiments presented here and previously [38], then gamma oscillations will not strongly correlate with BOLD.
Our simulation and empirical results are consistent with studies that varied chromatic contrast and spatial frequency while measuring MEG and BOLD. These studies found that BOLD and narrowband gamma activity did not match in stimulus specificity [18, 19]. It is likely that these stimulus manipulations, like ours, independently modulated narrowband gamma power and broadband power, although the studies did not quantify broadband fields, which are more challenging to measure with MEG than with ECoG [64]. We speculate that broadband fields spanning the gamma range would have shown a higher correlation with BOLD.
In our model, the LFP measures are highly sensitive to neuronal synchrony, whereas BOLD is not. In our simulations, increases in neuronal synchrony drove narrowband gamma oscillations in the field potential. There are other cases of population activity with a high degree of neuronal synchrony. One example is the steady-state evoked potential associated with a periodic stimulus [65, 66]. Previous studies have described discrepancies between evoked potentials and the BOLD signal, such as in the case of spatial summation [11], directional motion selectivity [7, 8] and spatial attention [9, 10]. Our modeling framework suggests that the neural sources generating the steady-state potential (synchronous neural activity) are likely to be only weakly related to the BOLD signal (depending largely on asynchronous signals), as these sources will primarily affect the second term on the right-hand side of Eq 4 (cross-power). This does not imply that the two measures are always or even usually discrepant; the BOLD signal and steady-state potentials are likely to correlate any time that the steady state signals correlate with other measures of neural activity. When measures do dissociate, we do not conclude that one measure is more accurate; instead, the measures offer complementary views of the population activity, emphasizing the degree of synchrony or the average level of the response. An intriguing question is how each of the two signals contributes to perception and behavior [67].
Neural synchrony can also emerge without being time-locked to the stimulus, often called “induced synchrony” or “induced oscillations” [68]. In our simulation, we assumed that narrowband gamma LFP changes were induced by increases in synchrony between neurons and not by changes in the level of gamma power within the individual neurons. In contrast, we assumed that broadband LFP increases were induced by increased broadband activity in individual neurons and not by increased broadband coherence between neurons. (In Eq 4, a change in the left-hand side, LFP power in the gamma band, can be produced by a change in either the first or second term on the right). This explains why, in our simulation, the broadband power was correlated with BOLD, whereas the LFP gamma power was not, findings that were also confirmed by the data. Were our assumptions justified?
In principle, an increase in narrowband gamma power in the LFP could arise because the neurons synchronize in the gamma band or because ongoing gamma oscillations within each neuron increase in amplitude, independent of coordination between neurons. There is strong experimental support for the former. Experiments that measure both intracellular membrane potential from single neurons and the extracellular LFP show that when there is an increase in narrowband LFP gamma power, the gamma power from individual neurons becomes more coherent with the LFP [47]. Moreover, the coherence between local spiking and the LFP also increases in the gamma band when LFP gamma power increases [43]. These results are consistent with our assumption that a significant part of the increase in gamma LFP power arises from a change in population coherence. To our knowledge, it is not certain whether there is also some increase in the level of gamma signals within individual neurons when the narrowband gamma band LFP power changes. However, since we can attribute a large part of the change in gamma LFP to a change in coherence, we infer that we can only attribute, at most, a small part of the change in gamma LFP to the level of gamma power within neurons.
In our simulation, we made two simple but extreme assumptions. First, we assumed that gamma oscillations occur with no change in the total level of neural activity, and hence no change in metabolic demand or BOLD. Second, we assumed that broadband responses occur with no change in neural synchrony. While these assumptions are likely incorrect at the limit, the simulations nonetheless captured the pattern of ECoG and fMRI results obtained in our datasets. Alternative models in which the broadband response was caused by a change in synchrony were much less accurate (S7 Fig). Models in which gamma responses were caused by a change in level were only slightly less accurate and cannot be ruled out entirely (S7 Fig). However, the regression models fit to our data (Fig 9) show that the power of narrowband gamma oscillations does not predict the BOLD response. Hence the most parsimonious explanation is that these responses in the LFP are caused in large part by changes in synchrony.
Both our measurements and our simulations showed that broadband electrophysiological responses were related to, but did not fully account for, the BOLD signal. This was especially evident in Simulation 2 and extrastriate data (V2/V3). In these cases, the amplitude of low-frequency oscillations (8–15 Hz) was negatively correlated with the BOLD signal, independent of broadband signals. Numerous previous studies have reported that low-frequency oscillations are anticorrelated with BOLD, including measurements in motor, visual, and language areas [20–22, 69–71]. This result may appear to conflict with the prior discussion, since we argued that oscillations (to the degree that they reflect neuronal synchrony) should have little to no effect on metabolic demand or the BOLD signal. It is therefore important to ask why low-frequency oscillations sometimes correlate with the BOLD signal, both in data and in simulation.
One explanation is that alpha oscillations, or a mechanism that generates the oscillations, affect the BOLD signal indirectly by inhibiting cortical activity. According to this explanation, an increase in alpha power results in a decrease in local spiking activity, in turn reducing metabolic demand and the BOLD signal [72]. Alpha oscillations may indeed co-occur with reduced cortical excitation [73]. However, if this coupling between alpha power and spiking was the only explanation for the relationship between alpha power and BOLD, then a more direct measure of neuronal excitation, such as broadband or multiunit activity, would adequately predict the BOLD signal; alpha power would negatively correlate with the BOLD signal but would provide no additional predictive power. Our data and model do not support this explanation, as we find that for most cortical sites, the most accurate predictor of the BOLD signal is a combined model including both the amplitude of alpha oscillations and broadband power.
We therefore propose that in addition to the indirect effect of modulating cortical excitability, alpha oscillations are also accompanied by a mean shift in membrane potential, making it less depolarized (i.e., closer to the equilibrium potential), and this shift reduces metabolic demand. Indirect evidence for a mean shift comes from MEG and ECoG studies [49, 50, 74], which refer to alpha oscillations as being asymmetrical (i.e., they are not centered at 0—there is a mean shift). This can be explained by a simple process: if alpha oscillations reflect periodic inhibitory pulses, then on average, they will cause a hyperpolarization (or less depolarization). If the neuron was slightly depolarized before the inhibitory alpha pulses, then the pulses would push the neuron toward equilibrium and hence a lower-energy state. In this view, large alpha oscillations reflect larger inhibitory pulses, reducing depolarization. We suggest that this reduced depolarization affects metabolic demand in two ways: by reducing spiking (as discussed above) and by maintaining a less-depolarized state, reducing metabolic demand. In our model, the contribution to the BOLD signal from each neuron is the power in the time series (Eq 3), and the mean contributes to power. The idea that a mean shift in the membrane potential affects metabolic demand (in addition to altering excitability) is consistent with the observation that slowly changing currents (<0.5 Hz) correlate with BOLD fluctuations [12, 75]. Moreover, if alpha oscillations are associated with a mean shift in membrane potential, this would explain why cortical excitability depends on the phase of the alpha cycle: at one phase, the membrane potential is more depolarized, and hence cortex is more excitable, and in the opposite phase, cortex is more hyperpolarized and hence less excitable. This is consistent with the observations that the threshold for eliciting a phosphene with TMS changes with alpha phase [76, 77] and that the alpha phase at the time of stimulus presentation influences the size of the BOLD response in visual cortex [78].
Inhibition takes two neurons—one to inhibit and one to be inhibited. In our simulations, the alpha oscillations (C3) were associated with inhibitory fluctuations in the membrane potential (mean below 0), which in turn was associated with decreases in BOLD. It is important to note that these fluctuations are meant to capture the effect of local inhibition on the postsynaptic neurons (the neurons being inhibited). The inhibitory neurons themselves are presynaptic, and the action of inhibiting other neurons is presumably an active process that consumes energy. Therefore, inhibition is expected to increase energy demand in some neurons (the presynaptic neurons) and decrease energy demand in other neurons (postsynaptic neurons). We did not model the inhibitory neurons explicitly; however, the neural activity associated with active inhibition would be expected to contribute to the measured broadband signal in the ECoG data and is implicitly included in the broadband inputs in our simulations (C1). More complex models (see paragraph 3.6) in which the circuitry of excitatory and inhibitory neurons is explicitly represented (such as [63, 79, 80]) may provide insight into how the balance between excitation and inhibition influences the field potential and the BOLD signal.
We found that the relationship between the BOLD signal and features of the ECoG data differed across cortical areas. For example, broadband changes in ECoG responses explained more variance in the BOLD data in V1 than in V2/V3. Conversely, low-frequency power decreases (alpha, 8–13 Hz) explained more variance in the BOLD signal in V2/V3 than in V1. In the absence of a model, we might have interpreted the results as evidence that neurovascular coupling differs across sites. Many previous studies have reported differences in the relation between LFP and BOLD as a function of site or condition, for example, between cortical and subcortical locations [81], across cortical regions [82, 83], between cortical layers [84], and as a function of medication [85]. Here, we showed that a difference in the relationship between LFP and BOLD need not arise because of a difference in neurovascular coupling. In our results, Simulations 1 and 2, like V1 compared to extrastriate areas, showed differences in the relationship between LFP and BOLD, yet we used the identical model of neurovascular coupling in all simulations. The systematic differences in the two simulations arose because of differences in the neuronal population activity, not because of differences in neurovascular coupling. While our results do not exclude the possibility of differences in neurovascular coupling across locations or states, they do caution against interpreting differences in the relationship between field potentials and BOLD as evidence for a difference in neurovascular coupling, since they show that a single model can account for a variety of patterns. More generally, the V1 versus V2/V3 discrepancies bolster the argument that one cannot predict the exact relationship between BOLD and field potentials without also specifying the neuronal population activity.
A complete description of the biophysical processes giving rise to the BOLD signal and the field potential is far beyond the scope of this paper and is likely premature given the enormous complexity in the nervous system, the vascular system, and the coupling mechanisms between them. Instead, the purpose of our modeling framework was to first begin with a general principle, namely that BOLD and field potentials sum neural activity according to a different sequence of operations; second, to instantiate this principle in simple mathematical rules; third, to combine these rules with a minimal model of neural population activity; and finally, to ask to what extent such a model can account for the patterns in our data.
Our model omits many biophysical components, such as compartmentalized neurons, multiple cell types and vessel types, neurotransmitters, the dynamics of blood flow, and so on; hence, it is not a detailed simulation of the nervous system or vascular system. We modeled the BOLD signal as a function of dendritic currents summed across neurons within an imaging region. The logic motivating this is straightforward. Neuronal activity consumes a large amount of energy, and this energy demand is dominated by the cost of restoring the membrane potential following ionic flows from synaptic potentials and action potentials [86, 87]. As a result, the increased energy demand from neuronal responses is related to the dendritic currents. Neurovascular coupling is the process of increasing blood flow to meet this energetic demand; a failure of the hemodynamic response such as stroke can cause neuronal and glial cell death, highlighting the importance of the relationship between blood supply regulation and neuronal activity [33].
We modeled the hemodynamic response as being proportional to the energy demand from dendritic currents. This model was proposed as a computational summary of the approximate relationship between the BOLD signal and neuronal activity, not as a hypothesis about a causal mechanism. Recent work suggests that energy consumption per se (e.g., the change in the cerebral metabolic rate of oxygen consumption) is not the triggering mechanism for the increased blood flow, rather neurotransmitters and other molecules associated with synaptic events are part of a complex cascade that causes vasodilation and changes in blood flow [33, 88, 89]. The exact biological mechanism responsible for neurovascular coupling is an area of highly active, ongoing research [35]. The key assumptions in the model—that the BOLD signal is correlated with changes in membrane potential and that the order of operations differs for the BOLD signal and the LFP—makes accurate predictions for our dataset. A separate and important research question is how closely the biophysical mechanisms match this computational-level description, and what these mechanisms are.
The simplicity of the model has benefits. It facilitates an understanding derived from basic principles, similar to the advantages in building computational, rather than biophysical, models of neural responses [90–93]. Both types of models and empirical studies are valuable. Here, we emphasize that even with a highly simplified model of the BOLD signal, the field potential, and neuronal population activity, we are able to reconcile a wide range of findings in a complicated and technical literature. The model accounts for differences in how broadband field potentials and gamma oscillations relate to the BOLD signal. It can explain differences between cortical areas in the relationship between field potentials and BOLD. The model also provides an explanation for why the amplitude of alpha rhythms is negatively correlated with BOLD, even after accounting for the relationship between broadband signals and BOLD.
We note that drastic simplifications are the norm in many fields of neuroscience, such as receptive field modeling of visual neurons; most such models omit fixational eye movements, optical properties of the eye, retinal and cortical circuitry, etc., instead modeling responses as a few simple mathematical computations of the stimulus (filtering, thresholding, and normalization) [94]. These highly simplified models will certainly fail under some conditions [95], yet they have proven to be of immense value to the field [93], in part due to their simplicity and in part because the alternative in which the responses of visual neurons are computed from a complete, neurobiologically realistic model of the nervous system simply does not exist.
To test competing computational theories about the relation between the visual input, the LFP, and the BOLD response, it is essential to make sample data and code available for others [38, 53]. Following standard practices of reproducible research [96–98], the Matlab code of the simulation and sample data and code to reproduce the Figs in this manuscript can be downloaded at https://github.com/dorahermes/Paper_Hermes_2017_PLOSBiology.
To understand how the electrophysiology and BOLD responses are related, it is necessary to specify both the manner in which population activity transfers to the two signals and the neuronal population activity itself. The former shows that the covariance between neuronal time series has a large influence on the field potential and not the BOLD signal. Based on our simulations and empirical results, we made several inferences about the neuronal population responses mediating the BOLD signal and the LFP: that narrowband gamma oscillations in visual cortex likely arise more from synchronization of neural responses than a change in level of the neural response and hence have a large influence on the field potential and little influence on the BOLD signal, that responses that are asynchronous across neurons manifest in broadband field potentials and an elevated BOLD signal, and that low-frequency oscillations observed in field potentials are likely accompanied by a widespread hyperpolarization, which in turn reduces metabolic demand and the BOLD signal. Our model-based approach brings us a step closer to a general solution to the question of how neural activity relates to the BOLD signal.
Informed, written consent was obtained from all subjects. The fMRI protocols were approved by the New York University IRB and ECoG protocols were approved by the Stanford University IRB, according to the principles expressed in the Declaration of Helsinki.
Simulations were computed for a population of 200 neurons. Each simulation trial was 1 second long with millisecond sampling. The time series for each neuron was derived by summing three inputs, each 1 second long, followed by leaky integration with a time scale of 10 milliseconds to simulate temporal integration in the dendrite (Fig 4). Each simulation was fit to ECoG data from one electrode and consisted of 240 trials, 8 repeats of 30 stimulus conditions. A condition in the simulation was defined by the parameter settings for the three inputs (Fig 4): C1 (broadband), C2 (gamma), and C3 (alpha). Variations in these three inputs resulted in power changes in the broadband, gamma, and alpha LFP. The inputs were fit to data such that the simulated LFP power changes matched the ECoG data power changes for a particular electrode and stimulus.
Stimuli for ECoG experiments were reported previously [38]. In brief, for one subject, the stimuli came from 8 classes of patterns (30 exemplars per class, 20x20°), including high contrast vertical gratings (0.16, 0.33, 0.65, or 1.3 cycles per degree square wave) noise patterns (spectral power distributions of k/f4, k/f2, and k/f0), and a blank screen at mean luminance (S2 Fig). For the second ECoG subject, there were the same 8 classes as well as two other stimulus classes–a high contrast white noise pattern and a plaid at 0.65 cpd. The fMRI subjects had the same 10 stimulus classes as the second ECoG subject.
ECoG data were measured from two subjects who were implanted with subdural electrodes (2.3 mm diameter, AdTech Medical Instrument Corp) for clinical purposes at Stanford Hospital. Informed, written consent was obtained from all subjects. ECoG protocols were approved by the Stanford University IRB. In 22 electrodes in V1 V2 and V3, broadband and narrowband gamma responses were quantified as before [38], and alpha power changes were calculated.
fMRI data was measured from four subjects (three female, ages 22–42) with normal or corrected-to-normal vision at the Center for Brain Imaging at NYU. Informed, written consent was obtained from all subjects. The fMRI protocols were approved by the New York University IRB. fMRI data were preprocessed and analyzed using custom software (http://vistalab.stanford.edu/software). Disc regions of interest (ROIs) (radius = 2 mm) were defined in fMRI subjects to match the position of the electrodes in ECoG subjects using a combination of anatomy, pRF centers, and visual field maps. The similarity between the ROI position and electrode position was compared via visual inspection of anatomical images and pRF centers (S3 Fig).
The relationship between fMRI and ECoG signals was analyzed using a linear regression model. The cross-validated coefficient of determination (R2) was used as a metric for model accuracy, and the regression coefficients were used to test whether ECoG predictors (broadband, gamma, and alpha) had a positive or negative relation with BOLD.
The relationship between fMRI and ECoG signals was analyzed using a linear regression model:
y=Xb+c+ε
where y is a vector of fMRI amplitudes (beta estimates), with n entries for the n different stimuli; X is a matrix of ECoG responses, n by 1, 2, or 3, where the columns correspond to one or more of broadband, gamma, and alpha estimates; b are the 1, 2, or 3 beta weights for the broadband, gamma, and alpha estimates; c is a constant (the y-intercept); and ε is the residual error term. The model was fitted separately for each cortical site (electrode/ROI pair) and for different combinations of predictors—broadband alone, gamma alone, alpha alone, each pairwise combination, and all three predictors together.
The n stimuli in the regressions included the contrast patterns and the blank stimulus. Inclusion of the blank stimulus is important for capturing the sign of the mean response. For example, if all contrast patterns induced a BOLD response of a particular level (say, +1) and induced ECoG responses of a particular level (say, –1) and we did not include the blank stimulus in the regression, then after subtracting the mean from each measure, all beta estimates would be approximately 0. This would mask a systematic relationship between ECoG and BOLD measures (in this example, an anticorrelation) arising from viewing stimuli with contrast compared to viewing a blank screen.
Models were evaluated by split-half cross-validation. First, the regression model yi = Xibi + ci + ε was fit using half of the fMRI subjects (1 and 2) and half of the ECoG stimulus repetitions (even repetitions). To cross-validate this model, the beta values (bi) were then applied to the left out half of the ECoG data (odd stimulus repetitions) to predict the left out half of the fMRI data (fMRI subjects 3 and 4). The same procedure was applied by reversing the training and testing data. This resulted in two testing datasets with BOLD responses predicted from ECoG for each stimulus condition (Xibi + ci) and an actual measured BOLD value. For each cortical site, the coefficient of determination (see below) was calculated between the concatenated predictions and BOLD data values of the two test sets. All R2 values reported in the results are cross-validated in this manner. The same pattern of results was achieved if instead of cross-validation, we solved the models on the complete datasets and computed the R2 adjusted for the number of regressors.
To test whether different ECoG predictors (broadband, narrowband, alpha) had a positive or negative relation with BOLD, we tested whether the regression coefficient was significantly larger or smaller than 0. The regression coefficient was considered to be significantly different from 0 using a bootstrap statistic: for each model, the median of the beta values across sites was calculated after resampling 10,000 times. If <2.5% of the resampled statistics were smaller than zero, the beta values were considered significantly positive, and similarly, if <2.5% of the resampled statistics were greater than 0, the beta values were considered significantly negative.
All model predictions were quantified using the coefficient of determination on cross-validated predictions. For predicting BOLD data from simulations of population neuronal activity (Fig 7, S7 Fig), the predicted BOLD has arbitrary units. In these cases, the observed BOLD and the predicted BOLD were both normalized by subtracting the mean and then dividing by the vector length. When predicting BOLD responses from features of the LFP data (broadband, gamma, and alpha) by regression, the predicted BOLD data were in the same units as the measured BOLD, and no normalizing or rescaling was done.
To quantify the accuracy of the models, we calculated the cross-validated coefficient of determination, R2:
R2=1–SSresidualsSSdata
SSresiduals=∑i(yi−fi)2
SSdata=∑i(yi−y¯)2
where y are the data values and f are the prediction values. Because the model fits are cross-validated, it is possible for the model errors (residuals) to be larger than the data values, hence R2 can be lower than 0, and spans (−∞,1]. In the case in which the model predictions and the data are unrelated and each are normally distributed with equal variance, R2 will tend to –1.
|
10.1371/journal.pmed.1002585 | The association of lifetime alcohol use with mortality and cancer risk in older adults: A cohort study | While current research is largely consistent as to the harms of heavy drinking in terms of both cancer incidence and mortality, there are disparate messages regarding the safety of light-moderate alcohol consumption, which may confuse public health messages. We aimed to evaluate the association between average lifetime alcohol intakes and risk of both cancer incidence and mortality.
We report a population-based cohort study using data from 99,654 adults (68.7% female), aged 55–74 years, participating in the U.S. Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. Cox proportional hazards models assessed the risk of overall and cause-specific mortality, cancer incidence (excluding nonmelanoma skin cancer), and combined risk of cancer and death across categories of self-reported average lifetime alcohol intakes, with adjustment for potential confounders. During 836,740 person-years of follow-up (median 8.9 years), 9,599 deaths and 12,763 primary cancers occurred. Positive linear associations were observed between lifetime alcohol consumption and cancer-related mortality and total cancer incidence. J-shaped associations were observed between average lifetime alcohol consumption and overall mortality, cardiovascular-related mortality, and combined risk of death or cancer. In comparison to lifetime light alcohol drinkers (1–3 drinks per week), lifetime never or infrequent drinkers (<1 drink/week), as well as heavy (2–<3 drinks/day) and very heavy drinkers (3+ drinks/day) had increased overall mortality and combined risk of cancer or death. Corresponding hazard ratios (HRs) and 95% confidence intervals (CIs) for combined risk of cancer or death, respectively, were 1.09 (1.01–1.13) for never drinkers, 1.08 (1.03–1.13) for infrequent drinkers, 1.10 (1.02–1.18) for heavy drinkers, and 1.21 (1.13–1.30) for very heavy drinkers. This analysis is limited to older adults, and residual confounding by socioeconomic factors is possible.
The study supports a J-shaped association between alcohol and mortality in older adults, which remains after adjustment for cancer risk. The results indicate that intakes below 1 drink per day were associated with the lowest risk of death.
NCT00339495 (ClinicalTrials.gov).
| The detrimental health impacts of heavy alcohol intakes are well known, and even light-moderate alcohol intakes have been linked to increased risks of cancer.
However, light-moderate drinking has also been suggested to be protective for cardiovascular disease, which has led to contradictory public health messages. We conducted this study to further understand the complex relationship between light-moderate alcohol consumption, cancer, and mortality.
We analysed whether risk of cancer or death differed in individuals with different lifetime alcohol intakes, using data from approximately 100,000 individuals in the United States.
The results suggested that risk of some cancers increased with each additional alcoholic drink per week consumed.
However, combined risk of cancer or death was lowest in light drinkers consuming less than 1 drink per day, rather than drinkers with higher intakes.
These results could help inform future US guidelines, which currently recommend less than 2 drinks per day for men and less than 1 drink per day for women.
This evidence should not be taken to support a protective effect of light drinking.
| Alcohol consumption appears to have a complex and somewhat controversial relationship to health [1–4]. The “J-shaped” relationship between alcohol intake and mortality, particularly from cardiovascular disease, observed in various cohort studies has been well cited within both scientific and mainstream publications [5–8]. However, concerns over the quality of the evidence supporting the J-shaped association [1,2], combined with alcohol’s role as a risk factor for a range of cancers [9,10], has led to conflicting messages surrounding the health implications of light to moderate alcohol consumption.
Criticisms over the quality of evidence supporting a J-shaped association with mortality often surround the methods used to assess and categorise alcohol intakes [2,11]. Many of the previous cohort studies assessing the role of alcohol in mortality and cancer have focussed on alcohol intakes measured at baseline, with former drinkers excluded from current drinker categories [11]. However, this approach may lead to an underestimation of the negative health effects of drinking alcohol, as former drinkers may have stopped drinking due to adverse health effects or health scares related to their drinking habits [1,2]. Measurement of alcohol intakes over an individual’s lifetime and classification according to average lifetime alcohol intakes helps to avoid the bias that occurs when separating former drinkers from current drinkers and gives a more comprehensive assessment of an individual’s drinking habits.
To date, few studies [5,6,12] have evaluated average lifetime alcohol consumption and mortality [13], which could help clarify the association. No studies of lifetime alcohol intake, to our knowledge, have examined the impact on mortality and cancer risk simultaneously, to allow a direct comparison of alcohol’s association with both important health outcomes.
A more comprehensive overview of alcohol and risk of cancer and death outcomes would also help to provide clearer guidelines on alcohol consumption. The 2015–2020 U.S. Dietary Guidelines for Americans currently recommend less than 1 drink per day in women and less than 2 drinks per day in men [14].
In this study, we aimed to simultaneously examine the association between average lifetime alcohol intake and mortality as well as cancer risk in a large cohort with prospective follow-up data. We also conducted a novel analysis of combined risk of cancer incidence or death, to account for differences in timing and absolute risk of these outcomes.
The Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial is a randomised trial designed to evaluate the impact of screening modalities on cancer mortality, as described previously [15]. Briefly, 154,952 individuals, aged 55–74 years, were recruited via 10 centres in the US (Birmingham, Alabama; Boulder, Colorado; Detroit, Michigan; Honolulu, Hawaii; Los Angeles, California; Minneapolis, Minnesota; Pittsburgh, Pennsylvania; Salt Lake City, Utah; St. Louis, Missouri; Marshfield, Wisconsin; Washington, DC) between 1993 and 2001. All participants provided written informed consent and the study was approved by the Institutional Review Boards at the National Cancer Institute and the 10 recruitment centres. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (S1 Checklist).
At study entry, participants self-completed a baseline questionnaire, which enquired information on demographic variables, including age, gender, race/ethnicity, marital status, educational attainment, personal and family medical history, tobacco smoking habits (including duration and frequency), medication use, and anthropometry.
Alcohol consumption was assessed using the validated PLCO Diet History Questionnaire (DHQ) [16], which was introduced into the trial in December 1998 and was completed by approximately 77% of all participants in both arms of the trial. Participants recruited after 1998 completed the DHQ at baseline for the nonscreening arm and approximately 3 years after baseline for the screening arm, whereas previously recruited participants were invited to complete the DHQ in 1999 or 2000.
The DHQ assessed historical drinking by inquiring about the amount of beer (12 ounce bottles or cans: 1 U.S. Department of Agriculture My Pyramid cup equivalent [17]), wine (5 ounce glasses: 1 cup equivalent), and liquor (1.5 ounce shots, including mixed drinks: 1 cup equivalent) the participant consumed per week between the ages of 18 and 24, 25 and 39, 40 and 54 years old, and at 55+ years, respectively. The DHQ also inquired about the amount and frequency the participant drank beer, wine (or wine coolers), and liquor (including liquor in mixed drinks) over the previous summer, as well as the rest of the previous year, to account for seasonal differences. The frequency was converted into days and an average daily alcohol intake of each type of alcohol was calculated as follows (using midpoints where necessary):
(amount*frequencyinsummer+(amount*frequencyfortherestoftheyear*3))/4
Average lifetime alcohol intake was calculated as a weighted average by multiplying the daily alcohol intake by the number of years in each age category and adding the categories together (e.g., intakes at 18–24 accounted for 7 years and intakes in the year prior to DHQ accounted for 1 year). Never drinkers were defined as having reported no alcohol consumption at any age. For some analyses, categories of drinking (infrequent drinkers [>0–<1 drinks/week], light drinkers [1–<3 drinks/week], somewhat light drinkers [3–<5 drinks/week], light-moderate drinkers [5–<7 drinks/week], moderate drinkers [1–<2 drinks/day], heavy drinkers [2–<3 drinks/day], very heavy drinkers [3+ drinks/day]) were chosen to be narrow enough to identify differences between groups and to allow relative comparability to previous studies [5], whilst being easily interpreted without limiting statistical power. The light drinker (1–<3 drinks/week) category was chosen as the reference category for mortality and combined outcomes, as these individuals were hypothesised to have the lowest mortality. The never drinker category was chosen as the reference category for risk of cancer, as it was hypothesised that these individuals would have the lowest cancer risk.
Vital status during follow-up was ascertained by annual mailed questionnaires and periodic linkage to the National Death Index. We used International Classification of Diseases, 9th Revision (ICD-9), codes to classify the underlying cause of death specified on the death certificate. Causes of death, in order, were defined as follows: cardiovascular disease, including heart disease (ICD-9 codes 390–398, 402, 404, and 410–429) and stroke (ICD-9 codes 430–434 and 436–438); cancer (ICD-9 codes 140–239).
Cancer diagnoses were ascertained through assessment of medical records from follow-up investigations following screening by trained personnel, through annual questionnaires, and linkage to the National Death Index. All cancers were pathologically confirmed through medical record abstraction by trained personnel. Total cancer includes incident cancers diagnosed after completion of DHQ but prior to death, excluding nonmelanoma skin cancer. Alcohol-related cancers include breast cancer, colorectal cancer, head and neck cancer, liver cancer, and esophageal cancer [10,18].
The primary outcomes of interest were overall mortality (death from any cause), mortality from cardiovascular disease and cancer, as well as total cancer risk (incidence of any cancer, excluding nonmelanoma skin cancer) and a novel analysis, which classed either incidence of any cancer (excluding nonmelanoma skin cancer) or death from any cause as the event of interest. Deaths from homicides, suicide or accidents, and other causes of death were reported as secondary outcomes. Follow-up time was calculated from the date of DHQ completion to the first of either death, 13 years after randomisation or 31 December 2009, or date of first incidence of any cancer (excluding nonmelanoma skin cancer) in analyses including cancer as an outcome.
A total of 154,952 individuals agreed to participate in the trial. Individuals who died or were diagnosed with cancer prior to or on the date of DHQ were excluded (n = 10,096). Individuals who did not complete the baseline questionnaire (n = 4,920) or the DHQ (n = 33,245), did not have valid DHQ (n = 5,221), or had missing information for alcohol or key confounders (n = 1,761) were excluded. After exclusions, 99,654 individuals were eligible for analysis, of whom 9,599 died from any cause and 12,763 were diagnosed with any cancer (excluding nonmelanoma skin cancer). A STROBE flowchart of participant selection is included in Fig 1.
Chi-squared tests (categorical variables) and ANOVA tests (continuous variables) were used to assess the association between baseline characteristics and average lifetime alcohol intakes. Cox proportional hazards models were used to estimate the hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) for the association between average lifetime alcohol intake and each outcome (all-cause mortality, cardiovascular-related mortality, cancer-related mortality, and cancer incidence, including total and alcohol-related cancers), in men and women separately. Age was used as the underlying timescale in all analyses [19]. Random effects meta-analyses using the inverse variance as the weight were used to provide a combined estimate across men and women for each level of intake.
Adjusted models included race/ethnicity (non-Hispanic white, non-Hispanic black, Asian, Hispanic/Other), study centre, randomisation group (control group or screening group), body mass index (BMI), year of DHQ completion, smoking status by pack-years (never smokers, former smoker reporting ≤25 pack-years, former smoker reporting ≥25 pack-years, current reporting ≤25 pack-years, current reporting ≥25 pack years, cigar/pipe only smokers), marital status, education, family history of cancer (cancer outcomes only), intakes of total energy (kcal/day), calcium, red meat, processed meat, fiber, coffee, fruit and vegetables, and menopausal hormone replacement therapy use (HRT, in women only). No model selection for confounders was done, as recommended by the reviewer; all variables were included. At the reviewer’s suggestion, BMI, intakes of total energy, calcium, red meat, processed meat, fiber, coffee, fruit and vegetables were included as continuous variables instead of categorical variables.
Restricted cubic spline models [20] were fitted with varying number of knots to assess the dose-response trend in the association between average lifetime alcohol intake (as a continuous variable) and each outcome after full adjustment. Never drinkers were classified as the reference category for all restricted cubic spline analyses due to methodological constraints. Akaike's information criterion [21] and likelihood ratio tests were used to investigate if there was a significant improvement to the model fit when fitting restricted cubic spline models, with varying number of knots, compared to a linear model with average lifetime alcohol intake as a continuous variable. If the linear model provided the best fit for an outcome, Cox proportional hazards models were conducted, including average lifetime alcohol intakes as a continuous variable.
A priori stratified analyses were conducted by age (<65 versus 65+), randomisation group (control versus screening arm), BMI (>18.5–<25 versus >25 kg/m2), and smoking status (ever versus never) among the entire cohort (without gender stratification to ensure adequate statistical power) using restricted cubic splines for total mortality and total cancer risk. Secondary analyses reported results for primary outcomes and secondary outcomes from Cox proportional hazards models using alternative reference categories to those used in main analyses: never drinkers for mortality outcomes and light drinkers for cancer incidence outcomes. Further outcomes evaluated included mortality from accidents/injuries/homicide/suicide and mortality from any other cause, as well as risk of alcohol-related cancers, other cancers, and individual types of cancer (including all alcohol-related cancers and PLCO cancers). Post hoc sensitivity analyses based on suggestions from reviewers also assessed overall mortality in current and former drinkers relative to never drinkers and the association between lifetime alcohol consumption with and death from any cause within cancer patients.
Restricted cubic splines were also used to assess the association between different types of alcohol (beer, wine, and liquor) and overall mortality and cancer incidence (accounting for intakes of other types of alcohol). A 2-tailed p-value < 0.05 was considered significant and analyses were conducted using Stata/SE statistical software (version 14.0, College Station, TX, USA). A copy of the research proposal is included in the supplementary material (S1 Text).
Our analysis included 99,654 individuals, including 51,306 women (68.7%) and 48,348 men (31.3%). The median follow-up time was 8.9 years, with 9,599 deaths and 12,763 primary cancers recorded over a total 836,740 person-years of follow-up. Drinking patterns differed between men and women, with the proportion of women compared to men higher amongst never and infrequent drinkers and lower amongst moderate, heavy, and very heavy drinkers (Table 1). The average median lifetime alcohol intake was 1.78 drinks per week (0.25 per day), with men reporting higher intakes (4.02 drinks/week) than women (0.80 drinks per week).
In men, heavier drinkers were less likely to be married or have completed a university degree; more likely to be smokers or obese; and tended to have lower intakes of fruits and vegetables, dietary fibre, and total calcium. Similar patterns were apparent in women, although never and infrequent drinkers tended to be more similar to very heavy drinkers (3+ drinks per day) than light-moderate drinkers in terms of university education and obesity. HRT use was least common amongst never drinkers (Table 1).
Analyses indicated that average lifetime alcohol intakes were linearly associated with total risk of cancer (Fig 2D) and risk of ‘alcohol-related’ cancers (Fig 2E). Light-moderate drinkers (HR 1.09, 95% CI 1.00–1.20) and heavy drinkers (HR 1.11, 95% CI 1.00–1.24) were at an increased risk of cancer overall (when combining men and women) when compared to never drinkers (Fig 4). When average lifetime alcohol was assessed as a continuous variable, each additional drink per day was associated with a small increase in total risk of cancer in men (HR 1.01, 95% CI 1.00–1.02). No major differences were observed by age, smoking status, BMI, or randomisation group (S3 Fig), but differences by type of alcohol were observed (S4 Fig). Risk of ‘alcohol-related’ cancer was 66% higher in very heavy drinkers than in never drinkers in men (HR 1.66, 95% CI 1.16–2.38). Each additional drink per day was associated with an increased risk of ‘alcohol-related’ cancers in men (HR 1.06, 95% CI 1.04–1.08). Analyses by type of cancer revealed very heavy drinking was associated with increased risk of head and neck cancers (HR 3.12, 95% CI 1.54–6.34), liver cancer (HR 3.53, 95% CI 1.26–9.87), and esophageal cancer (HR 3.99, 95% CI 1.47–10.82) (S3 Table).
A J-shaped relationship was apparent between average lifetime alcohol intakes and combined risk of cancer or death, with the lowest risk observed at intakes between 1 and 5 drinks per week (Fig 2F, S1 Fig). Never, infrequent, moderate, and very heavy drinkers had a higher combined risk of cancer or death than light drinkers (1–<3 drinks/week) overall (men and women combined) with HRs of 1.07 (95% CI 1.01–1.13) for never drinkers, 1.08 (95% CI 1.03–1.13) for infrequent drinkers, 1.10 (95% CI 1.02–1.18) for heavy drinkers, and 1.21 (95% CI 1.13–1.30) for very heavy drinkers. Results were similar for women alone, whereas in men the differences were only statistically significant for heavy (HR 1.09, 95% CI 1.01–1.18) and very heavy drinkers (HR 1.19, 95% CI 1.10–1.28) (Fig 5).
In this analysis, we observed nonlinear, “J-shaped” associations between average lifetime alcohol consumption and overall and cardiovascular-related mortality. Light lifetime alcohol consumption was associated with reduced overall and cardiovascular-related mortality compared to never drinking. Higher average lifetime alcohol consumption was also linearly associated with increased cancer-related mortality and cancer incidence. Combining the risk of cancer and death into a single analysis attenuated the modest reduction in risk observed for light drinkers, suggesting only a small benefit in risk for light drinkers compared to never drinkers.
Whilst existing research is largely consistent as to the harms of heavy drinking in terms of both cancer incidence and overall mortality, there are disparate messages regarding the safety of light to moderate alcohol consumption. Some argue that there is ‘no safe limit of alcohol’, largely on the basis of increased cancer incidence [22,23], and others emphasise the potential benefits for reducing cardiovascular mortality [3,24]. The results for lifetime alcohol consumption when mortality and cancer incidence were assessed separately in the current study were similarly disparate. The analysis of combined risk of death and cancer incidence aimed to account for these disparate associations. The results indicate that the J-shaped association and risk reductions observed for light alcohol intakes remained, albeit slightly attenuated, with small reductions in combined risk at light intakes (between 1 and <5 drinks per week). This study adds to the limited evidence base for women [13] and indicates the lowest combined risk of death and cancer incidence is at lighter intakes in both men and women. These results could help inform future US guidelines, which currently recommend less than 2 drinks per day for men and less than 1 drink per day for women [14]. The United Kingdom recently lowered their guidelines for men to recommend less than 14 units (about 1 drink) per week, several alcohol-free days per week, and limiting the total amount of alcohol in one session [23]. Mendelian randomisation studies indicating beneficial effects of alleles associated with reduced alcohol intakes on cardiovascular outcomes are consistent with our findings and could also inform these debates [25]. Further studies incorporating other major sources of morbidity, such as dementia [4,26], may better inform public health guidelines on the health impact of light drinking.
This large cohort study had a number of strengths and limitations. Previous studies reporting a J-shaped association between lifetime alcohol consumption and mortality have been criticised for measuring alcohol consumption at a single time point and separating ‘unhealthy’ former drinkers from current drinkers [11,27]. By examining the use of lifetime alcohol consumption, we were able to limit this potential bias. The present study is also, to our knowledge, the first study with prospective follow-up to assess the association between lifetime alcohol consumption and overall cancer risk.
Other studies have been criticised for having broader categories of consumption and advocating a reduction in mortality at implausibly low intakes [11]. In our study, we included a separate category for infrequent lifetime drinkers (>0–<1) to examine infrequent drinking more closely. The increased mortality in infrequent drinkers compared to light drinkers in our study suggests that these groups are indeed different and could explain why a systematic review found no reduction in mortality amongst ‘low volume drinkers’, which includes both infrequent and moderate drinkers (up to 2 drinks per day) [11]. The narrower classification of alcohol intakes used in the current study may have limited the statistical power to assess associations by cancer type.
The current study also had a larger proportion of never drinkers than a previous analysis of lifetime alcohol consumption from Europe [5]. This allowed for a more powerful assessment of the relative mortality between never drinkers and light drinkers. However, as an indirect consequence, the assessment of heavy drinking in women was limited by a smaller number of women reporting average lifetime alcohol consumption over 2 drinks per day in the current study.
Unlike many previous analyses, which focused on baseline or current consumption [24], this study assessed the risk associated with average lifetime alcohol consumption from the age of 18. However, persons were not enrolled in the study until after the age of 55 years and after DHQ completion. Thus, the study reflects mortality in older adults and the impact of alcohol consumption on mortality at younger ages could not be assessed, which could have been impacted by deaths from accidents, violence, or suicide [11,27], in particular. However, similar associations between alcohol and overall mortality have been seen previously in younger adults [24], which may alleviate concerns of an impact due to selection bias.
It is also possible that a certain degree of misclassification of alcohol consumption occurred due to social desirability bias or recall issues, particularly amongst individuals with infrequent alcohol consumption [28]. In addition, an assessment of episodic, binge drinking was not possible with the data collected in this study, although binge drinking is unlikely to exaggerate reductions in risk within light drinkers. We also did not have information on individuals who did not complete the DHQ, which also raises the potential for selection bias.
The role of socioeconomic status in health is also a potential source of confounding, as high socioeconomic status is associated with light-moderate alcohol intakes and better mortality outcomes [29,30]. Whilst we adjusted for educational attainment, smoking status, and various dietary factors that may be reflective of socioeconomic status, residual confounding is possible. Similarly, we were unable to adjust for physical activity, which could also be a source of residual confounding. Consistent with other screening trials, participants in the PLCO trial are more highly educated and less likely to smoke than the general population [31], suggesting that they are more homogenous than the US population at large. Thus, whilst the generalizability of findings may be weaker, the impact of such residual confounding is likely to be small. Finally, we were unable to evaluate cardiovascular disease incidence in this study, which would complement our analysis.
The study supports a J-shaped association between alcohol and mortality in older adults, which remains after adjustment for cancer risk. The results indicate that intakes between 1 and <5 drinks per week were associated with the lowest combined risk of cancer or death. This study provides further insight into the complex relationship between alcohol consumption, cancer incidence, and disease mortality and may help inform public health guidelines.
|
10.1371/journal.ppat.1004507 | The Toll-Dorsal Pathway Is Required for Resistance to Viral Oral Infection in Drosophila | Pathogen entry route can have a strong impact on the result of microbial infections in different hosts, including insects. Drosophila melanogaster has been a successful model system to study the immune response to systemic viral infection. Here we investigate the role of the Toll pathway in resistance to oral viral infection in D. melanogaster. We show that several Toll pathway components, including Spätzle, Toll, Pelle and the NF-kB-like transcription factor Dorsal, are required to resist oral infection with Drosophila C virus. Furthermore, in the fat body Dorsal is translocated from the cytoplasm to the nucleus and a Toll pathway target gene reporter is upregulated in response to Drosophila C Virus infection. This pathway also mediates resistance to several other RNA viruses (Cricket paralysis virus, Flock House virus, and Nora virus). Compared with control, viral titres are highly increased in Toll pathway mutants. The role of the Toll pathway in resistance to viruses in D. melanogaster is restricted to oral infection since we do not observe a phenotype associated with systemic infection. We also show that Wolbachia and other Drosophila-associated microbiota do not interact with the Toll pathway-mediated resistance to oral infection. We therefore identify the Toll pathway as a new general inducible pathway that mediates strong resistance to viruses with a route-specific role. These results contribute to a better understanding of viral oral infection resistance in insects, which is particularly relevant in the context of transmission of arboviruses by insect vectors.
| Pathogenic microbes can enter their hosts through different routes. This can have a strong impact on which host defensive mechanisms are elicited and in disease outcome. We used the model organism Drosophila melanogaster to understand how resistance to viruses differs between infection by direct virus entry into the body cavity and infection through feeding on food with the virus. We show that the Toll pathway is required to resist oral infection with different RNA viruses. On the other hand this pathway does not influence the outcome of viral infection performed by injection. Together our results show that the Toll pathway has a route-specific general antiviral effect. Our work expands the role of this classical innate immunity pathway and contributes to a better understanding of viral oral infection resistance in insects. This is particularly relevant because insect vectors of emerging human viral diseases, like dengue, are infected through feeding on contaminated hosts.
| Pathogens can infect their hosts through many different routes. In humans, for instance, microbes can directly enter the host through skin lesions or mediated by insect vectors. However, most of human infections start at mucosal surfaces of the respiratory, digestive or genital tracts. Pathogens specialize in different transmission strategies involving different host tissues. On the other hand, hosts mount distinct immune responses in different tissues, involving specialized cells and structures. Therefore, pathogen entry route can have a strong impact on the result of infection in animals, from humans to insects [1]–[4].
In Drosophila melanogaster oral or systemic infection with bacteria trigger different responses and have different outcomes (see [5] for review). Injection of bacteria into the haemocoel induces a systemic immune response based on the secretion of proteins into the haemolymph by the fat body [6]–[9]. Oral infections prompt a local immune response in the gut, and in some cases also a systemic response [10]–[13]. In both these responses the immune deficiency (Imd) signaling pathway can be activated and many antimicrobial peptides are produced [13]. However, these responses differ in other activated pathways and induced genes [13], [14]. Notably, the Toll pathway, a major mediator of systemic immune responses, is not involved in the gut local response. Injection of bacteria is generally more pathogenic than oral infection, with lower titres of bacteria being required for a lethal effect [2], [15]. Interestingly, the bacteria Serratia marcescens administered through oral infection can cross the gut barrier and enter the haemolymph, however these systemic bacteria have a lower pathogenicity than corresponding titres directly injected [15]. These findings indicate that natural infections lead to more structured and effective immune responses. These functional differences are also reflected in evolutionary processes since Drosophila adaptation to pathogenic bacteria is dependent on infection route [2].
Viral infections in insects have also been show to differ with infection route. For example, honeybees infected by Deformed Wing virus (DWV) through vertical transmission or horizontal oral transmission have no apparent disease symptoms. However, if horizontally transmitted by the parasitic Varroa mite, presumably from the mite saliva to the bee haemocoel, DWV is highly pathogenic [16]–[18]. Understanding the common and unique characteristics of insect defence against viral pathogens delivered through different routes is important in order to explain this differential pathogenicity. Moreover, resistance to viral oral infection in insects is also of particular interest since vectors of arboviruses are mainly infected through feeding on contaminated hosts.
Drosophila melanogaster has become an important model organism to study innate antiviral immunity in insects [19]–[21]. Some Drosophila viruses are vertically transmitted (e.g. Sigma virus) [22] and others can be infective by feeding, such as Nora virus [23], [24] and Drosophila C virus (DCV) [25]–[29]. ERK has recently been shown to be involved in resistance to RNA viruses by oral infection [29]. However, most of D. melanogaster antiviral immunity research has been done on systemic infection with viruses. The best characterized antiviral mechanism in Drosophila is the RNA interference (RNAi) pathway that has a strong influence on infection by a wide range of viruses, including RNA and DNA viruses [30]–[35]. Consistent with the important role of RNAi, several viruses express suppressors of this mechanism [30], [32], [36]. Other important mediators of antiviral protection are the intracellular bacteria Wolbachia [37], [38]. Presence of these endosymbionts increases resistance to several RNA viruses [37]–[40].
The role of classical Drosophila inducible immune pathways in antiviral defence seems less broad or well defined. The JAK/STAT pathway is required for resistance to DCV and Cricket Paralysis virus (CrPV) but not to other viruses [35], [41]. Similarly, mutants in the Imd pathway are less resistant to Sindbis virus and CrPV [42], [43] but not to DCV [44]. The role of the Toll pathway in antiviral immunity is less clear. This pathway is initiated by the binding of the cytokine Spätzle to Toll which triggers an intracellular signalling cascade involving the adaptor proteins dMyD88 and Tube and the kinase Pelle, and leads to activation of the NF-κB transcription factors Dorsal and Dorsal-related immunity factor (Dif) [45]–[47]. These transcription factors are normally sequestered in the cytoplasm and translocate to the nucleus upon Toll pathway activation. No phenotype was observed with DCV or Sindbis in Dif mutants or Dif and dorsal double mutants, respectively [43], [44]. However, Dif mutants are more susceptible to Drosophila X virus (DXV) [48]. On the other hand, the role of the whole pathway in resistance to DXV is not clear since loss-of-function mutants in Toll (Tl), spätzle (spz), tube (tub) and pelle (pll) show no phenotype [48]. Moreover, constitutive activation of the pathway, in a Toll gain-of-function mutant, also leads to higher susceptibility to DXV [48].
Data in other insects support an antiviral role for the Toll pathway. In honeybees dorsal-1A knockdown increases titres of DWV [49]. Also, in the mosquito Aedes aegypti the Toll pathway is induced upon ingestion of a dengue virus infected blood meal and inactivation of the pathway resulted in increased viral loads [50]. These studies raise the possibility that the Toll pathway is generally involved in the response to viruses in insects and prompt further analysis of its function in Drosophila antiviral immunity.
Here we investigate the role of the Toll pathway in immune response to several RNA viruses on Drosophila melanogaster comparing a natural infection route (i.e. by feeding) and systemic infection. We show that several Toll pathway components, including the extracellular cytokine Spätzle, the membrane receptor Toll, the kinase Pelle and the NF-kB-like transcription factor Dorsal, are required to resist natural viral infections in Drosophila but not systemic infection. These data provide evidence that the inducible Toll pathway has a route-specific general antiviral effect.
DCV is a non-enveloped virus with a single-stranded, positive-sense RNA genome that belongs to the Dicistroviridae family [51]. This virus is a natural pathogen of D. melanogaster that can be found in both wild and laboratory fly populations [22]. On most Drosophila studies using DCV the virus is injected directly into the body cavity, bypassing putative natural barriers and immune defences. In order to infect Drosophila flies with DCV through a natural route, we developed a protocol for oral DCV infection in adults. The protocol consisted in keeping adult flies with a mix of DCV and yeast for 24 hours in a vial. After this period, defined as 0 days post-infection (dpi), flies were transferred to vials containing standard Drosophila food and their survival scored daily. We found that DCV oral infection in adult DrosDel w1118 isogenic (hereafter called w1118 iso) [52] flies can cause a lethal infection in both females and males, killing up to 25% of flies in 20 days (Fig. 1A, 1B and Dataset S1). We observed that flies started to die 5 to 6 dpi, similarly to infection by injection or pricking. We fitted the survival data with a Cox proportional hazard mixed effect model and compared the relative risk of dying of infected flies with non-infected controls (mock). In order to compare the different doses with each other we performed a Tukey's test on the resulting Cox hazard ratios. Lethality is dose-dependent since we observed that higher DCV doses induce significantly different higher lethality rates (Fig. 1A, 1B, S1 and Dataset S1).
We observed that both females and males that become lethargic and inflated die within one day (Fig. 1C). In order to identify the reason of the observed overinflated body, particularly the abdomen, we dissected these flies at 5 dpi. Moribund flies exhibit an oversized crop when compared with healthy flies (Fig. 1D, 1E and S2). Using immunofluorescence we detected DCV infecting crop-associated muscle cells (Fig. 1E), suggesting that viral infection of this visceral muscle is the reason of the crop oversize.
To further characterize the course and the tropism of DCV upon oral infection we investigated which tissues were infected at 0 dpi (immediately after the 24 hours DCV exposure), 2 dpi and 5 dpi. We analysed oesophagus, crop, proventriculus, midgut, Malpighian tubules, hindgut, male and female reproductive organs, haemocytes, fat body, trachea and thorax skeletal muscle. At 0 dpi we were able to detect virus particles in the lumen of the midgut (Fig. 2A), indicating that the virus is reaching at least as far as the midgut. However, we were not able to detect any DCV infected cell, including epithelial and visceral muscle cells of the midgut (Fig. 2A). At 2 dpi the only tissue in which we could detect infection was the fat body (Fig. 2B). This DCV infection was confined to some regions of the fat body, mostly in the abdominal region. At 5 dpi DCV was also detected in the fat body (Fig. S3A) and the extent of the infection there was much greater than at 2 dpi. Analysis of flies 5 dpi also revealed the presence of DCV in the visceral muscle surrounding the crop, midgut and hindgut (Fig. 1E, 2C and 2D). Despite the presence of DCV in the midgut visceral muscle, we were not able to detect any virus in the gut epithelium. We also detected DCV in the muscle surrounding the Malpighian tubules at the junction point with the gut, but we were not able to detect DCV in the Malpighian tubules cells (Fig. S3B). The visceral muscle cells of the ovarian and testis peritoneal sheaths were also infected with DCV (Fig. 2E and F). We detected DCV in the abdominal muscle rarely (less than 1/40 flies) (Fig. S3C) but we never found DCV in thorax skeletal muscle (Fig. S3D). DCV was also detected in small sections of the tracheal system, mostly frequently in the abdominal region (Fig. S3E). Additionally, we observed that DCV was present in some circulating haemocytes (Fig. 2G). This could indicate that DCV efficiently infects haemocyte cells. Another possible explanation is that haemocytes phagocytose infected cells. Overall, these immunofluorescence results show that by oral infection, DCV infects specific tissues of D. melanogaster.
In order to compare DCV tropism upon oral and systemic infections we examined all the above tissues in 20 males 2 and 5 dpi for both infection protocols using immunofluorescence. For this we pricked flies with a relatively low dose of DCV (105 TCID50/ml). Analysis of orally infected flies confirmed that at 2 dpi only the fat body was infected with DCV (Fig. 3A and Table 1). There was a restriction to the fat body in some systemically infected flies at 2 dpi, however in other flies the infection was also present in other tissues (Fig. 3B and Table 1). At 5 dpi DCV was detected in the same tissues for both virus-delivery methods (Fig. 3C, 3D and Table 1). These results show that independently of delivery route DCV tropism is largely the same, although it is less restricted to the fat body in the early stages of systemic infection.
In order to analyse the role of the Toll pathway in the response to viral oral infection we tested a collection of mutants in different genes of the Toll pathway: spz (spz4/spz4), Tl (Tlrv1/Tlr3), pll (pll2/pll21), dorsal (dl1/dl1) and Dif (Dif1/Dif1). To limit putative effects of different genetic backgrounds each mutation was introgressed into the w1118 iso background. This introgression was done by chromosome replacement and backcrossing (see Materials and Methods). We orally infected these mutant lines with DCV and their survival was compared to the control line w1118 iso. Flies mutant in the genes spz, Tl and pll were more susceptible to DCV oral infection than w1118 iso control flies (Fig. 4A, 4B, S4A, S4B and Datasets S2, S3, S4). For pll mutants we further show increased sensitivity, compared with w1118 iso, at several DCV infection doses and dose-dependent lethality (Fig. S5A–D and Dataset S5). Mutations in the genes encoding the two NF-κB homologues known to be downstream of the Toll pathway give different results. Dif mutants do not show a phenotype in this assay and are as sensitive to DCV infection as the w1118 iso control flies while dl mutants show high susceptibility, to the same degree as spz and pll mutants (Fig. 4A and S4A). These results contrast with the requirement of Dif but not dl in adult flies to resist bacteria and fungi [46], [53]–[56]. The high lethality observed for the several mutations in genes of the Toll pathway when compared to the control background are a consequence of DCV infection, since in the absence of viral infection and in the timeframe of this analysis these mutations have no effect on survival, except for the dl mutant that seems to be slightly deleterious by itself (Fig. 4C and S4C). In summary, these data show that the Toll pathway is important to survive DCV infection and Dorsal, but not Dif, is the downstream transcription factor required.
To investigate whether the increased lethality rates of pll-deficient flies was due to decreased resistance or tolerance to DCV we analysed the viral levels by Western blot at 1, 3 and 5 dpi. We observed that at 3 and 5 dpi pll mutant flies had more viral protein than w1118 iso flies (Fig. 4D and S6). We confirmed these results by measuring DCV RNA levels of single flies at 2, 5, 10 and 20 dpi using reverse transcription quantitative PCR (RT-qPCR). A greater number of pll mutant flies exhibited high quantities of DCV RNA when compared with w1118 iso flies at 2 and 5 dpi (Fig. 4E, S5E and Dataset S6). DCV titres are significantly different between these lines at these days and the median of viral RNA load was approximately one thousand to ten thousand times higher in pll mutant flies (Fig. 4E and S5E). All the flies analysed are infected with DCV, even the ones that survive the infection for 20 days. This shows that there is no clearance of the virus in this timeframe. These results show that a mutant in the Toll pathway has lower resistance to DCV upon oral infection.
In order to investigate if the Toll pathway is also required to resist DCV systemic infection we pricked Toll pathway mutants and w1118 iso flies with DCV and followed their survival for 20 days. We found that Toll pathway mutant lines were not more susceptible to DCV systemic infection when compared with w1118 iso flies (Fig. 4F, S7A–C and Datasets S7, S8). We further analysed the pll mutant infected at several doses in order to rule out a dose-specific lack of effect. pll mutant was not more sensitive than w1118 iso to DCV systemic infection in any of the doses (Fig. 4G and Dataset S9). These results suggest that, contrary to oral infection, the Toll pathway is not important in the immune response to DCV systemic infection.
To explore whether Toll pathway mutant flies have altered patterns of infection, we analysed DCV tropism at 2 and 5 dpi in pll mutant flies after oral infection. As before, 20 males were analysed per time point. In this mutant we can detect DCV in a higher proportion of flies at 2 dpi and 5 dpi than w1118 iso (40% compared with 15 or 25%, Fig. 3A, 3C, 4H, 4I and Table 1), in agreement with the RT-qPCR data. In contrast to w1118 iso flies (Fig. 3A), at 2 dpi DCV is not restricted to the fat body and can also be detected in muscle surrounding the crop and the midgut (Fig. 4H). At 5 dpi pll mutant flies showed the same DCV tropism as the observed in w1118 iso (Fig. 3C and 4I). We were also unable to detect DCV in crop or midgut epithelial cells in pll mutants. These results show that although DCV seems to be spreading faster in pll mutant flies than in w1118 iso, overall there is no difference in DCV tropism.
Wolbachia induces resistance to infection by RNA viruses in D. melanogaster and other insects [37], [38], [57]. In mosquitoes this protection has been suggested to be dependent on the Toll pathway [58]. To test this hypothesis in D. melanogaster we compared the survival of w1118 iso and pll-deficient flies, infected and non-infected with Wolbachia. The results show that in D. melanogaster Wolbachia also protects against viral oral infection (Fig. 5A, S8A–C and Datasets S10, S11, Cox Proportional Hazards Model, p<0.001). We observed that Wolbachia protects pll mutants against DCV infection to the same extent as w1118 iso flies (Fig. 5A and S8A–C). The pll mutation does reduce survival of Wolbachia-positive flies when orally infected with DCV but to the same extent as in Wolbachia-free flies and there is no interaction between the two factors (Cox Proportional Hazards Model, Wolbachia*Genotype interaction; p = 0.67). The same lack of interaction is observed for systemic DCV infection (p = 0.69). Therefore in D. melanogaster the Toll pathway is not absolutely required for Wolbachia protection to DCV, confirming previous data with dengue virus [59].
Other Drosophila-associated microbiota could also indirectly affect DCV oral infection and Toll pathway mediated protection. The Toll pathway could, for instance, be important to control a secondary infection with bacteria upon viral infection induced damaged. w1118 iso and pll-deficient flies were raised and maintained with antibiotics and susceptibility to DCV was compared with conventionally-reared flies (Fig. 5B and C, S8D–F and Dataset S12). There is no significant effect of the antibiotic treatment on the susceptibility to viruses (Cox Proportional Hazards Model; p = 0.28) and pll-deficient flies are still more susceptible to DCV infection in the absence of bacteria (Cox Proportional Hazards Model; p<0.001). Hence, in our experimental setup the Drosophila-associated microbiota does not play a role in the susceptibility to DCV oral infection or Toll-mediated resistance.
To test the specificity of the Toll pathway in D. melanogaster antiviral immune response, we tested its requirement for resistance to other insect RNA viruses. Cricket Paralysis virus (CrPV) is closely related to DCV, also belongs to the Dicistroviridae family, and causes a lethal infection in adult flies [32], [42], [60]. Upon CrPV oral infection we observed that pll-deficient flies were more susceptible than control flies (Fig. 6A, S9A and Dataset S13). As with DCV oral infections, we found that a greater number of pll mutant flies exhibited higher amounts of CrPV RNA when compared with w1118 iso flies (Fig. 6B and Dataset S14). pll-deficient flies showed the same susceptibility to CrPV systemic infection as control flies at different viral infection titres (Fig. 6C, S9B and Dataset S15).
We also tested whether pll-deficient flies are more sensitive to Nora virus oral infections. Nora virus is a picorna-like, non-enveloped virus, with a positive-sense single-stranded RNA genome [23]. This virus naturally infects D. melanogaster and causes persistent infection without any evidence of pathology [23], [24]. Nevertheless, we compared the lethality rates of pll-deficient flies with w1118 iso control flies upon Nora virus oral infection. As show in Fig. 6D and S9C (Dataset S16), we did not observe any lethality associated with Nora oral infection, even in pll mutants. However, when we measured Nora RNA levels of single flies 5 dpi we found that a greater number of pll mutant flies exhibited high amounts of viral RNA when compared with w1118 iso flies (Fig. 6E and Dataset S14).
Finally, we investigated the importance of Toll pathway in the immune response to Flock House virus (FHV). FHV is a non-enveloped, positive-sense RNA virus that belongs to the Nodaviridae family of insect virus [61]. Although FHV is not a natural pathogen of D. melanogaster it can replicate and cause lethality in adults when injected [31]. pll-deficient flies were more susceptible to FHV oral infection than w1118 iso control flies (Fig. 6F, S9D and Dataset S17) and had higher levels of viral RNA (Fig. 6G and Dataset S14). We also tested whether pll mutant flies were more susceptible to FHV systemic infection. In concordance with the DCV results, pll mutant flies were not more susceptible to FHV systemic infection when compared with w1118 iso flies across several doses of infection (Fig. 6H, S9E and Dataset S18).
These analyses with different viruses indicate that the Toll pathway is required to resist a broad range of RNA viruses. Moreover, this requirement seems to be specific to oral infection and not relevant in the context of a systemic infection. To establish whether the increased sensitivity to viral infection extended to other pathogens, we orally infected pll-deficient flies with Pseudomonas entomophila and compared their survival to w1118 iso. pll-mutant flies were not more susceptible to these Gram-negative bacteria than w1118 iso (Fig. S10A and Dataset S19). This was expected since the Toll pathway is not required for the transcriptional immune response to Gram-negative bacteria gut infection [13]. We also analysed the feeding rates of pll mutant and w1118 iso flies. When exposed to DCV mixed with yeast both lines had the same feeding rate (Fig. S10B and Dataset S20). These data show that Toll pathway mutant flies are not generally more susceptible to oral infection by all pathogens. Testing further pathogens will allow assessing if increased susceptibility of Toll pathway mutants is restricted to viruses.
Since we observed that flies mutant in genes of the Toll-Dorsal pathway have increased sensitivity to DCV oral infection, we investigated whether Dorsal is activated during viral infection. We probed if Dorsal was translocated from the cytoplasm to the nucleus upon DCV infection by using an antibody specific against its C-terminal domain [62] (Fig. S11A and B). At 5 days after oral infection, but not 2 dpi, we were able to detect nuclear import of Dorsal in fat body cells infected with DCV (20 flies were analysed in each time point) (Fig. 7A). This is only observed in infected cells although many fat body cells infected with DCV do not show Dorsal nuclear localization (Fig. 7B). However, we never detected Dorsal enrichment in the nuclei of non-infected fat body cells, even in infected flies (Fig. 7C and S11A). This nuclear translocation upon DCV infection seems specific to the fat body since we do not observe it other tissues, including gut epithelial and muscle cells (in the same 2 dpi and 5 dpi samples) (Fig. S11C and S11D). We have also failed to detect Dorsal translocation in haemocytes of w1118 iso flies orally or systemically infected (the w1118 iso infected flies analysed in Table 1) (Fig. S11E). We also analysed Dorsal localization after DCV systemic infection (2 dpi and 5 dpi, 10 flies each). Dorsal is translocated to the nuclei of fat body cells in response to DCV infection at 2 dpi (Fig. 7D) but not in epithelial or muscle cells of the gut in the same flies. Finally, we tested Dorsal translocation upon oral viral infection in a pll mutant line (Fig. 7E). We do not see Dorsal translocation in 16 DCV infected flies that are pll−/− but we see translocation in 4 out of 14 infected w1118 iso control flies (chi-square test, p = 0.037). This shows that Dorsal translocation in response to viral infection is dependent on the Toll pathway. In summary, these results show that Dorsal is translocated from the cytoplasm to the nucleus in fat body cells in response to DCV infection, suggesting that the Toll pathway is involved in an antiviral inducible immune response.
Drosomycin (Drs) encodes an antimicrobial peptide and is a target gene of immune activation of the Toll pathway [7]. We probed expression of a Drs reporter gene [47] in response to viral infection. We observed Drs-GFP expression in the fat body of 8 out of 8 DCV infected flies but not in gut muscle or epithelium (Fig. 7F and G and S11F and G). Out of 8 non-infected flies none showed activation of Drs-GFP expression (Fig. 7H). The Drs-GFP fat body expression is present in infected and non-infected cells in DCV-infected flies, unlike Dorsal translocation, indicating a systemic activation of Toll pathway. This result further shows that the Toll pathway is activated in the fat body upon viral infection.
In order to test if inactivation of the Toll pathway in the fat body or other tissues (muscle, visceral muscle, enterocytes and haemocytes) would increase sensitivity to viruses we expressed three RNAi constructs for pll with different drivers and compared survival after DCV oral infection with control (Fig. S12 and Dataset S21). We failed to see any significant increase in lethality upon viral infection in these lines. Based on this negative result with RNAi it is not possible to conclude on the need of the Toll pathway anti-viral response in specific tissues.
In Drosophila the Toll pathway plays a fundamental role in the response to systemic infection by fungi and bacteria [7], [54], [63]. Here we show that this pathway is also required to resist oral viral infections. Mutants in genes that encode the ligand Spätzle, the Toll transmembrane protein receptor, the cytoplasmic kinase Pelle and the NF-κB transcription factor Dorsal succumb faster to DCV infection and have higher titres of this virus. This demonstrates that a functional canonical Toll pathway is required for flies to survive a natural viral infection.
Two very similar NF-κB homologues, Dif and Dorsal, can be downstream of the Toll pathway in flies. Our genetic analysis shows that Dorsal but not Dif is required for viral resistance. In contrast, Dif but not Dorsal is required for adult systemic response to infection by bacteria and fungi [7], [46], [53], [63] and has been regarded as the Toll pathway transcription factor involved in adult immune systemic response. Nonetheless, other data also indicate a role for Dorsal in the immune response. In larvae both Dif and Dorsal are translocated into fat body cell nuclei upon systemic infection with bacteria [6], [45], [64] and Dorsal is upregulated in infected larvae [65]. In larvae Dif and Dorsal may be redundant in resistance to bacteria with the double mutant being very susceptible to normal Drosophila-associated microbiota [66]. Dif and Dorsal can form homo and heterodimers and these recognize different DNA sequences and differentially activate target genes [67], [68]. However, overexpression of one or the other transcription factor many times is sufficient to rescue mutant or double mutant phenotypes [66], [69] as well as activate expression of antimicrobial peptides [69]. Overall, the exact role and contribution of either transcription factor in the several types of immune responses of Drosophila is not known. On the other hand Dorsal also has a clear role in development in early embryogenesis, which Dif does not seem to share [45], [70]. The Toll pathway, although not necessarily through Dorsal, also has a role in muscular and neuromuscular development [71]–[75] and hematopoiesis [76]. This raises the possibility that the phenotypes we observed are due to development problems. This hypothesis would be particularly relevant for oral infection with viruses if development problems were to lead to alterations in feeding. However, we observe no differences in feeding rates between pll mutants and control. Moreover, we show that pll mutants are not more susceptible to oral infection with a bacterial pathogen. These data indicate that it is not a major digestive system development problem that leads to lower resistance to oral viral infection. We cannot absolutely rule out a development problem; however, we detect Dorsal translocation into the nuclei of DCV infected fat body cells and expression of a Drosomycin reporter gene in the fat body of infected flies. This shows that the Toll-Dorsal pathway is induced upon viral infection and, together with the genetic data, strongly supports a Toll pathway mediated anti-viral response. Identification of the target genes of the Dorsal transcription factor after viral infection will be important in the future, as well as understanding how they contribute to resistance to viruses.
During embryonic development and systemic immune response to fungi and bacteria the extracellular pro-Spätzle is proteolytically cleaved, leading to binding to the Toll receptor and activation of the pathway. In the case of infection, specific pattern recognition receptors present in the haemolymph are activated by microbial ligands and start a proteolytic cascade that culminates in pro-Spätzle cleavage [55], [56], [77]. Fungal and bacterial proteases can also lead to Spätzle cleavage through a different proteolytic cascade involving Persephone [77], [78]. At this point it is unclear how activation of the Toll pathway by viral infection works and it probably differs significantly from activation by bacteria and fungi. Putative pathogen associated molecular patterns associated with viruses and recognized by Drosophila must be different from the cell wall components of bacteria and fungi involved in Toll pathway activation. Moreover, viruses are intracellular parasites while the previously studied microbial elicitors of the Toll pathway are extracellular and present in the haemolymph. Previous work has shown that in Drosophila the Toll pathway also responds to tumours [79] and to a block in apoptosis, via Persephone [80], while in mosquitoes it can be activated by reactive oxygen species [58]. Viral infection could be indirectly detected by the Toll pathway through recognition of tissue damage and share a mechanism of activation with the above situations. Drs expression in response to viral infection is widespread in the fat body of infected flies and not restricted to infected cells. This indicates that Spätzle activation is systemic upon viral infection and that the Toll pathway is generally activated in the fat body of these flies. This is in agreement with previous published data showing up-regulation of Drs and Toll pathway genes upon DCV infection [28], [35]. As a further layer of complexity, our results show that Dorsal translocation is restricted to viral infected cells and is not observed throughout the fat body. This is at odds with a systemic activation of Spätzle and how the Toll pathway responds to bacteria and fungi. It is possible that Dorsal is activated throughout the fat body but that is not visible in the translocation assay. However, Dorsal activation and translocation to the nucleus may depend on Toll activation and a second cell-autonomous signal. In mammals RIG-I-like receptors (RLRs) and NOD-like receptors are involved in cell-autonomous activation of innate immunity in response to viral infection. There are no homologues of these cytoplasmic pattern recognition receptors in Drosophila. However, Dcr2 has a helicase domain homologous to helicase domains in RLRs and has been suggested to act as a pattern recognition receptor in Drosophila [81]. Toll-like receptors in mammals are also able to detect viral infection through binding to nucleic acids in vesicular compartments. Toll-7 in Drosophila can bind vesicular stomatitis viruses and induce antiviral autophagy [82]. Unravelling the signal that leads to Dorsal translocation in virally infected cells will be important to understand antiviral immunity in Drosophila.
Our results show that the increased lethality rates observed in the Toll pathway deficient flies are associated with higher DCV loads. Thus, the Toll pathway is involved in resistance to viruses. Furthermore, we demonstrate in this study that Toll requirement to control viral loads is not specific to DCV and extends to other RNA viruses, such as FHV, CrPV and Nora virus. Previous work did not see an effect of a pll mutant in a Nora virus infection assay [83]. The difference in our results may be due to different control of the genetic background or differences in the assay. We analysed the response to a new Nora virus oral infection while Habayeb and colleagues analysed the capacity to clear the viruses in a chronically infected Drosophila stock [83]. The median increase in viral titres we observe in pll mutants can be up to ten thousand fold. The magnitude of the difference is comparable or higher to differences between wild type flies and RNAi mutants [30]–[33] and between flies with and without Wolbachia [37]. The strength and generality of the interaction between the Toll pathway and viruses indicates that this is a major antiviral pathway in Drosophila. This is consistent with previous studies showing Toll pathway antiviral effect in mosquitoes and honeybees [49], [50].
The increased sensitivity to viruses in Toll pathway mutants is only manifested upon oral infection and not systemic infection. This is not a result of different infection titres with the two modes of infection because Toll pathway mutants are not more sensitive to a low dose of virus by systemic infection. Therefore, we have identified a pathway with a route-specific role. Nonetheless, we observe Dorsal nuclear translocation in fat body cells after both routes of infection. This indicates that the pathway is activated regardless of type of infection but it is only effective in a scenario of oral infection. In order to understand the differential requirement of the Toll pathway we performed a detailed analysis of the dynamics of DCV oral and systemic infections. Overall we found no major differences in the tissue distribution of DCV between the two infection routes. In both DCV is present in the fat body, trachea and visceral muscle of the crop, midgut and hindgut, and gonads. Although we can detect DCV particles in the midgut lumen shortly after oral ingestion, we could not determine its point of entry. We were unable to detect DCV infection in the epithelium of the digestive system at any time point. This could indicate that the DCV is transported across gut epithelial cells to the body cavity (haemocoel) without infecting the epithelial cells themselves. Transcytosis of virions has been described in mammals and insects [84]–[86]. An alternative explanation would be that DCV rapidly kills infected epithelial cells, therefore hindering their detection. Apoptosis of midgut cells following viral infection has been observed in Drosophila and in mosquitoes [87], [88]. However, a recent study in Drosophila reported that upon oral ingestion DCV was able to infect midgut epithelial cells [29]. The difference between these results may reflect differences in the feeding protocol: Xu and colleagues continually exposed flies to DCV for several days [29], while we only infect flies for one day. In our setup the fat body seems to be the first tissue to be infected; all infected flies have DCV in the fat body and some infected flies only have DCV in the fat body. This is more evident in orally infected flies that at 2 dpi only have DCV in this tissue. This may reflect a difference in the dynamics of the two infection routes and in systemic infected flies DCV seems to disseminate faster. The detection of Dorsal translocation only in fat body cells and the probable early restriction of DCV to this tissue when delivered by oral infection may be part of the explanation of the differential requirement of the Toll pathway in the two routes of infection.
Our results show that the Toll pathway is required to resist viral infections, which adds to the previously known requirement of the Toll pathway to resist bacteria, fungi, and parasitoids. This contributes to the idea that Spätzle may work more as a cytokine involved in general response to infection than to specific pathogens [5]. This Toll antiviral resistance is dependent on Dorsal and not Dif and we show Dorsal activation in virus-infected cells. The specificity of the immune response to difference pathogens may therefore rely on which transcription factors are activated downstream of the Toll pathway. Finally, we show that Toll requirement is restricted to viral oral infection and therefore route specific. This demonstrates that the interaction of viruses with Drosophila varies with mode of infection. Oral infection with viruses may be subject to more layers of control since it is probably the most frequent route of infection. Understanding this complexity is particularly relevant because arboviruses are transmitted to arthropod vectors of human diseases through feeding.
Flies were maintained on standard cornmeal diet at a constant temperature of 25°C unless otherwise stated. All fly lines were cleaned of possible chronic viral infections as described elsewhere [22], [37]. Briefly, flies were aged to 30 days at 25°C and their eggs were collected in agar plates, treated with 50% bleach for 10 min, washed with water, and transferred to fresh vials.
Fly lines used in this study were free of Wolbachia except if otherwise stated. To mark midgut epithelial we used flies carrying the driver Myo1A-Gal4 (expressed in the enterocytes [89]) combined with UAS-GFP. We have analysed the following homozygous or heterozygous combination of mutants in the Toll pathway: spz4/spz4 (spz4 is a loss of function allele) [90], Tlrv1/Tlr3 (Tlrv1 is a loss of function allele and Tlr3 is a hypomorphic allele) [91], pll2/pll21 (pll2 is loss of function allele and pll21 is a hypomorphic) [90], [92], dl1/dl1(dl1 is a loss of function allele) [93], Dif1/Dif1 (Dif1 is a loss of function allele) [53]. To reduce genetic background effects these mutations were isogenized to the DrosDel w1118 isogenic background [52]. For each line the non-mutated chromosomes were replaced using balancer chromosomes whereas the mutation was recombined to the respective DrosDel w1118 isogenic chromosome for seven generations. We confirmed that the isogenized lines retained the mutation of interest by the associated development phenotype (lethality or maternal effect) or by DNA sequencing in the cases of absence of phenotype. For Drosomycin expression we used y w drs-GFP dpt-LacZ flies. For tissue specific pll knockdown the following drivers were used: C7-Gal4 (fat body driver [94]), 24B-Gal4 (visceral muscle driver [95]), Myo1A-Gal4 (midgut epithelium [89], [96]), mef2-Gal4 (somatic, visceral and cardiac muscle [97]) and hml(delta)-Gal4 (haemocyte driver [98]). Tlr3 (#3238) and dl1 (#3236) were obtained from the Bloomington stock center (http://flystocks.bio.indiana.edu/). Three independent UAS-pll-IR constructs and control UAS-mCherry-IR flies from TRiP collection [99] were used y1 sc* v1; P{TRiP.HMS01213}attP2 (#34733), y1 sc* v1; P{TRiP.GL00150}attP2 (#35577), y1 sc* v1; P{TRiP.HMS02332}attP40 (#41935), y1 sc* v1; P{VALIUM20-mCherry}attP2 (#35785). MyoIA-Gal4 was kindly given by Nicolas Tapon, spz4 and y w drs-GFP dpt-LacZ by Bruno Lemaitre, Tlrv1 by Kathryn Anderson, pll2 and pll21 by Steven Wasserman and Dif1 by Dominique Ferrandon.
DCV was produced either in cell culture or in flies. Cell culture DCV production and titration were performed as described in [37]. DCV production in flies was done in w1118 iso flies that were clear from viruses and Wolbachia infection [37], [100]. Flies were afterwards orally infected with DCV, which led to the establishment of a chronically infected stock. This stock was kept for at least five generations before extracting DCV from it. Because DCV infected stocks show a high lethality rate at pupal stage, we perform DCV extraction from pupae. We squashed 50 g of pupae in 50 ml of 50 mM Tris-HCl, pH 7.5. The extract was frozen at −80°C, thawed and centrifuged twice for 20 min at 27000 g at 4°C, keeping the supernatant. The supernatant was aliquoted and stored at −80°C and later titrated in Schneider's Line 2 (SL-2) cells as described in [37]. FHV and CrPV was produced and titrated in Schneider Drosophila line 2 (DL2) as in [37] and in [101], respectively, with minor changes. Nora virus extract was prepared from a naturally infected Oregon R stock [37]. One hundred adult flies were squashed in 1 ml of 50 mM Tris-HCl, pH 7.5. Extract was then frozen at −80°C, thawed and twice centrifuged for 10 min at 20000 g, at 4°C. The supernatant was aliquoted and stored at −80°C.
Infections were performed on 3–6 days-old flies. To perform oral infection with virus we used empty plastic vials with 1×3 cm pieces of filter paper (Whatman gel blotting papper GB003) placed in the bottom. We loaded on the filter paper 350 µl of a mix of 75% virus extract and 25% of yeast (Saccharomyces cerevisiae, Sigma-Aldrich). Ten flies were placed per vial and left feeding for 24 hours at 25°C. For mock oral infections flies were exposed to buffer (50 mM Tris-HCl) mixed with yeast (25%). After this infection period we transferred the flies to new vials containing standard cornmeal diet. For viral systemic infections CO2 anesthetized flies were pricked in the thorax. The 0.15 mm diameter needles used for infection (Austerlitz Insect Pins) were dipped into a virus solution diluted to the desired concentration in 50 mM Tris-HCl, pH 7.5. After systemic infections flies were transferred to vials containing standard cornmeal food, 10 flies per vial. After both protocols of infection flies were kept at 25°C, checked for survival daily and vials changed every 5 days.
Pseudomonas entomophila was grown in LB at 30°C overnight. P. entomophila cultures were then concentrated by centrifugation and adjusted to OD600 = 75. For oral infections with P. entomophila flies were exposed to a 1∶1 solution of bacteria culture and 5% sucrose in water. In control mock infections, flies were exposed to LB with 5% sucrose. Survival was followed every 12 hours for 3 days. Micrococcus luteus was grown in LB at 37°C overnight, concentrated by centrifugation and adjusted to OD600 = 3. For systemic infections with M. luteus flies were pricked in the thorax with fine needles dipped in bacterial suspension. The P. entomophila and M. luteus strains used in this study were kindly provided by Bruno Lemaitre and Thomas Rival, respectively.
Flies were dissected to expose the internal tissues, fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) for 15 min, washed in PBS, then incubated with 1% Triton-X-100 and 5% FBS in PBS (PTX-FBS) for 30 min. Samples were then incubated overnight with primary antibody at 4°C. Rabbit polyclonal antibodies raised against purified DCV (kindly given by Peter Christian) was used at 1∶200 dilution. Dorsal antibody developed by Ruth Steward was obtained from the Developmental Studies Hybridoma Bank, created by the NICHD of the NIH and maintained at The University of Iowa, Department of Biology, Iowa City, IA 52242 [62], was used at 1∶5 dilution. The samples were washed with PTX-FBS, and then incubated in PTX-FBS with secondary antibodies conjugated with Alexa Fluor 488 or Alexa Fluor 568 (both by Molecular Probes) for 1 h. Samples were then washed with PTX-FBS, and incubated with Alexa Fluor 594 Phalloidin and DAPI or TOTO-3 (all by Molecular Probes) for 15 min. The samples were then washed in PTX-FBS, dissected and mounted in Vectashield Mounting Medium for microscopy. Confocal images were taken with either a Leica SP5 or Zeiss LSM 510 META confocal microscopes and processed in Fiji [102].
3–6 day old males of each line were orally infected with DCV (1011 TCID50/ml), collected 1, 3 and 5 days later for protein extraction, and probed in a Western blot with anti-DCV antibody. Ten males were pooled per sample. Rabbit polyclonal antibodies raised against purified DCV was kindly given by Dr. Peter Christian. E7 mouse monoclonal anti-β-tubulin was acquired from Developmental Studies Hybridoma Bank [103].
For each sample RNA was extracted from one male fly using the Zymo Research Direct-zol RNA MiniPrep kit according to manufacturer's instructions. RNA concentrations were determined using NanoDrop ND-1000 Spectrophotometer. cDNA was prepared from 1 µg of total RNA using Random Primers and M-MLV Reverse Transcriptase (both Promega). Primers were allowed to bind to the template RNA for 5 min at 70°C and the reaction proceeded at 25°C for 10 min, 37°C for 60 min and 80°C for 10 min.
Each cDNA sample was analyzed in triplicate using a 7900HT Fast Real-Time PCR System (Applied Biosystems) instrument. We performed each reaction in a 384-well plate (Applied Biosystems), using 7 µl of iQ SYBR Green supermix (Bio Rad), 0,5 µl of each primer solution at 3,6 µM and 5 µl of diluted cDNA. Viral amplification was performed using the following thermal cycling protocol: initial 50°C for 2 min, denaturation for 10 min at 95°C followed by 40 cycles of 30 s at 95°C, 1 min at 56°C and 30 s s at 72°C. Melting curves were analysed to confirm specificity of amplified products. We obtained Ct values for manual threshold of 10 using the program SDS 2.4. Relative amounts of viral RNA were calculated by the Pfaffl Method [104] using Drosophila Rpl32 as a reference gene. The following primers were used: DCV forward 5′- TCATCGGTATGCACATTGCT-3′; DCV reverse 5′-CGCATAACCATGCTCTTCTG-3′; FHV forward 5′- ACCTCGATGGCAGGGTTT-3′; FHV reverse 5′- CTTGAACCATGGCCTTTTG-3′; CrPV forward 5′-ACGAGGAAGCAACTCAAGGA-3′; CrPV reverse 5′-GAGCCCGCTGAGATGTAAAG-3′; Nora forward 5′-TTTCACTTTACTGTTGGTCTCC-3′; Nora reverse 5′-ATTCCATTTGTGACTGATTTTATTTC-3′; Rpl32 forward 5′- CCGCTTCAAGGGACAGTATC-3′; Rpl32 reverse 5′-CAATCTCCTTGCGCTTCTTG-3′.
Flies w1118 iso and pll−/− were raised for one generation in food with a mix of antibiotics (100 µg/ml of streptomycin, 200 µg/ml of rifampicin and 100 µg/ml of tetracycline) [79], [105] and progeny was used to test susceptibility to virus. Flies were maintained in antibiotic food until the end of survival analysis. Elimination of bacteria was confirmed by plating homogenates of pll−/− flies that died during the time-course of infection. For each condition, a pool of 3 dead flies was homogenized with a pestle in 100 µl of LB. The homogenized extract was plated with the help of a 10 µl inoculation loop in Lactobacilli MRS broth and Mannitol culture media, which are able to grow Lactobacillus and Acetobacter, respectively [106]. The plates were incubated for 4 days at 25°C and subsequently scanned.
All statistical analyses were done using R (2.10.1) [107].
To compare survival rates we used a Cox's proportional hazards mixed effect model (coxme in R). Fixed effects include sex, viral dose, genotype, presence/absence of Wolbachia, antibiotic treatment, and repeat of the experiment. To account for variation between vials of the same line in the same experiment, replicate vials were considered as a random effect. This method accounts for variation between vials of the same line in the same experiment and variation between replicates of the experiment.
To assess the significance of the different fixed factors and their interactions, we performed stepwise backward model selection, and compared the difference in the log-likelihood of the different models with a χ2 distribution, with the appropriate degrees of freedom.
To compare the different doses or the different genotypes with each other we performed either pairwise comparisons between all levels of the factors (Tukey-like contrasts [108]) or contrasted the genotype of interest with the respective genetic backgrounds, averaging for the effect of the remaining factors. When the interaction between factors was statistically significant, the factor of interest was compared independently for the different levels of the interacting variable. When needed, and in order to obtain independent estimates of the hazard ratios (e.g. between different genotypes, with and without infection), we calculated the hazard ratios in models which included the interaction term, despite the interaction being non-significant. Multiple comparisons were performed using the “multcomp” (function glht) and “lsmeans” (function lsmeans) packages in R.
In order to compare viral loads between genotypes we used a Wilcoxon rank sum test (wilcox.test in R).
To analyze the feeding rates we used a generalized linear model (GLM) with a binomial response, with the proportion of fed versus unfed flies as a dependent variable and genotype and time as fixed factors.
The p-value of the chi-square test (chisq.test in R) was computed for a Monte Carlo test with 109 replicates.
|
10.1371/journal.ppat.1004514 | PUL21a-Cyclin A2 Interaction is Required to Protect Human Cytomegalovirus-Infected Cells from the Deleterious Consequences of Mitotic Entry | Entry into mitosis is accompanied by dramatic changes in cellular architecture, metabolism and gene expression. Many viruses have evolved cell cycle arrest strategies to prevent mitotic entry, presumably to ensure sustained, uninterrupted viral replication. Here we show for human cytomegalovirus (HCMV) what happens if the viral cell cycle arrest mechanism is disabled and cells engaged in viral replication enter into unscheduled mitosis. We made use of an HCMV mutant that, due to a defective Cyclin A2 binding motif in its UL21a gene product (pUL21a), has lost its ability to down-regulate Cyclin A2 and, therefore, to arrest cells at the G1/S transition. Cyclin A2 up-regulation in infected cells not only triggered the onset of cellular DNA synthesis, but also promoted the accumulation and nuclear translocation of Cyclin B1-CDK1, premature chromatin condensation and mitotic entry. The infected cells were able to enter metaphase as shown by nuclear lamina disassembly and, often irregular, metaphase spindle formation. However, anaphase onset was blocked by the still intact anaphase promoting complex/cyclosome (APC/C) inhibitory function of pUL21a. Remarkably, the essential viral IE2, but not the related chromosome-associated IE1 protein, disappeared upon mitotic entry, suggesting an inherent instability of IE2 under mitotic conditions. Viral DNA synthesis was impaired in mitosis, as demonstrated by the abnormal morphology and strongly reduced BrdU incorporation rates of viral replication compartments. The prolonged metaphase arrest in infected cells coincided with precocious sister chromatid separation and progressive fragmentation of the chromosomal material. We conclude that the Cyclin A2-binding function of pUL21a contributes to the maintenance of a cell cycle state conducive for the completion of the HCMV replication cycle. Unscheduled mitotic entry during the course of the HCMV replication has fatal consequences, leading to abortive infection and cell death.
| Cyclin A2 is a key regulator of the cell division cycle. Interactors of Cyclin A2 typically contain short sequence elements (RXL/Cy motifs) that bind with high affinity to a hydrophobic patch in the Cyclin A2 protein. Two types of RXL/Cy-containing factors are known: i) cyclin-dependent kinase (CDK) substrates, which are processed by the CDK subunit that complexes to Cyclin A2, and ii) CDK inhibitors, which stably associate to Cyclin A2-CDK due to the lack of CDK phosphorylation sites. Human cytomegalovirus (HCMV) has evolved a novel type of RXL/Cy-containing protein. Its UL21a gene product, a small and highly unstable protein, binds to Cyclin A2 via an RXL/Cy motif in its N-terminus, leading to efficient degradation of Cyclin A2 by the proteasome. Here, we show that this mechanism is not only essential for viral inhibition of cellular DNA synthesis, but also to prevent entry of infected cells into mitosis. Unscheduled mitotic entry is followed by aberrant spindle formation, metaphase arrest, precocious separation of sister chromatids, chromosomal fragmentation and cell death. Viral DNA replication and expression of the essential viral IE2 protein are abrogated in mitosis. Thus, pUL21a-Cyclin A2 interaction protects HCMV from a collapse of viral and cellular functions in mitosis.
| HCMV (also referred to as human herpesvirus-5, HHV5) is widely distributed in the human population. Acute HCMV infection can cause severe complications in immunocompromised individuals, like neonates, transplant recipients and AIDS patients. Persistent HCMV infection has been implicated as a contributing factor in the complex etiology of chronic disorders like inflammatory bowel disease, atherosclerosis and cancer [1], [2]. Treatment of HCMV is limited by severe side effects of available virostatics and by the emergence of resistant strains [3].
At the cellular level, HCMV can establish either a latent or lytic, productive infection, depending on the cell type and differentiation status. Lytic HCMV infection is accompanied by dramatic changes in host cell physiology, which are induced by the virus to promote its replication and dissemination. To this end HCMV has evolved an arsenal of regulatory factors that interact with central control mechanisms of the host cell. Besides metabolic pathways [4], cell death programs [5], intrinsic and innate immune responses [6], one of the main targets of HCMV is the cell division cycle [7], whose proper function is essential for the maintenance of genomic stability and cell growth control.
Progression through the somatic cell cycle relies on the periodic activation of cyclin-dependent kinases (CDKs) [8]. Fundamental to this periodicity is the temporal and spatial regulation of cyclin proteins, which are required for both CDK activation and substrate recognition. In brief, mitogen-dependent induction of Cyclin D1–CDK4/6 activity in the early stages of the cell cycle (G1-phase) leads to phosphorylation of the retinoblastoma protein (pRb) family of transcription factors and hence to a de-repression of growth-promoting, pRb–E2F-controlled genes, including those encoding E, A and B-type cyclins. Increasing levels of Cyclin E1–CDK2 activity in late G1 trigger S phase entry by further up-regulation of pRb–E2F-dependent gene expression and by promoting the process of replication licensing [9], [10]. In S phase, Cyclin E1 is marked for proteolysis by SCF-dependent ubiquitination [11], [12], whilst Cyclin A2 and B1 proteins become stabilized by inactivation of the APC/C ubiquitin ligase [13]. Cyclin A2–CDK1/2 activity catalyzes the initiation and regular progression of cellular DNA synthesis as well as entry into mitosis [14]–[18]. The importance of Cyclin A2 for the G2/M transition is in large part due to orchestrating regulatory processes controlling the major mitosis-promoting factor, Cyclin B1–CDK1 [19]–[24]. This complex is held inactive until G2/M transition by inhibitory phosphorylation of CDK1 and cytoplasmic localization of Cyclin B1 [23], [25]. Once activated, it drives mitosis up and through metaphase. Further transition to anaphase can only occur after relief of the spindle checkpoint-mediated restriction of the APC/C. As a consequence, Cyclin B1 is rapidly destructed to allow mitotic exit [26]. Evidence from conditional knockout mice suggested that CDK1 has a ubiquitous essential role for cell division [27]. Cyclin A2 is essential in embryonic and hematopoietic stem cells but not in mouse embryonic fibroblasts where its loss can be compensated by E-type cyclins [28]. On the other hand, combined conditional knockout of Cyclin A2 and CDK2 is sufficient to impair fibroblast proliferation [29]. This correlates with the non-overlapping role of Cyclin A2 in many cultured somatic cell lines [14]–[18], [19]–[24], [30].
HCMV disturbs the regular order of events in the cell cycle control machinery. It represses Cyclin D1 [31] and instead employs viral gene products to achieve pRb hyperphosphorylation [32], [33] and E2F target gene activation [34]. It induces Cyclin E1-CDK2 activity [35]–[38] but at the same time inhibits replication licensing [39]–[41]. It inhibits the APC/C ubiquitin ligase leading to the accumulation of numerous APC/C substrate proteins including Cyclin B1 [42]–[44]. However, HCMV blocks expression of the APC/C substrate Cyclin A2 [44]–[46]. Moreover, Cyclin B1-CDK1, although expressed and activated, is retained to the cytoplasm unable to trigger mitotic entry [47]. The net effect of these HCMV-induced alterations is a unique cell cycle state, characterized by a block of cellular DNA synthesis and cell division in the presence of a fully induced nucleotide metabolism as well as up-regulated replication and repair factors. Such conditions are considered to be most favorable for efficient replication of the HCMV genome [7], [48], [49].
Before HCMV can take control over the cell cycle, it is itself subjected to cell cycle-dependent regulation [50]. In the pre-immediate early phase of infection, i.e. the time between virus entry and de novo expression of viral gene products, HCMV is blocked by Cyclin A2-CDK2 activity [51], [52]. This block relies on a Cyclin A2 binding site in the HCMV inner tegument protein pp150 [53], matching the RXL/Cy consensus motif found in cellular Cyclin A2-CDK substrates and inhibitors [54]. Apparently, HCMV has acquired this motif to make the incoming virus particle sensitive to cellular Cyclin A2, thereby restricting the start of lytic gene expression to cells with low or absent Cyclin A2-CDK activity like quiescent, differentiated and G1 cells [53]. Besides this general effect on lytic gene expression, Cyclin A2-CDK interferes more selectively with expression of the essential IE2-86 kDa protein [51]. This might be one of the reasons why HCMV suppresses Cyclin A2 once the lytic gene expression has started [44]–[46].
In this study we aimed to deepen our understanding of HCMV-Cyclin A2 interaction by investigating the cause and consequences of Cyclin A2 repression during lytic infection. We found that HCMV encodes a second RXL/Cy motif that targets the highly unstable pUL21a protein [55] to Cyclin A2, resulting in its proteasomal destruction. This mechanism turned out to be of crucial importance for the HCMV-induced cell cycle arrest, preventing the virus from mitotic entry and resultant abortive infection.
To identify potential HCMV gene products with Cyclin A2 inhibitory function we extended our previous search for Cyclin A2-interaction motifs in virion components [53] to the whole virus proteome [56]. We found that the small HCMV early-late protein pUL21a contains two promising candidate motifs, resembling already validated RXL/Cy sequences of human Cyclin A2-CDK substrates and inhibitors (Fig. 1A). In particular, both candidate motifs possess a bulky hydrophobic residue at either position +1 or +2 relative to the RXL core. This feature is important for docking to the hydrophobic patch region in Cyclin A2 [57] and therefore can serve as a suitable criterion for the identification of candidate RXL/Cy motifs.
To test if pUL21a can physically interact with Cyclin A2 we performed a pull-down experiment, employing His-purified pUL21a as bait. We found that pUL21a-WT was able to precipitate Cyclin A2 from cell extracts with similar efficiency as the viral Cyclin A2 substrate pp150 (Fig. 1C). Pull-down of an RXL-binding deficient Cyclin A2 hydrophobic patch mutant [58] occurred only at background levels, consistent with a pUL21a-RXL-dependent recruitment of Cyclin A2. In fact, point mutation (RXLAXA) of the candidate motif RRLFQ (RXL2) abolished pUL21a-Cyclin A2 binding, whereas the other candidate motif, RRLAF (RXL1), was dispensable for this interaction (Fig. 1C). Cyclin B1, which binds inefficiently to RXL/Cy motifs [59], failed to co-precipitate with pUL21a (Fig. 1D), further demonstrating the specificity of pUL21a-Cyclin A2 binding.
PUL21a lacks full consensus (S/T-P-X-K/R) CDK phosphorylation sites and contains only one minimal consensus site (S/T-P) for CDK and mitogen-activated protein kinases (MAPK)-dependent phosphorylation near the N-terminus (Fig. 1B). Thus, it resembles Cyclin-CDK inhibitors (p21, p27, p57) rather than substrates. To investigate whether pUL21a functions as a Cyclin A2-CDK substrate or inhibitor in vitro, we included the protein in Cyclin A2 kinase assays, using pp150 as a substrate. We found that pUL21a severely compromised Cyclin A2-dependent phosphorylation of pp150 but was itself only weakly phosphorylated. In accordance with the pull-down experiment, the pUL21a-RXL2 mutant was completely inert to Cyclin A2-CDK activity (Fig. 1E). Thus, pUL21a only qualifies as a poor substrate but instead as a strong competitive inhibitor of Cyclin A2-dependent phosphorylation in vitro.
In order to validate the pUL21a-Cyclin A2 interaction under more physiological conditions, we aimed to co-immunoprecipitate both proteins from pUL21a-transfected human cells (Fig. 2A). To improve detection of pUL21a, we stabilized the protein by MG132-mediated inhibition of the proteasome. Under these conditions, we were able to co-precipitate a significant fraction of pUL21a with endogenous Cyclin A2-CDK2 complexes. Like in the preceding experiments, an intact RXL2 motif was required for this interaction. The amount of Cyclin A2-bound pUL21a was increased when using a previously described pUL21a-PRAA mutant (APCmut), deficient in APC/C binding [42]. This was most probably due to the higher availability of pUL21a-APCmut and suggests that Cyclin A2-CDK binding to pUL21a was not saturated.
When comparing cyclin protein levels of MG132-treated and untreated cells, it became apparent that in the presence of an active proteasome, pUL21a transfection resulted in a marked, RXL2-dependent decrease of Cyclin A2 and an RXL1-dependent decrease of Cyclin D1 (Fig. 2B). These effects were specific since Cyclins B1 and E1 were unaffected, and they were similar in strength as the APC/C binding site-dependent down-regulation of APC5 [42]. Furthermore, both Cyclin A2 and APC5 mRNA levels were not influenced by pUL21a (Fig. 2C). Thus, destabilisation of the Cyclin A2 protein appears to be the primary function of pUL21a-Cyclin A2 binding in vivo. In addition, the RXL1-dependent down-regulation of Cyclin D1 pointed to a distinct function of the other putative cyclin docking motif.
We then set out to investigate the functional consequences of pUL21a-Cyclin A2 interaction in the context of viral infection. To this end we introduced the RXL2 mutation into the HCMV genome by traceless BAC-mutagenesis and analyzed the recombinant virus in direct comparison to WT, UL21a deletion (ΔUL21a) and UL21a-PRAA mutant (APCmut) viruses. To ensure a synchronous start of lytic gene expression in infected cells, we used a high multiplicity of infection (MOI) and fibroblasts that were arrested in G0/G1 by contact inhibition. Accordingly, except the G1-specific Cyclin D1, cell cycle factors were barely detectable at the time of infection (Fig. 3A). Whereas Cyclin D1 expression was largely unaffected by HCMV infection and all four viruses led to a strong and sustained induction of Cyclin E1, they markedly differed with respect to Cyclin A2, Cyclin B1 and APC5 protein regulation. Both UL21a deletion and RXL2 point mutation abolished the characteristic block of Cyclin A2 expression observed in HCMV-WT infected cells. Whereas Cyclin A2 mRNA levels were only slightly elevated (see Fig. 3B), the effect was most pronounced at the level of protein expression, consistent with the observed proteasome dependency of Cyclin A2 down-regulation in pUL21a-transfected cells (Fig. 2). Remarkably, the RXL2 point mutation resulted in considerably higher peak levels of Cyclin A2 protein expression than the deletion of the whole UL21a open reading frame. This may be explained by the fact that the APC/C inhibitory function of UL21a is not affected by the loss of Cyclin A2 binding (see APC5 levels in Fig. 3A) and thus can contribute to Cyclin A2 stabilization. Cyclin A2 induction in HCMV-UL21a-RXL2mut-infected cells was followed by a strong up-regulation of Cyclin B1 mRNA and protein expression, most likely due to Cyclin A2-dependent transcriptional activation of the Cyclin B1 promoter [23]. Again, the effects of UL21a deletion were much weaker, hardly exceeding the moderate levels of Cyclin B1 induction in HCMV-WT-infected cells. Thus, the simultaneous loss of Cyclin A2 and APC/C-binding sites in pUL21a masks the full potential of its Cyclin A2 and B1-directed negative control function.
We next asked what impact, if any, the increased amounts of Cyclin A2 and B1 at early-late times of infection have on viral replication. We found that the UL21a-RXL2 mutation caused a clear delay of up to 24 h in the overall accumulation of late (pp28, pp150) gene products. Furthermore, the characteristic late increase in IE2 protein expression was also suppressed in HCMV-UL21a-RXL2mut-infected cells (Fig. 3A). This correlated with an almost fourfold reduction in viral progeny production compared to HCMV-WT (Fig. 3C).
The changes induced by the loss of pUL21a-mediated control over Cyclin A2 became more evident when the cell cycle profiles of infected cells were analyzed. Between 24 and 48 h post infection, both UL21a deletion and RXL2 mutation caused a rapid increase in the rate of DNA synthesis, leading to an accumulation of cells with 4n DNA content (Fig. 4A). This increase was due to cellular rather than viral DNA replication, as indicated by BrdU pulse-labeling experiments. Both in HCMV-WT and HCMV-APCmut-infected cells the sites of BrdU incorporation co-localized with pUL44, which is the processivity factor of the viral DNA polymerase and a well-accepted marker of ongoing viral DNA replication [60], [61]. This restriction of DNA synthesis to viral replication compartments was overcome in the majority of HCMV-ΔUL21a and HCMV-RXL2mut-infected cells. There, nucleotide incorporation was detected throughout the nucleus, both in pUL44-positive and negative regions (Fig 4B–C). The presence of cellular DNA synthesis in these cells was confirmed by suppressing viral replication with foscarnet (PFA), a selective inhibitor of herpesviral DNA polymerases. Although PFA treatment, consistent with published data [62], caused some delay in S-phase progression, a large number of HCMV-ΔUL21a and HCMV-UL21a-RXL2mut-infected cells had doubled their DNA content after 96 h (Fig. S1). Thus, the pUL21a-Cyclin A2 interaction is essential for the block of cellular DNA synthesis in HCMV-infected cells.
We then examined whether the Cyclin A2-induced progression through S-phase is followed by mitotic entry of infected cells. Entry into mitosis is marked by cell rounding and chromatin condensation and therefore can be readily visualized by phase contrast microscopy and DAPI staining. Both methods revealed a high number of mitotic cells at 48 h after infection with the UL21a-RXL2 mutant virus (Fig. 4C, 5A). To quantitatively assess the number of mitotic cells, we performed flow cytometry of histone H3-serine 10 phosphorylation (pH3ser10), a marker of condensed chromatin. Parallel analysis of DNA content and IE1/2 protein expression in combination with an appropriate gating strategy ensured that only single, IE-positive cells were evaluated (Fig. S2). We found that the percentage of HCMV-UL21a-RXL2mut-infected cells displaying H3(ser10) phosphorylation increased from about 20% at day 2 to almost 30% at day 3 before it declined again to just under 10% at day 5 (Fig. 5B–C). Some cells had undergone chromatin condensation prematurely, i.e. before completion of DNA replication (Fig. 5B). In case of UL21a deletion, the proportion of mitotic cells was also increased compared to HCMV-WT, but remained constantly below 3% and showed no signs of premature mitotic entry. This low rate of mitosis is well supported by the only moderate strength of Cyclin B1 induction in these cells (Fig. 3A). As it is rather unusual that a point mutation has a stronger phenotype than a whole gene knockout, it was important to ensure that cyclin up-regulation and mitotic entry were not caused by unwanted secondary mutations in the genome of the UL21a-RXL2-mutant virus. We therefore repaired the RXL2 mutation and analyzed the cell cycle effects of the UL21a-RXL2-revertant virus. Cyclin A2 and B1 protein expression as well as mitotic cell numbers were reverted to HCMV-WT levels (Fig. S3), demonstrating that the observed changes are specifically linked to the pUL21a-RXL2 motif. We concluded that pUL21a-Cyclin A2 interaction is not only required for the inhibition DNA synthesis in HCMV-infected cells but also for the inhibition of mitotic entry.
It is important to note that even in case of the RXL2 mutant the majority of infected, Cyclin A2 expressing cells did not traverse the G2/M boundary but remained in S/G2. In contrast to the mitotic cell population, these pH3(ser10)-negative cells appeared to support viral DNA replication to a reasonable extent as many of them acquired a greater than 4n DNA content in a PFA-sensitive manner (Fig. 5B, Fig. S1). This may explain why Cyclin A2 up-regulation has no stronger effect on virus growth (Fig. 3C).
To test whether the observed effects of UL21a-RXL2 mutation on cell cycle progression and virus replication are a TB40-specific phenomenon, we introduced the same mutation in the highly fibroblast-adapted HCMV strain AD169. Following infection of confluent fibroblasts, the AD169-UL21a-RXL2 mutant virus induced similar levels of mitotic entry (Fig. S4A) as the TB40 counterpart (Fig. 5A, Fig. S3B) and also had a similarly moderate growth defect (Fig. S4B). These results argue against a virus strain dependency of the UL21a-RXL2mut phenotype.
Under normal circumstances, the duration of mitosis in cultured fibroblasts is about 1 h. The finding of HCMV-UL21a-RXL2mut-infected cells remaining at high numbers in mitosis over days pointed to a blockade or at least a severe prolongation of the mitotic process. We reasoned that the previously reported restriction of Cyclin B1-CDK1 activity to the cytoplasm of HCMV-infected cells [47] might prevent the nuclear envelope breakdown in late prophase [63]. To test this possibility, we analyzed the nucleo-cytoplasmic distribution of Cyclin B1-CDK1 and the integrity of the nuclear lamina. As expected, we were able to confirm the cytoplasmic sequestration of CDK1 by HCMV-WT. However, we observed a high proportion of CDK1 in the nuclear fraction of UL21a-RXL2mut-infected cells (Fig. 6A), suggesting that Cyclin A2 down-regulation by pUL21a is also responsible for the limited nuclear availability of this essential mitotic kinase during HCMV infection [47]. Interestingly, when we checked the nucleo-cytoplasmic distribution of Cyclin A2 and pUL21a (Fig. S4), we found that Cyclin A2 resembled CDK1 in a way that nuclear translocation was only possible in the absence of pUL21a binding. In contrast, pUL21a itself was found in both cellular compartments. In accordance with the observed nuclear entry of Cyclin A2, Cyclin B1 and CDK1, chromosome condensation in mitotic HCMV-UL21a-RXL2mut-infected cells was accompanied by a loss of the nuclear envelope (Fig. 6B). Hence, there was no evidence for a block of prophase-prometaphase transition.
In order to identify the step in mitosis that is impeded in cells infected with the UL21a-RXL2 mutant, we next visualized mitotic figures by fluorescent staining of chromosomes (DAPI), centromeres (CENP-A) and the spindle apparatus (α-Tubulin). The results were clear-cut. At 2 days post infection, the large majority of mitotic cells had accumulated in metaphase whilst anaphase and telophase figures were completely absent (Fig. 7A–B). Since in HCMV-ΔUL21a-infected cells all mitotic stages were represented, this metaphase arrest appeared to be a direct consequence of the still intact APC/C-inhibitory function in pUL21a-RXL2mut. Remarkably, not all metaphase cells contained a regular, bipolar mitotic spindle. About one third displayed an aberrant, mono- or multipolar spindle formation (Fig. 7A–B). This phenomenon was not specific for the UL21a-RXL2 mutant as it was also observed in case of the UL21a deletion virus (Fig. 7B). Over time, more and more of the metaphase-arrested cells acquired additional pathological features and ultimately disintegrated entirely. The most obvious alteration was the appearance of dispersed chromosomal material that no longer co-localized with the centromere-specific histone H3 variant CENP-A (Fig. 7A–B). Whereas the CENP-A-containing foci were still in contact with the mitotic spindle, the respective chromosomal material was detached (Fig. 7A). This indicates that these chromosomes had lost their centromeric region, most likely as a result of chromosomal instability. The centromere loss occurred in HCMV-infected cells as demonstrated by IE1/IE2-costaining (Fig. S6). Thus, a decomposition of chromosomes appeared to characterize the end-stage of the aberrant mitotic process induced by UL21a-RXL2-deficient HCMV.
To examine the structural integrity of mitotic chromosomes in more detail, we analyzed metaphase spreads of HCMV-UL21a-RXL2mut-infected cells by various methods. First, we re-applied CENP-A and DAPI co-staining to confirm the assumed chromosomal disintegration. Numerous acentric chromatid fragments were found dispersed to the periphery of the spreads, whereas the residual, CENP-A-positive material often clustered in central, metaphase plate-like structures (Fig. 8). Non-infected, prometaphase-arrested control cells, in contrast, showed regular pairs of sister chromatids and centromeres (Fig. 8, upper panel). The sheer extent of chromosome shattering in many HCMV-UL21a-RXL2mut-infected cells (Fig. 8, lower panels) pointed towards a general, and not a chromosome-specific destruction process.
Giemsa staining revealed further chromosomal abnormalities, contrasting with the characteristic X-shaped chromosomal appearance of prometaphase-arrested control cells (Fig. 9A, upper left panel). Besides chromosomal fragmentation, HCMV-UL21a-RXL2mut-infected, mitotic cells displayed incompletely condensed chromosomes (Fig. 9A, magnified view #2) or showed signs of precocious separation of sister chromatids (Fig. 9A, magnified view #1). In addition, the extent of chromosomal fragmentation increased over time and “pulverized” chromosomes dominated the picture of mitotic cells at 3–4 days post infection (Fig. 9A, lower panel). Notably, the remaining non-mitotic HCMV-UL21a-RXL2mut-infected cells appeared at this time with fully developed viral replication compartments, visible as dark-stained intranuclear regions. This is in accordance with the observation of viral DNA replication in the G2-arrested, non-mitotic subpopulation (Fig. 5B, Fig. S1) and with the only moderate growth defect of pUL21a-Cyclin A2 binding-deficient viruses (Fig. 3C). It appears that not Cyclin A2 expression per se but mitotic entry had deleterious consequences for HCMV-infected cells.
Chromosome fragmentation is often difficult to distinguish morphologically from premature, incomplete chromosome condensation [64]. To obtain further proof for the presence of chromosomal breakage and to enable a more quantitative assessment of the different chromosomal abnormalities, we performed fluorescence in situ hybridization (FISH) analysis of two large chromosomes, 1 and 3 (Fig. 9B–C). Incompletely condensed chromosomes were recognized by their extended conformation or beads-on-a-string appearance. A clear indication of chromosome breakage was the random distribution of chromosome 1 or 3 fragments over the whole metaphase spread. The whole chromosome FISH analysis confirmed the presence of both phenotypes. At 2 days post infection, some HCMV-UL21a-RXL2mut-infected mitotic cells even showed a mixed phenotype with features of both prematurely condensed and partially fragmented chromosomes (Fig. 9B), suggesting that in these cells premature condensation gives way to chromosomal breakage [65]. In support of this view, the percentage of mitotic cells with fragmented chromosomes almost doubled between 2 and 3 days post infection (from 34% to 59%), at the expense of cells with incompletely condensed and normal metaphase chromosomes (Fig. 9C). Notably, about one third of mitotic cells showed no major signs of chromosomal damage at day 2 to 3 post HCMV-UL21a-RXL2mut infection but instead contained precociously separated sister chromatids (Fig. 9B–C). This points to a phenomenon called cohesion fatigue, which is often found in cells arrested or delayed at metaphase [66].
Having completed the initial characterization of virus-induced mitotic aberrations, we were next interested to gain insight into the implications of mitotic entry for HCMV replication. First, we analyzed the consequences for major immediate early proteins IE1 and IE2 (Fig. 10). Both nucleoproteins were easily detectable in the non-mitotic fraction of HCMV-UL21a-RXL2mut-infected cells and showed a regular nuclear localization pattern, with IE1 being evenly distributed and IE2 accumulating in viral replication compartments [67]. In mitotic cells, however, the picture was dramatically different. Whereas IE1 was found associated with metaphase chromosomes, as expected from published work [68], [69], IE2 protein expression was below detection limit (Fig. 10). The lack of this essential viral transcription factor suggests that virus replication is aborted in mitosis. This was further supported when we examined the morphology and activity of viral replication compartments in metaphase-arrested cells (Fig. 11). Judged on the basis of pUL44 staining, the replication compartments appeared much smaller compared to interphase cells and were displaced to the cellular periphery. Moreover, no incorporation of BrdU was detectable suggesting that viral DNA synthesis and accordingly progression of the infectious cycle was stalled. We concluded that the pUL21a-dependent arrest at the G1/S-transition eventually protects HCMV-infected cells from entering an abortive mitotic stage in which the virus can neither replicate nor enable the infected cell to exit from.
The antagonism between Cyclin A2-CDK activity and viral gene expression has been recognized as a characteristic hallmark of HCMV-cell cycle interactions [7], [46], [51], [52]. Recently, we discovered that HCMV employs an RXL/Cy motif in the pp150 tegument protein to sense the cellular Cyclin A2 status right at the beginning of an infection [51], [53]. This sensing mechanism is required to restrict the onset of IE gene expression to the G0/G1 phase, where Cyclin A2-CDK is inactive. Here, we found how HCMV makes use of another RXL/Cy motif in the early gene product pUL21a to maintain the status of low Cyclin A2-CDK activity after lytic gene expression has started. This mechanism is not only required to block cellular DNA synthesis but ensures that viral replication is not abrogated by entry into mitosis. Possibly, it also protects pp150 at the stage of virion assembly from Cyclin A2-CDK-dependent phosphorylation so that pp150 can support the onset of lytic gene expression in newly infected G0/G1 cells. Such cooperative action would explain why the progeny of UL21a-deficient HCMV has a reduced infectivity despite pUL21a itself is not a component of viral particles [55]. Thus, we suggest that both virus-encoded interactors of Cyclin A2 constitute a control circuitry (Fig. 12) that synchronizes the HCMV lytic cycle to the cell division cycle in a way that secures efficient viral growth.
Cyclin A2 destruction is after APC4/5 destruction [42] the second cell cycle regulatory function ascribed to pUL21a. This small protein appears to be perfectly designed to confer its own instability to every protein it binds to. Importantly, we found (Fig. 3A–B, 6A) that Cyclin A2 down-regulation by pUL21a also accounts for the only moderate expression of Cyclin B1 and the previously reported cytoplasmic sequestration of CDK1 in HCMV-infected cells [47]. Because the APC/C ubiquitin ligase is required for proteasomal degradation of Cyclin A2 and Cyclin B1 from M to G1 phase [13], the two functions residing in pUL21a, APC/C inhibition and Cyclin A2 degradation, inversely regulate Cyclin A2 and Cyclin B1 abundance (Fig. 13). This leads to the seemingly paradoxical situation that pUL21a-RXL point mutation has a much stronger effect on Cyclin A2 and B1 expression (Fig. 3A–B), CDK1 nuclear translocation (Fig. 6A) and mitotic entry (Fig. 5) of infected cells than deletion of the whole UL21a open reading frame (Fig. 13). Why has HCMV concentrated two potent antagonistic cell cycle activities on the same protein? Probably, HCMV benefits from stabilization of the many other APC/C substrates [44], [70] and, by integrating a Cyclin A2-binding motif into the same regulatory module, has found an elegant way to counteract the negative consequences of increased Cyclin A2 and B1 stability. Furthermore, it might be beneficial for virus replication if the APC/C-inhibitory function of pUL21a, which alone elicits mitotic entry and fatal metaphase arrest, is balanced by the G1 arrest function of pUL21a-mediated Cyclin A2 degradation (Fig. 13). Our data do not exclude the existence of additional mechanisms contributing to Cyclin A2 down-regulation at the level of transcription [50]. In fact, repression of the Cyclin A2 promoter in the early phase of infection [46] might account for the only moderate effect of pUL21a-RXL mutation on Cyclin A2 protein expression at 24 hpi (Fig. 3A).
Although Cyclin A2 is an important regulator of S phase [71], [72], the necessity of pUL21a-Cyclin A2 interaction for the inhibition of cellular DNA synthesis in HCMV-infected cells is an unexpected result. Previous evidence suggested that HCMV interferes primarily with the process of replication licensing, by inhibiting the loading of the MCM2-7 helicase complex onto chromatin [39], [41]. Two mechanisms have been proposed: up-regulation of the cellular licensing inhibitor Geminin [39] and expression of the viral licensing inhibitor pUL117 [40]. In addition, a post-licensing inhibition of the MCM2-7 complex by direct physical interaction between the MCM3 acetylase MCM3AP and HCMV-IE2 has been reported [73]. Conversely, low Cyclin A2-CDK activity is known to promote, rather than constrain, MCM2-7 loading [74]–[76], arguing against a causative role of pUL21a-Cyclin A2 interaction in the viral inhibition of pre-replicative complex formation. Only recently, Cyclin A2-MCM7 binding was described to be critical for the S-phase promoting function of Cyclin A2 [77]. Two scenarios can be considered to explain the apparent dominance of Cyclin A2 over other virus-encoded control mechanisms: either the MCM-directed viral inhibitors alone are not sufficient to maintain a tight and stable block to the onset of cellular replication or they are negatively influenced by Cyclin A2 expression.
Notably in this context, Cyclin A2 over-expression [51] as well as UL21a deletion [78] have been shown to specifically impair mRNA expression of the essential viral trans-activator and S phase inhibitor IE2 [38], [73]. The causal chain from pUL21a via Cyclin A2 to IE2 expression and inhibition of cellular DNA synthesis was confirmed by another report by Caffarelli et al that appeared while this manuscript was in preparation [79]. Importantly, Caffarelli et al were able to overcome the negative effects of pUL21a-RXL2 mutation on IE2 expression and virus replication by Cyclin A2 knockdown [79]. In the present study, IE2 protein accumulation at late times of infection was suppressed by UL21a-RXL point mutation but only marginally affected by UL21a deletion (Fig. 3A), correlating well with their different impact on Cyclin A2 expression and mitotic entry (see above). In fact, with the elimination of IE2 protein in mitosis (Fig. 10) we discovered a further level of IE2-specific regulation, contributing to the overall decrease in IE2 expression of HCMV-UL21a-RXL2mut-infected cells. Caffarelli et al [79] possibly missed mitotic entry of UL21a mutant-infected cells because their cell cycle analysis was carried out at early times of infection (24–48 hpi) and in the presence of cell cycle-retarding concentrations of phosphonoacetic acid [51]. An exact side-by-side comparison, however, of Cyclin A2-mediated effects on IE2 and the cell cycle in this and previous studies is difficult, given that all previous analyses employed the highly laboratory-adapted strain AD169 [51], and used either proliferating [78], [79] or Cyclin A2-overexpressing cells [51] for infection. Here, in contrast, the low-passage endotheliotropic strain TB40 and growth-arrested cells were used as starting materials for most experiments. Given that in our cellular system an AD169-UL21a-RXL2 mutant showed a very similar cell cycle and virus growth phenotype (Fig. S4) as the corresponding TB40 mutant (Fig. 3C, Fig. 5A), it appears that different host cell conditions at the time of infection are the most likely explanation for the more severe growth defects reported for UL21a-RXL and deletion mutants in the AD169 background [78], [79]. Although we provided clear evidence for an abrogation of viral replication after entry into mitosis (Fig. 10–11), it appears that the remaining two thirds of cells in interphase (mainly G2 phase, see Fig. 4–5, Fig. S1) supported viral DNA replication and release of HCMV progeny to a reasonable extent. It remains to be determined how distinct HCMV genotypes and varying infection conditions influence the actual outcome of pUL21a-Cyclin A2 interaction on IE2 expression and virus growth.
Animal CMVs lack a Cyclin A2-binding site in their pp150 homologues and accordingly can initiate viral gene expression independent of the host cell cycle state [53]. Except for primate CMVs, where the Cyclin A2-destabilizing function of pUL21a is conserved [79], animal CMVs lack also a pUL21a homologue and therefore, not surprisingly, murine CMV (MCMV) has been found to induce not only Cyclin E1 but also Cyclin A2-dependent kinase activity [51]. However, Cyclin A2 up-regulation does not result in mitotic entry of MCMV-infected cells, which instead become arrested in G1 and G2 by a yet unknown IE3-dependent mechanism [80]. Both G1 and G2-arrested cells support MCMV DNA replication with similar efficiency [80]. This resembles the situation in HCMV-UL21a-RXL2mut and HCMV-ΔUL21-infected G2 cells where PFA-sensitive viral DNA synthesis leads to a greater than 4n DNA content (Fig. 5B, Fig. S1). Thus, it appears that animal CMVs have evolved their own, pUL21a-independent cell cycle arrest mechanisms to prevent mitotic entry, thereby counteracting the negative consequences that Cyclin A2 up-regulation can have for viral replication.
The removal of pUL21a-dependent Cyclin A2 repression reveals the full mitogenic potential of HCMV. The UL21a-RXL-mutant virus was able to force density-arrested fibroblasts to re-enter the cell cycle, to traverse through S phase and enter mitosis. The presence of anaphase and telophase figures (Fig. 7B) even suggested that some cells, infected by the UL21a deletion virus, were able to divide. Thus, Cyclin A2 provides the missing link between virus-induced G1/S-promoting [32], [34], [38], [81] and G2/M-promoting activities [47], [82] that alone were unable to drive cell cycle progression to completion. Of particular interest in this context is the fact that Hertel and Mocarski have already described a “pseudomitotic” phenotype in the late phase of HCMV infection. This is characterized by an upregulation of mitosis-related gene expression and formation of abnormal mitotic spindles in the presence of an intact nuclear envelope [82]. This phenotype fits very well to previous findings by Sanchez et al showing HCMV-mediated activation but also cytoplasmic sequestration of the mitotic kinase Cyclin B1-CDK1 [47], whose nuclear translocation is an absolute prerequisite for nuclear envelope breakdown in prophase [63]. Considering that Cyclin A2 induction by UL21a-RXL2 mutation overcome these limitations (Fig. 6), Cyclin A2 clearly appears to be the rate-limiting factor for mitosis entry in HCMV-infected cells.
Primary, non-transformed cells possess potent surveillance mechanisms that arrest the cell cycle in response to deregulated DNA replication [83]. Similarly, entrance into mitosis is blocked if DNA replication is not completed [84]. Unscheduled cellular DNA synthesis (Fig. 4) and premature chromatin condensation (Fig. 5B) in UL21a-RXL-mutant-infected cells indicate that HCMV overrides these checkpoints, most likely by abrogating p53 and Chk2-dependent signaling [85], [86]. This and the partitioning of DNA repair enzymes to viral replication compartments predisposes the host cells to the accumulation of DNA damage [49], [87], [88]. It is yet unclear to what extent the intrinsic genotoxicity of HCMV contributes to the severe chromosomal instability seen after mitotic entry of infected cells. However, it is remarkable that checkpoint bypass by dual deficiency of p53 and Chk1 was recently shown to result in a mitotic catastrophe that closely resembles the extensive fragmentation of chromosomes and centromeres in HCMV-UL21a-RXL2mut-infected cells [89], [90]. An alternative explanation is that the prolonged mitotic arrest gives rise to the progressively increasing chromosomal damage seen in these cells [91]. This view is supported by the simultaneous loss of sister chromatid cohesion in many cells (Fig. 9), which is also considered as a characteristic result of extended metaphase arrest [92], [93]. Regardless to what extent the different mechanisms contribute to the severe chromosomal aberrations induced by the HCMV-UL21a-RXL mutant, the inherent capacity of this virus to promote both cell cycle progression and genetic instability may fuel the ongoing discussion about its potential role in cancer development and progression [94], [95].
All procedures involving animals and their care were approved (protocol class 003-08/13-01/14; number 2170-24-01-13-03) by the Ethics Committee for Biomedical Research of the University of Rijeka Faculty of Medicine (Croatia) and were conducted in compliance with institutional guidelines as well as with national (Animal Protection Act 135/06 and 37/13; Regulations on the Protection of Animals Used for Scientific Purposes 55/2013) and international (Directive 2010/63/EU of the European Parliament and of the Council of 22 September 2010 on the protection of animals used for scientific purposes; Guide for the Care and Use of Laboratory Animals, National Research Council, 1996) laws and policies. Animals were bred and raised at the Laboratory for Mouse Breeding and Engineering Rijeka (LAMRI), University of Rijeka Faculty of Medicine. The authorization for breeding of and experiments on laboratory mice in this facility has been obtained by the competent national body (Veterinary Department of the Croatian Ministry of Agriculture – authorization No. HR-POK-004).
Human embryonic lung (HEL) fibroblasts (Fi301, obtained from the Institute of Virology, Charité, Berlin, Germany) and human embryonic kidney (HEK) 293 cells (obtained from the Leibniz Institute DSMZ – German Collection of Microorganisms and Cell Cultures, Braunschweig, Germany) were maintained as described previously [52]. Where indicated, cells were treated with 2.5 µM MG132 or 50 ng/ml nocodazole (20 h before harvest) or were labeled with 10 µM 5′-bromo-2′-deoxyuridine (BrdU) for 1 h before harvest. Viruses were derived from the HCMV strain TB40-BAC4 [96] and modified by bacterial artificial chromosome (BAC) mutagenesis (see below). Viruses were propagated on HEL fibroblasts and infectious titers were determined as follows. Fibroblasts were seeded on 6-well plates and grown to confluence. Then, cells were infected with a series of virus dilutions in 0.5 ml of growth medium per well. The virus inocula were replaced by fresh culture medium after an adsorption period of 1 h. Cells were harvested at 24 hpi by trypsinization, counted with a hemocytometer and permeabilized with ice-cold PBS/80% ethanol for at least 5 min. After washing with PBS/1% BSA, cells were stained with an Alexa Fluor 488-conjugated IE1/2-specific antibody (clone 8B1.2, Merck-Millipore). The percentage of IE-positive cells was determined by flow cytometry. Only virus dilutions resulting in 1–10% IE-positive cells were used to calculate virus titers. This was done by multiplying the percentage of IE-positive cells by total cell number and dilution factor. For experiments, a multiplicity of infection (MOI) of 5–10 IE-protein forming units per cell was used. Fibroblasts were grown to confluence before infection to synchronize them in early G1 phase.
A traceless mutagenesis method [97] was used to generate TB40-UL21a-RXL2mut, AD169-UL21a-RXL2mut, TB40-UL21a-APCmut and TB40-ΔUL21a. In brief, a kanamycin resistance cassette (KanR) was amplified from the plasmid pEPkanS2 [97], using the primer pairs UL21a-RRLFQmut-fw/rev, UL21a-PRmutAA-fw/rev, UL21a-del-fw/rev (table 1). The PCR products were transformed into Escherichia coli GS1783 [97], containing either TB40-BAC4 or AD169-pHB5 (both kindly provided by Jens von Einem, Ulm, Germany), and selected for integration of KanR. Subsequently, the KanR was removed by induction of λRed-recombinase and I-SceI expression. The HCMV-TB40-UL21a-RXL2 mutation was reverted using the primer pair UL21a-RRLFQ-revert-fw/rev (table 1). All steps of BAC mutagenesis were controlled by PCR and sequencing of the modified region. BACs were prepared and purified over NucleoBond Xtra-Midi columns (Macherey-Nagel) following the manufacturer's instructions. To reconstitute recombinant virus, the BACs were cotransfected with pp71/pUL82 and Cre recombinase expression plasmids into HEL fibroblasts by Amaxa nucleofection (Lonza).
The pUL21a coding sequence was amplified from TB40-BAC4, using primers UL21a-fw-BamH1 and UL21a-rev-Sal1 (see table 1). After BamH1/Sal1-mediated insertion into the prokaryotic expression vector pET-52b(+) (Merck), the resulting construct pET-52b(+)-UL21a served as template for site-directed inverse PCR mutagenesis. To introduce the RXL1 and RXL2 mutations, the 5′-phosphorylated primer pairs UL21a-RXL1-ARAAF-BssH/UL21a-RXL1-rev and UL21a-RXL2-ARAFQ-BssH/UL21a-RXL2-rev were used. The amplified, nicked plasmids were incubated with T4 DNA Ligase for 1 h at room temperature before being transformed into the Escherichia coli strain XL1-blue. UL21a-WT, UL21a-RXL1mut and UL21a-RXL2mut cDNAs were subcloned from pET52b(+) into the eukaryotic expression vector pCI-neo (Promega) in-frame to an N-terminal triple-hemagglutinin (3HA) epitope tag. To introduce a point mutation (PRAA) in the APC/C binding site (UL21a-APCmut), the pCI-neo-3HA-UL21a-WT plasmid was subjected to site-directed mutagenesis (see above) using the 5′-phosphorylated primers UL21a-APC-AA-NarI and UL21a-APC-rev (see table 1). All plasmids were confirmed by sequencing and purified by CsCl-ethidium bromide equilibrium centrifugation.
The pET-52b(+)-derived expression vectors were transformed into the Escherichia coli strain BL21-CodonPlus (DE3)-RILP (Stratagene). Recombinant Strep-pUL21a-His and Strep-pp150c-His proteins were expressed, purified and used for His-pull down assays as previously described [53], except for the following modifications: (i) pUL21a expression was induced by addition of isopropyl-β-D-1-thiogalactopyranoside (IPTG) to a final concentration of 0.5 mM; ii) after binding to Ni-NTA beads, pUL21a protein was washed 7 times with buffer containing increasing concentrations (up to 200 mM) of imidazole.
Kinase assays were performed as described, using recombinant pp150c as substrate [53]. Where indicated, equal amounts of either pUL21a-RXL2 or pUL21a-WT protein was added to the kinase reactions.
Two BALB/c mice were immunized by subcutaneous injection of 50 µg of purified Strep-pUL21a-His protein in complete Freund's adjuvant. After three weeks, two mice were boosted with 50 µg of the same protein in complete Freund's adjuvant by injecting two-third volume subcutaneously and one-third volume intraperitoneally (i.p.). After two weeks, the sera of both animals were tested by ELISA for antibody titer against the recombinant pUL21a immunogen and the better responder was additionally boosted i.p. with 50 µg of the same protein dissolved in PBS. Three days later, spleen cells were collected and fused with SP2/O myeloma cells (ATCC: CRL 1581) at a ratio of 1∶1. The cells were seeded on 96-well tissue-culture plates in 20% Roswell Park Memorial Institute (RPMI) 1640 medium containing hypoxanthine, aminopterine, and thymidine for hybridoma selection. Supernatants of the generated mother-well cell lines were screened for antibodies, reactive against pUL21a immunogen by ELISA. Positive mother 4G12 was further cloned to generate UL21.02 cell line secreting monoclonal antibodies against pUL21a.
The pCI-neo-3HA-UL21a expression plasmids were transfected into proliferating HEK293 cells using the Turbofect reagent (Fermentas). To stabilize pUL21a, MG132 was added to the cell culture medium at 24 h post transfection. Cells were harvested at 48 h post transfection and extracted by freezing-thawing in immunoprecipitation buffer (IPB): 50 mM Tris–Cl pH 7.4, 150 mM NaCl, 10 mM MgCl2, 10 mM NaF, 0.5 mM Na3VO4, 0.5% Nonidet P-40, 10% glycerol, 1 mM dithiothreitol (DTT), 2 g/ml aprotinin, 1 mM leupeptin, 1 mM Pefabloc. Cyclin A2-containing protein complexes were immunoprecipitated by incubating extracts for 2 h with agarose-conjugated Cyclin A2 antibodies (sc-751 AC, Santa Cruz). After 4 washing steps with IPB, the agarose-bound proteins were analyzed by immunoblotting for the presence of Cyclin A2, CDK2 and HA-tagged pUL21a.
Nuclear and cytoplasmic fractions were prepared as described previously [53].
Immunoblot analysis was carried out as described [52], using antibodies against IE1 (6E1, Vancouver Biotech), IE2 (clone 12E2, Vancouver Biotech), GAPDH (clone 6C5, Santa Cruz), HA (clones 12CA5 and 3F10, Roche), Cyclin A2 (clone BF683, BD Biosciences), Cyclin B1 (clone GNS1, Santa Cruz), Cyclin D1 (clone EPR2241, Abcam), Cyclin E1 (clone H-12, Santa Cruz), CDK1 (PC25, Merck-Calbiochem), CDK2 (clone 55, BD Biosciences), APC5 (A301-026A, Bethyl Laboratories), Lamin A/C (clone 636, Santa Cruz), β-Tubulin (clone 2-28-33, Sigma-Aldrich), pp65 (clone CH-12, Santa Cruz), pp28 (clone 5C3, Santa Cruz), pp150 (clone XP1, a gift of Bodo Plachter). The Strep-tag HRP detection kit (IBA Lifesciences) was used to detect bacterially expressed Strep-pUL21a-His proteins.
Multi-probe ribonuclease protection assays were performed as described earlier [38]. The template set hCyc2 (BD Biosciences) was used for in vitro transcription of the radioactively labeled probe.
Total cellular RNA was prepared, quantified, reverse-transcribed and analyzed by real-time PCR as described [51]. To determine mRNA expression levels of Cyclin A2, APC5 and GAPDH, the primer pairs cyclinA2-fw/cyclin A2-rev, APC5-fw/APC5-rev and GAPDH-fw/GAPDH-rev were used (see table 1).
Cells were fixed, permeabilized and co-stained with propidium iodide and fluorescently labeled antibodies as previously described [52]. The following antibody combination was used: Alexa Fluor 488-conjugated anti-IE1/2 (clone 8B1.2, Merck-Millipore), anti-pH3(ser10) (clone 6G3, Cell Signaling) and BD Horizon V450-conjugated rat anti-mouse IgG1 (clone A85-1, BD Biosciences).
HEL fibroblasts were grown to confluency on glass coverslips before infection. After harvest, the cells were washed with PBS, fixed with 4% paraformaldehyde in PBS, permeabilized with PBST (PBS, 0.1% Triton X-100, 0.05% Tween 20), blocked and immunostained essentially as described elsewhere [98]. To enable immunofluorescence detection of chromosome-bound IE1 in mitosis (Fig. 6B, Fig. 10, Fig. S5), cells were fixed and permeabilized with ice-cold methanol according to Mücke et al [69]. To enable immunofluorescence microscopy of chromosome spreads, cells were swelled after harvest in hypotonic buffer (75 mM KCl) for 15 min at 37°C. Using a Cellspin I centrifuge (Tharmac), aliquots of 2.5×104 cells were then cytocentrifuged at 250 g for 10 min onto a 6 mm×6 mm square sample area of Superfrost Plus glass slides (Thermo Scientific). After attachment, cells were allowed to dry for 15 min. Subsequently, cells were fixed with 4% paraformaldehyde for 10 min at room temperature and subjected to immunostaining. Primary antibodies against CENP-A (clone 3–19, GeneTex), pH3-ser10 (clone 6G3, Cell Signaling), Lamin A/C (clone 636, Santa Cruz), α-Tubulin (clone YL1/2, Merck Millipore), pUL44 (clone CH16, Santa Cruz), IE1/2 (clone 8B1.2, Merck Millipore), IE1 (clone 6E1, Vancouver Biotech) and IE2 (clone 12E2, Vancouver Biotech) were used. To allow various combinations of mouse and rat primary antibody clones, highly cross-absorbed anti-mouse IgG and IgG isotype-specific antibodies (Life Technologies) were employed as secondary reagents. For instance, the staining approach used in Fig. 6B included three different pairs of mouse primary and isotype-specific secondary antibodies (anti-IE1/2 + anti-mouse IgG2a-Alexa 488, anti-Lamin A/C + anti-mouse IgG2b-Alexa 594, anti-pH3-ser10 + anti-mouse IgG1-Alexa 647). Interphase nuclei and mitotic chromosomes were always counterstained by the use of 4′,6-diamidin-2-phenylindol (DAPI). Images were acquired by an Eclipse A1 laser-scanning microscope, using NIS-Elements software (Nikon Instruments). Equal microscope settings and exposure times were used to allow direct comparison between samples.
Giemsa staining was performed essentially as described [80]. Fluorescence in situ hybridization (FISH) was performed with whole chromosome painting (WCP) probes (Metasystems) for chromosome 1 (spectrum orange-labeled) and chromosome 3 (spectrum green-labeled) following the manufacturer's instructions with slight modifications.
|
10.1371/journal.pgen.1005292 | DNA Damage Regulates Translation through β-TRCP Targeting of CReP | The Skp1-Cul1-F box complex (SCF) associates with any one of a number of F box proteins, which serve as substrate binding adaptors. The human F box protein βTRCP directs the conjugation of ubiquitin to a variety of substrate proteins, leading to the destruction of the substrate by the proteasome. To identify βTRCP substrates, we employed a recently-developed technique, called Ligase Trapping, wherein a ubiquitin ligase is fused to a ubiquitin-binding domain to “trap” ubiquitinated substrates. 88% of the candidate substrates that we examined were bona fide substrates, comprising twelve previously validated substrates, eleven new substrates and three false positives. One βTRCP substrate, CReP, is a Protein Phosphatase 1 (PP1) specificity subunit that targets the translation initiation factor eIF2α to promote the removal of a stress-induced inhibitory phosphorylation and increase cap-dependent translation. We found that CReP is targeted by βTRCP for degradation upon DNA damage. Using a stable CReP allele, we show that depletion of CReP is required for the full induction of eIF2α phosphorylation upon DNA damage, and contributes to keeping the levels of translation low as cells recover from DNA damage.
| Approximately 600 human genes encode enzymes that act as ubiquitin ligases, which facilitate the transfer of the small protein ubiquitin to thousands of substrate proteins; “tagging” with ubiquitin often promotes the degradation of the substrate by the proteasome. In this paper, we adapt a technique called Ligase Trapping for use in mammalian cells. Ligase Trapping is a highly accurate method for determining which substrates are targeted by a ubiquitin ligase. Here we use it to identify new substrates of the human cell cycle regulator βTRCP. Our screen was indeed highly accurate, as we were able to validate 88% of the candidate substrates we identified by mass spectrometry. Some of these new substrates were unstable proteins that were stabilized by inhibition of βTRCP, or of the entire class of ubiquitin ligases of which βTRCP is a part. However, others appear to be stable or redundantly-targeted substrates, which have been more difficult to identify with current techniques. This suggests that Ligase Trapping will be able to reliably identify new substrates of human ubiquitin ligases. Further, one of the new βTRCP substrates, CReP, is specifically depleted upon DNA damage, and depletion of CReP contributes to inactivation of the translational machinery upon DNA damage.
| E3 ubiquitin ligases, which facilitate the attachment of anywhere from one to a long chain of the small protein ubiquitin to substrate proteins, are important regulators of the cell cycle and the response to stress. The best-studied outcome of ubiquitination is destruction of the substrate by the proteasome. There has been a great deal of interest in the discovery of ubiquitin ligase substrates, with the recent introduction of techniques that either look for proteins whose levels change when a particular ubiquitin ligase is inhibited [1–5], or those that use mass spectrometry to look for proteins that interact physically with the ubiquitin ligase [6–11]. Unfortunately, some ligase-substrate interactions are likely too weak to purify by affinity. Moreover, once a list of associated proteins is identified, it is not always clear which are direct substrates. To address this, most studies have determined whether the half-life of the substrate is significantly altered upon inhibition of the ligase [11]. However, in many instances, only a select fraction of substrate is targeted. In addition, some substrates are targeted redundantly by multiple ligases [12]. These facts often make it impossible to verify candidates merely by examining their half-life. For ubiquitin ligases for which a consensus binding sequence is known, the presence of this sequence has been used frequently to separate true substrates from non-substrate or non-specific interactors. However, this method is not useful to discover substrates of the vast majority of ubiquitin ligases, for which no consensus sequence is known. To eliminate these problems, we developed a technique called Ligase Trapping [13] (Fig 1A), in which an E3 ubiquitin ligase is fused to a ubiquitin-associated (UBA) domain. This mediates an extended interaction between the E3 ligase and its ubiquitinated substrates, allowing their co-immunoprecipitation. To distinguish between substrates and other associated proteins, this immunoprecipitate is subjected to a second purification for 6xHIS-ubiquitin under denaturing conditions. These purifications can be used both for substrate identification and as a diagnostic for candidate confirmation, in cases where the bulk level of a protein is stable.
The SCF is a cullin-RING ligase (CRL) containing 3 core catalytic subunits: the RING finger protein RBX1, the cullin CUL1 and the adaptor SKP1 [14–17]. This catalytic base associates with a substrate adaptor called an F box protein, of which humans encode at least 69. F box proteins are thought to recognize their substrates only after substrate modification, typically by phosphorylation [14,17]. Several of these F box proteins have been characterized due to their well-established roles as tumor suppressors and oncogenes. βTRCP[18] is an F box protein that turns over substrates to control the G2/M transition (e.g. WEE1 [19]/CDC25 [20,21]), as well as the response to DNA damage (e.g. CDC25 [20,21], claspin [7,22]).
In this paper, we establish ubiquitin ligase trapping in mammalian cells. Of the 28 candidates identified using this technique, 12 were well-established substrates [6,20,21,23–33]. For the 16 remaining candidates, we examined 14 and found that 11 of these confirmed. Thus, 23 of the 26 known/tested candidates, (88%) appear to be substrates, suggesting that Ligase Trapping is a robust discovery technique. Further characterization showed that turnover of one of the βTRCP substrates, CReP, is exacerbated by DNA damage. CReP is a protein phosphatase 1 (PP1) specificity subunit that counteracts the phosphorylation of eukaryotic initiation factor 2 alpha (eIF2α) on serine-51 [34], a stress-induced modification that inhibits translation initiation on most transcripts [35,36]. Inhibiting the turnover of CReP after DNA damage significantly reduced the accumulation of serine-51 phosphorylated eIF2α, and increased translation after DNA damage, suggesting that CReP turnover is an important mechanism by which DNA damage regulates translation.
To establish Ligase Trapping in human cells, we created a stable HEK293 line in which 6xHIS-ubiquitin is expressed upon treatment with doxycycline. In this cell line, tagged ubiquitin accounts for a significant portion of the total ubiquitin pool when cells are treated with doxycycline (S1A Fig). In yeast, fusion of F box proteins, via a 3xFlag linker, to the UBA of Dsk2 or the two tandem UBAs of Rad23, led to enhanced purification of nascent ubiquitinated F box protein substrates [13]. We fused the human F box protein βTRCP to the human homologs of these UBA-containing proteins, and found that the RAD23B fusion increased the poly-ubiquitinated species purified by the βTRCP fusion most strongly (S1B Fig). Accordingly, we made a stable cell line that expressed both doxycycline-inducible 6xHIS-Ub and a Ligase Trap consisting of βTRCP fused on its C-terminus to 3xFlag and the C-terminal UBAs of RAD23B.
To determine whether the βTRCP trap was functional, we expressed an epitope-tagged allele of the βTRCP substrate ATF4 in our stable cell line. We were able to immunoprecipitate poly-ubiquitinated ATF4 with the βTRCP trap, but not with the Ligase Traps of two unrelated F box proteins, FBXO24 and Fbw7 (Fig 1B). We obtained a similar result with β-catenin (S2 Fig). We also purified ubiquitinated forms of the Ligase Traps, which was unsurprising as many ubiquitin ligases are themselves ubiquitinated. We also purified substantial unmodified forms of the Ligase Traps. This is likely a result of the very large amount of IP loaded relative to input (5,000:1 for the 2nd step), which is necessary to see the very small percentage of substrate that is poly-ubiquitinated. Even in cases where the unmodified band is equal in intensity in the input and 2nd step IP, this represents only 0.02% IP background. This phenomenon also occurs frequently with unmodified substrates, while the relevant purification of poly-ubiquitinated substrates is highly specific to the relevant Ligase Trap. To examine further whether the purification of β-catenin was specific, we made a stable cell line identical to our βTRCP ligase trap line, but with a mutation in the WD40 domain of βTRCP predicted to prevent binding to β-catenin [37]. As expected, this mutant trap failed to purify polyubiquitinated β-catenin (Fig 1C), showing that β-catenin purification by βTRCP represents a specific interaction. To make certain that the βTRCP Ligase Trap didn’t simply bind all ubiquitinated proteins more efficiently, we made a similar stable cell line expressing Fbw7-3xFlag-RAD23. Poly-ubiquitinated forms of the known Fbw7 substrate MED13 [10] were preferentially precipitated with the Fbw7 Ligase Trap (Fig 1D).
Having established the functionality of the βTRCP ligase trap cell line, we performed a large-scale, two-step purification and identified ubiquitinated co-precipitating proteins by mass spectrometry. Before collection, we treated cells with the proteasome inhibitor MG132 for four hours, as we had shown that this treatment increases the amount of poly-ubiquitinated material purified by the βTRCP ligase trap (S1C Fig). We defined candidate βTRCP substrates as those proteins identified in at least two of three purifications of the βTRCP ligase trap, but not in any of the negative control purifications. Twenty-eight proteins met these criteria (Table 1). Of these, twelve were previously-validated βTRCP substrates, and many others had been shown to interact with βTRCP in previously published large data sets, but had not been individually examined to determine if they were substrates [4,8,11,38–40]. SUN2 was purified in a large-scale screen for βTRCP substrates, and shown to be stabilized by the proteasome inhibitor MG132 [39] while this manuscript was under review. In addition, several other known βTRCP substrates, such as ß-catenin [41–45], were selectively identified in the βTRCP purification, but as some peptides were also identified in control purifications, these did not meet the stringent criteria that we had chosen for this initial analysis (bottom of Table 1). The large fraction of previously-published substrates (43%) that we purified confirms that Ligase Trapping accurately identified true substrates.
We also purified substrates of Fbw7 using a Ligase Trap. The Fbw7 Ligase Trap was expressed at a low level, suggesting that this trap was less stable. However, the proteins pulled down most abundantly and specifically by the Fbw7 Ligase Trap were MED13 and MED13L, two members of the Mediator complex shown to be Fbw7 substrates in a recent screen[10] in which Fbw7 interactors were precipitated and identified by mass spectrometry. (Our purification of MED13 is shown in Fig 1D) In that screen, the entire 26-member Mediator complex was purified, and MED13 and MED13L had to be identified as the direct Fbw7 substrates by a combination of degron prediction and careful validation; we did not purify any other members of the Mediator complex.
Ligase Trapping also provided a method to validate candidates beyond simply examining substrate turnover. Ligase Trapping is able to show that a ubiquitinated substrate specifically purifies with a particular ligase even if the substrate is redundantly targeted by multiple ligases, or if only a small fraction of the substrate (such as that in a particular complex) is ubiquitinated. To fully assay the accuracy of the Ligase Trap technique, we decided to validate candidate βTRCP substrates. Out of fourteen of the previously unknown/unvalidated candidates that we examined, eleven showed specific purification of polyubiquitinated material by the βTRCP ligase trap (Table 1 and Figs 2, S4 and S5). This strongly suggested that these candidates are true substrates of βTRCP, and that this technique accurately identified substrates with low background and thus will be an efficient way of identifying and validating substrates of other ubiquitin ligases in the future. Two βTRCP candidate substrates were not examined due to technical difficulties.
In order to determine whether βTRCP could bind its candidate substrates in the absence of the UBA domains present in the Ligase Traps, we co-expressed Flag-tagged versions of these F box proteins in HEK293 cells with HA-tagged versions of a subset of their candidate substrates. In all cases, the substrate was purified more efficiently by its cognate ligase than by the negative control ligase (Fig 3A).
Because a common outcome of ubiquitination by the SCF is proteasomal degradation of the ubiquitinated protein, we assayed whether a subset of the candidate substrates were degraded in a way that depended on the cognate ligase. For five of the βTRCP candidate substrates, we co-transfected cells with DNA encoding tagged substrate, as well as a negative control plasmid or a plasmid expressing an shRNA targeting both paralogs of βTRCP, then inhibiting bulk protein translation with cycloheximide and assaying substrate levels. Although the knockdown we achieved was quite modest, three of the five substrates were significantly stabilized (Fig 3B). One, RASSF3, was not stabilized, suggesting either that it is a better βTRCP substrate than the others, or that it is targeted by other ubiquitin ligases. UBE4B is a stable protein. (Note that we detected UBE4B with a specific antibody against this protein, and did not ectopically express it, so its stability is unlikely to be an artifact.) It is possible that either only a small pool of the substrate was targeted, or that the outcome of ubiquitination of UBE4B is not proteasomal degradation.
Several commonly-used approaches identify ubiquitin ligase substrates as those proteins whose abundance is increased by inhibition of the relevant ligase. One key advantage of ligase trapping is that, in contrast to these techniques, it can identify substrates whose bulk turnover is not affected by inhibition of the ligase. To determine more universally which substrates were quantitatively targeted for degradation by βTRCP, we expressed tagged versions of the substrates, inhibited protein synthesis with cycloheximide, and followed the turnover of the substrate in the absence or presence of MLN4924 (Table 1 and S6 Fig). Of the ten substrates examined, three (CReP, ZNF395, and SUN2) were unstable proteins that were stabilized by MLN4924, suggesting that their turnover is mediated by βTRCP alone or in combination with other cullin-RING ligases. (CReP was previously shown to be an unstable protein [34], as was SUN2.) Four (ZNF704, FNIP, RASSF3 and AEBP2) were not or only partially stabilized by MLN4924, suggesting that these might be redundantly targeted by βTRCP and a non-CRL ligase. Three proteins (HIVEP2, UBE4B, and TRIM9) appeared to be constitutively stable, although we cannot rule out that overexpression or epitope tagging of HIVEP2 and TRIM9 led to an artifactual stabilization. βTRCP could be promoting non-degradative ubiquitination of these substrates, or may only ubiquitinate a specific pool.
We were initially concerned that treating cells with MG132 would lead to increased background, or skewing of the results. Therefore, we performed two purifications of the βTRCP ligase trap in the absence of MG132. This purification generated a list with several of the same substrates, but lacking a subset, especially those shown to be unstable in Figs 3B and S6 (S7 Fig). In addition, all of our validations were performed in the absence of MG132 (Figs 2, S4 and S5).
We wished to further explore the biological significance of CReP turnover. First, we verified that the ubiquitinated CReP pulled down by the βTRCP ligase trap required SCF activity. Indeed, pre-treatment of cells with MLN4924 eliminated the ubiquitinated CReP (but not unmodified CReP) pulled down by the βTRCP ligase trap (Fig 4A). Second, we mutated CReP’s single well-conserved βTRCP-binding consensus, as well as the amino acids immediately downstream, which form a second less-well-conserved consensus. The βTRCP consensus is DpSGX(1–4)pS [46], with some substitution of acidic amino acids for phosphorylations tolerated. The sequence we mutated in CReP is DDGFDSDSSLSDSD (marked in S11 Fig). Although this sequence lacks the most-conserved DSG motif, many well-documented βTRCP substrates have variations in this part of the degron [18], and human CDC25A and CDC25B have well-validated degrons that contain DDG, just like CReP [25] (shown in Fig 4B). This mutant, CReP11A, was significantly stabilized relative to wild type CReP (Fig 4C and 4D), strongly suggesting that CReP turnover is dependent on βTRCP. The notable downshift of the mutant is likely due to mutation of several negatively-charged residues. Mutation of a portion of the same region was independently shown to stabilize CReP while our manuscript was in the review process[47].
Because both protein-folding stress and DNA damage have been shown to regulate eIF2α phosphorylation, we tested whether these stresses also regulated CReP levels. The proteostatic stress inducer thapsigargin had a very minor effect on CReP levels, consistent with a previous report showing no effect [34]. However, DNA damage provoked by either ultraviolet light (UV) or the topoisomerase inhibitor camptothecin (CPT) led to complete depletion of CReP (Fig 5A). Suggestively, the disappearance of CReP was coincident with the induction of eIF2α phosphorylation by these stressors. The depletion of CReP was not due merely to inhibition of translation by eIF2α phosphorylation, as DNA damage also decreases the half-life of CReP compared to no treatment or treatment with proteostatic stressors in a cycloheximide chase (S11 Fig), and CReP still disappears upon DNA damage in mouse embryonic fibroblasts in which Ser51 of eIF2α has been mutated to alanine (data not shown). CReP turnover and subsequent eIF2α phosphorylation is at least partially dependent on βTRCP, as transfection with shRNA against both paralogs of this ligase delays DNA damage-dependent induction of both CReP turnover and eIF2α phosphorylation (Fig 5B). CReP depletion is fully dependent on CRL-mediated degradation, because treatment of cells with the CRL inhibitor MLN4924 prevents CReP depletion (Fig 5C). The residual CReP turnover seen even in cells treated with βTRCP shRNA may reflect our inability to achieve sufficient knockdown of βTRCP, or additional turnover mediated by another CRL. Cullin-mediated turnover of CReP in response to DNA damage was not restricted to HEK293 cells, since it occurs in both primary human fibroblasts (Fig 5D) and immortalized mouse embryonic fibroblasts (MEFs) (S12 Fig).
The CReP11A mutant was not completely stabilized upon DNA damage (data not shown), possibly because DNA damage promotes βTRCP binding to additional sites on CReP. βTRCP has been shown to interact with non-consensus phosphodegrons in MDM2, suggesting that it may be difficult to identify degrons by sequence alone[48]. Therefore, we mapped phosphorylated residues on CReP to identify any additional degron sequences (S9 Fig). Notably, most phosphosites were observed both with and without CPT. It is possible that the increase in CReP turnover observed upon DNA damage is not due to increased phosphorylation, but to a change in a targeting factor or localization of CReP. However, phosphosites are still likely to be required for turnover. For clustered phosphosites and phosphosites that were near short acidic stretches, we mutated both the phospho-acceptor and all acidic and potential phospho-acceptors in the region. In addition, we mutated one additional weak βTRCP consensus site that was not covered in the phospho-mapping. We then tested the stability of these mutants, in various combinations, in DNA damage (data not shown). CReP31A (S10 Fig) was the least mutated allele that was completely stable upon treatment with DNA damage (Fig 5E and 5F). Importantly, this stabilization was not merely an artifact of high starting levels resulting from prioritized transcription or translation, as CReP31A is stable even upon pre-treatment with camptothecin followed by cycloheximide chase (Fig 5E). Like the 11A mutant, CReP31A migrates much more quickly than the endogenous protein, likely due to mutation of many negatively-charged amino acids.
To examine the physiologic role of the turnover of CReP upon DNA damage, we determined whether CReP stabilization had an effect on eIF2α phosphorylation. When CReP turnover was inhibited by knockdown of βTRCP, treatment with MLN4924, or mutation of CReP, phosphorylation of eIF2α was delayed or inhibited to an equivalent degree (Fig 5B, 5C and 5F). This is not specific to HEK293 cells, as MLN4924 also reduced eIF2α phosphorylation after UV treatment in immortalized mouse embryonic fibroblasts (MEFs) (S12 Fig). However, primary human fibroblasts (Fig 6D) had constitutively high levels of eIF2α phosphorylation, so the effect of CReP turnover was only subtle. This may reflect a greater need for this pathway in fast-growing cells, or the fact that these primary cells were under constant stress.
Upon proteostatic stress, eIF2α phosphorylation promotes the translation of the transcription factor ATF4 [49]. ATF4 activates the expression of the transcription factor CHOP [49], which in turn promotes the transcription of GADD34 [50]. Like CReP, GADD34 is a PP1 targeting subunit that acts on Ser51 of eIF2α [51,52]. These PP1 subunits appear to have a dedicated role in regulating eIF2α, since the lethal phenotype of knockout mice lacking both GADD34 and CReP can be rescued by mutating eIF2α Ser51 [51]. Previous reports suggested that GADD34 is induced at late time points after DNA damage in some cell types [53]. We were especially interested in whether DNA damage promoted the destruction of CReP only to replace it with GADD34. However, we found that UV treatment did not promote GADD34 protein expression, while ER stress induced by thapsigargin did (Fig 6A). This may reflect a cell-type difference between HEK293 cells and cells previously used to show GADD34 induction. Surprisingly, treating cells with UV and thapsigargin simultaneously blocked the thapsigargin-mediated increase in GADD34 protein levels, suggesting that DNA damage somehow dominantly prevents expression of this protein. Inhibition of GADD34 expression by UV treatment could be rescued by simultaneously treating cells with MLN4924, suggesting that a CRL is involved in blocking GADD34 accumulation.
Finally, we examined whether CReP turnover after DNA damage affected rates of translation. After treatment with DNA damage, translation rate was assayed via the SUnSET method [54], by adding puromycin to the cells for 10 minutes, then detecting the degree of puromycin incorporation into newly translating polypeptides via western blotting with an anti-puromycin antibody. We found that expression of CReP31A, which led to high CReP levels even after treatment with camptothecin and initial recovery from this damage, accelerated the recovery of translation after DNA damage, doubling the translation rate at 2 hours after CPT washout (Fig 6B and 6C). Notably, this effect was not seen with the unstable, ectopically expressed wildtype CReP, although it was expressed at the same level as CReP31A. This effect reproduced several times, although the exact timing varies, likely due to subtle variations in CReP expression levels during transfection.
We have identified and validated thirteen novel substrates of the well-studied ubiquitin ligase βTRCP via Ubiquitin Ligase Trapping. While we were unable to test two of the twenty-eight candidate substrates identified, 88% of the remaining twenty-six were either known or validated novel substrates. While affinity chromatography is often able to identify ligase substrates, these data suggest that Ligase Trapping provides an unprecedented hit rate, making it an especially efficient way to identify new ubiquitin ligase substrates. Moreover, this technology has allowed us to easily validate substrates even if their bulk stability is not affected by βTRCP ubiquitination.
Our results for FBW7 suggest another way in which Ligase Trapping can complement currently available techniques. In a previous study, the Clurman lab pulled out all 26 members of the Mediator complex with FBW7. They used degron prediction and follow-up experiments to identify MED13 and MED13L as the ubiquitylated Fbw7 substrates and carefully confirmed that they are direct substrates. Our mass spec of the Fbw7 ligase trap immunoprecipitation specifically purified MED13 (and MED13L) uniquely in the Fbw7 Ligase Trap, and not in any of the other purifications. Moreover, we pulled out none of the other 25 subunits. This underscores the usefulness of our technique, especially for the great majority of F box proteins for which no degron consensus is known. Thus, even in cases where Ligase Trapping identifies similar numbers of substrates compared to other techniques, it allows one to quickly identify the directly ubiquitylated substrates.
In addition to the substrate CReP, which we followed up in detail, turnover of several of the other substrates is likely to be regulated in response to cell cycle position or stress. Sun2 is a transmembrane protein that spans the inner nuclear envelope and has been implicated in the maintenance of nuclear structure and the regulation of DNA damage. Its turnover by βTRCP may regulate these processes, and its removal from the membrane after ubiquitination may also be a regulated step. Strikingly, four of the eleven novel substrates we validated, ZNF395, HIVEP1/2, ZNF704, and AEBP2, are transcription factors, as are several known βTRCP substrates, such as Nrf2 and ATF4. We also identified two substrates that are themselves ubiquitin ligases, UBE4B and TRIM9, which opens up the possibility of complex mutual regulation. While UBE4B ubiquitination depends on the SCF (data not shown), it is not highly ubiquitinated (Fig 2), and it appears that the majority of the UBE4B population is stable (Fig 3B). RASSF3 is a candidate tumor suppressor protein that activates p53-dependent apoptosis under appropriate conditions, including DNA damage [55]. Its regulation by βTRCP is consistent with the known role of βTRCP in responding to DNA damage, and may help explain the oncogenic effect of βTRCP overexpression [18] (along with other known tumor suppressor substrates of βTRCP, such as REST[45]). RASSF3 appears to have both stable and unstable pools. This may reflect the relatively small pool of cells undergoing stress at any particular time in an untreated culture. Perturbations such as DNA damage might drive RASSF3 turnover.
Our previous studies in yeast [13] showed that 56% of newly-identified SCF substrates were strongly stabilized when the F box in question was mutated. 25% showed small or moderate stabilization, but were still unstable in the F box gene mutant. Finally, 19% appeared stable even in wildtype. We find here that 45% of confirmed novel substrates were stabilized by treatment with a pan-CRL inhibitor, 18% showed no stabilization, and 27% were stable in wildtype. Thus, in both cases only half or fewer novel substrates were quantitatively turned over by the single ligase, although this is likely an underestimate overall, since previously characterized substrates may be biased for this category. While some of these effects could be due to the population assay employed, as noted above, substrates such as Cln3 and Gal4 in yeast, as well as PIP box-containing substrates in humans, are targeted in a way that is dependent upon the sub-cellular localization/context of the substrate [12,56]. Alternatively, some turnover events occur as part of quality control pathways that only target those proteins that are in some way defective.
We have implicated βTRCP in the regulation of translation initiation after DNA damage through its turnover of CReP, and shown that DNA damage-induced phosphorylation of eIF2α, because it uniquely requires the depletion of CReP, occurs via a different mechanism from the other stresses known to promote eIF2α phosphorylation, which all promote kinase activation. Previous work has shown that the phosphorylation of eIF2α after UV treatment depends on the kinase Gcn2 [57,58]. We propose that this phosphorylation requires both Gcn2 activation and CReP turnover.
Why does phosphorylation of eIF2α require CReP depletion after DNA damage, but not in response to proteostatic stress? One possibility is that eIF2alpha kinases are less active after DNA damage than after proteostatic stress. We observed that, once CReP levels begin to drop, eIF2α phosphorylation is much higher upon our UV treatment than after proteostatic stress (Fig 5A). This likely reflects both continued CReP activity and the induction of GADD34 upon proteostatic stress. We showed in Fig 6B and 6C that CReP turnover has a significant effect on translation rates after DNA damage, but substantial inhibition of translation happens even in the absence of CReP turnover. Translation rates are highly redundantly regulated, both via control of eIF2α phosphorylation and via regulation of eIF4. Our results are consistent with a model in which CReP turnover is important to enforce continued low levels of translation at later timepoints. Moreover, the high levels of eIF2α phosphorylation enabled by CReP turnover in response to DNA damage may allow translational reprograming that leads to induction of DNA damage repair proteins, even as global translation is downregulated. Indeed, translation of several DNA repair proteins has been shown to be resistant to inhibition of CAP-dependent translational inhibition by eIF2α phosphorylation [58].
Finally, how do CRLs prevent the induction of GADD34 after UV treatment? One possibility is that CReP turnover upon DNA damage (which requires CRLs) drives such strong eIF2α phosphorylation that translation of GADD34 or one of its upstream regulators ATF4 or CHOP is inhibited. Another possibility is that a CRL is turning over a specific protein to keep GADD34 levels low. βTRCP is known to target ATF4 [24] and the Cul3-associated ligase SPOP is reported to target CHOP [59]. GADD34 is also a known proteasome target, consistent with its being a substrate of βTRCP or another CRL [60]. Targeting of both CReP and Gadd34 for degradation upon DNA damage underscores the importance of limiting eIF2α phosphatase activity during DNA damage.
All plasmids were transfected into the 293 FlpIn TRex cell line (Life Technologies, Grand Island, NY, USA), which contains both a site for FRT-mediated recombination (which we did not use in this work) and expresses the tet repressor, which allows doxycycline-inducible expression from promoters that include tet operators. Mouse embryonic fibroblasts (MEFs) were immortalized by transduction with the SV40 large T antigen (kind gift of Morgan Truitt and Davide Ruggero). All cells were grown in DMEM with 10% heat-inactivated fetal bovine serum. For large-scale purifications, medium was supplemented with 500 U/mL penicillin and 500 μg/mL streptomycin.
6xHis-ubiquitin was expressed from pTB30, a modified pcDNA3.1 vector with a pCMV/TetO promoter expressing 6xHis-Uba52-IRES-6xHis-RPS27A. The parent of this construct was the kind gift of Zhijian Chen, UT Southwestern. The construct was linearized with Pvu I and transfected into 293 FlpIn TRex cells. Stable transfectants were selected with G418 and a clone was selected that expressed at a high level only upon treatment with doxycycline.
To make the ligase trap fusion proteins, F box proteins were fused on the C-terminus to 3xFlag followed by the C terminal half of human RAD23B (Accession #BC020973.2, amino acids 185–410), encoding two UBA domains. Ligase traps βTRCP2 (FBXW11; Accession #BC026213.1, pTB53), Fbxo24 (Accession #NM033506.2, pBEN20), and Fbxo6 (Accession #NM018438.5, pBEN5) were expressed as hygromycin resistance-T2A-ligase trap fusions driven by the mouse PGK1 promoter. Each of these constructs also expresses an shRNA against the relevant F box protein (to which the fusion protein is resistant), driven by the mouse U6 promoter. These cassettes were linearized by digestion with Pac I. Fbw7 (Accession# NM_033632.3, pTB59) Ligase Trap was expressed from a pcDNA3.1 vector, under the control of the CMV promoter. The vector was linearized with BglII. All linearized plasmids were transfected into the HisUb cell line and stable transfectants were selected with hygromycin. We selected clonal cell lines that expressed moderate levels of the relevant ligase trap.
All substrate proteins were tagged on the N-terminus with the 5xHA epitope, and expressed from the CMV promoter in pcDNA3.1, except SUN2, AEBP2, ALDH2, and RASSF3, which were tagged on the C-terminus. They were transiently transfected into the relevant cell line using Fugene HD at 3 μL/μg DNA (Promega Corporation, Madison, WI, USA) or polyethyleneimine (at 18 μg/μg DNA) 24–48 hours before the experiment. βTRCP was knocked down with an shRNA targeting both BTRC and FBXW11, expressed from the pSUPER-puro-retro vector (under the H1 promoter)[61].
MG132 is used at 5 μM. MLN4924 is used at 1 μM. Camptothecin is used at 3 μg/mL, unless otherwise specified.
Medium was removed from adherent cells and set aside. Cells were covered in 37°C 1X PBS with 0.9 mM CaCl2 and 0.5 mM MgCl2, then exposed to 300 J/m2 UV-C, PBS was aspirated, and medium was replaced.
For western blotting, cells were lysed in 1X RIPA buffer with protease and phosphatase inhibitors for 30 minutes on ice, insoluble material was spun out, then protein concentrations were measured with BCA Reagent (Pierce, Thermo Scientific, Rockford, IL, USA) and normalized before addition of SDS sample buffer with DTT. For Figs S7 (except for RASSF3) and 5C, cells were lysed directly in SDS sample buffer with DTT or βMe.
All gels were Criterion Tris-HCl 4–20% gradients (cat. #345–0034, BioRad, Hercules, CA, USA), except for the gel for the α-HA blot in Fig 2C, which was a 7.5% gel (BioRad cat. #345–0007).
Antibodies used were α-HA 16B12 at 1:1,000–1:2,000 (cat. #MMS-101R, Covance, Emeryville, CA, USA), α-6xHis at 1:1,000–1:2,000, α-ubiquitin P4D1 at 1:100, α-Flag M2 at 1:2,000 (cat. #F3165, Sigma, St. Louis, MO, USA), α-Cul1 at 1:1,000, α-vinculin at 1:1,000–1:5,000, α-βactin at 1:1,000–1:10,000 (Sigma cat. #A5441 for Fig 4A, Abcam, Cambridge, UK, cat.#ab8226 for all others), α-PPP1R15B (CReP) at 1:1,000–1:5,000 (cat. #14634-1-AP, Proteintech, Chicago, IL, USA), and α-GADD34 (cat. # 10449-1-AP, Proteintech, Chicago, IL, USA). α-phosphoS51-eIF2α (cat. #9721), α-eIF2α (cat. #9722), α-phosphoS317Chk1 (cat. #2344), and α-Chk1 (cat. #2360) antibodies were all from Cell Signaling Technologies, Danvers, MA, USA. The α-puromycin antibody 12D10 was from EMD Millipore (cat. #MABE343).
Western blots in Figs 1, 2A, 2B and 3A were incubated with secondary antibodies fused to horseradish peroxidase and visualized by treatment with Western Lightning ECL (Perkin Elmer, Waltham, MA, USA). Western blots in Figs 2C, 3B and 4 were incubated with fluorescent secondary antibodies and visualized with an Odyssey scanner (Licor, Lincoln, NE, USA).
Unless otherwise noted, stable cell lines expressing Ligase Traps were treated with 5 μM MG132 for 4 hours before collection. We grew 100–200 barely sub-confluent 15 cm dishes for each purification, representing approximately 1–3 x 109 cells. Pellets were lysed in 25 mM Hepes-KOH, pH8, 150 mM K Oac,10 mM MgCl2, 5 mM CaCl2, 20 mM iodoacetamide, 30 μM MG132, protease inhibitors, and phosphatase inhibitors by sonication, then treated with DNase (660 U/mL) at 4°C for 30 minutes before addition of Nonidet P-40 to 0.1%. Samples were spun to remove insoluble material, then incubated with α-Flag M2 magnetic beads (Sigma, St. Louis, MO, USA) at 4°C overnight. Beads were washed 5 times in 1X PBS+0.1% Nonidet P-40, then eluted in this wash buffer+0.5 mg/mL 3xFlag peptide. The eluate was denatured by addition of 2X volume Buffer B (216 mM NaH2PO4, 16 mM Tris, 9.37 M urea, pHed to 8). The sample was then incubated with NiNTA agarose for 3 hours at room temperature. The beads were washed 3X in Buffer B diluted to 8M urea+10 mM imidazole, then 2X in Buffer B diluted to 1 M urea+10mM imidazole. Samples were eluted in 0.5 M urea, 300 mM imidazole, 0.1% rapigest (or Nonidet P-40 if not to be used for mass spectrometry), 108 mM NaH2PO4, 8 mM Tris (pHed to 8 before adding imidazole).
The immunopurified protein complexes were mixed in a ratio of 1:1 with digestion buffer (100 mM Tris-HCl, pH 8.5, 8M urea), reduced, alkylated and digested by sequential addition of lys-C and trypsin proteases as previously described[62,63]. For identification of phosphorylation site, proteins were digested directly in the excised gel slice using trypsin[62]. Peptide digests desalted and fractionated online using a 50 μM inner diameter fritted fused silica capillary column with a 5 μM pulled electrospray tip and packed in-house with 15 cm of Luna C18(2) 3 μM reversed phase particles. The gradient was delivered by an easy-nLC 1000 ultra high pressure chromatography system (Thermo Scientific). MS/MS spectra were collected on a Q-Exactive mass spectrometer (Thermo Scientific) [64,65]. Data analysis was performed using the ProLuCID, DTASelect2, and Ascore algorithms as implemented in the Integrated Proteomics Pipeline—IP2 (Integrated Proteomics Applications, Inc., San Diego, CA) [66–69]. Phosphopeptides were identified using a differential modification search that considered a mass shift of +79.9663 on serines, threonines and tyrosines. Protein and peptide identifications were filtered using DTASelect and required at least two unique peptides per protein and a peptide-level false positive rate of less than 5% as estimated by a decoy database strategy[70]. Normalized spectral abundance factor (NSAF) values were calculated as described and multiplied by 105 to improve readability [71].
We followed the SUnSET protocol [54]. Puromycin was added to culture medium at a final concentration of 10 μg/mL, incubated for 10 minutes at 37°C and 8% CO2, then medium was replaced with ice-cold PBS with 5 mM EDTA, and cells were sprayed from the dish on ice, spun down at 4°C and flash-frozen. Samples were normalized by protein concentration, and puromycin incorporation was detected by western blotting with a monoclonal anti-puromycin antibody (12D10) and quantified by densitometry.
|
10.1371/journal.ppat.1002750 | Molecular and Electrophysiological Characterization of a Novel Cation Channel of Trypanosoma cruzi | We report the identification, functional expression, purification, reconstitution and electrophysiological characterization of a novel cation channel (TcCat) from Trypanosoma cruzi, the etiologic agent of Chagas disease. This channel is potassium permeable and shows inward rectification in the presence of magnesium. Western blot analyses with specific antibodies indicated that the protein is expressed in the three main life cycle stages of the parasite. Surprisingly, the parasites have the unprecedented ability to rapidly change the localization of the channel when they are exposed to different environmental stresses. TcCat rapidly translocates to the tip of the flagellum when trypomastigotes are submitted to acidic pH, to the plasma membrane when epimastigotes are submitted to hyperosmotic stress, and to the cell surface when amastigotes are released to the extracellular medium. Pharmacological block of TcCat activity also resulted in alterations in the trypomastigotes ability to respond to hyperosmotic stress. We also demonstrate the feasibility of purifying and reconstituting a functional ion channel from T. cruzi after recombinant expression in bacteria. The peculiar characteristics of TcCat could be important for the development of specific inhibitors with therapeutic potential against trypanosomes.
| The use of high-resolution electrophysiological techniques to study ion channels has provided a large amount of information on functional aspects of these important membrane proteins. However, the study of ion channels in unicellular eukaryotes has been limited to detection of ion conductances in large cells, gene identification studies, and pharmacological treatments to investigate the potential presence of different ion channels. In this paper we report the first identification, functional expression, purification, reconstitution, and electrophysiological characterization with single-molecule resolution of a novel cation channel (TcCat) from Trypanosoma cruzi. This is a novel channel that shares little sequence and functional similarities to other potassium channels and its peculiar characteristics could be important for the development of specific inhibitors with therapeutic potential against trypanosomiasis. Surprisingly, the parasites have the unprecedented ability to rapidly change the localization of the channel when they are exposed to different environmental stresses. We demonstrated the feasibility of purifying and reconstituting a functional ion channel from T. cruzi after recombinant expression in bacteria. In addition, we obtained yeast mutants that will provide a useful genetic system for studies of the assembly and composition of the channel.
| Trypanosoma cruzi is a unicellular parasitic eukaryote and the etiologic agent of Chagas disease, which currently affects millions of people in North, Central and South America, and is becoming frequently diagnosed in non-endemic countries [1], [2].
T. cruzi has a complex life cycle involving insect and mammalian hosts and different morphological and functional stages: epimastigotes and metacyclictrypomastigotes in the insect vector, and intracellular amastigotes and bloodstream trypomastigotes in the mammalian host. During its life cycle, the parasite finds extreme fluctuations in environmental conditions to which it must adapt in order to survive. A wide range of ionic concentrations, osmolarities and pHs are major challenges to cope with when it transits through the vector gut to the excreta, and from this highly concentrated environment to the interstitial fluid of the mammalian host. Particularly, the concentration of K+in the vector can vary between 40 to 358 mM depending on the feeding cycles of the insect [3], and from 5 to 140 mM between the extra and intracellular environments of the mammalian stages.
In previous studies [4], [5] we demonstrated that a plasma membrane H+-ATPase is the major regulator of intracellular pH (pHi) in all stages of T. cruzi. However, in contrast to epimastigotes, whose pHi is not affected by extracellular cations [4], trypomastigotes possess a cation-dependent pHi control. We proposed [5] that, as occurs in plants [6], [7] and other protists [8], in these trypomastigote stages an inward rectifier K+ channel functions in K+ uptake dissipating the plasma membrane potential(Vm) generated by the H+-ATPase thereby increasing its efficiency. This putative channel could be blocked bythe addition of Cs2+ or Ba2+ [5].The plasma membrane H+-ATPase also plays a significant role in the regulation of Vm in all stages of T. cruzi [9]. In contrast to epimastigotes the Vm of trypomastigotes is markedly sensitive to extracellular Na+ and K+. In support of the presence of a K+permeable channel, the Vm is hyperpolarized by K+-free buffer in trypomastigotes [9]. Interestingly, trypomastigotes are able to maintain a negative Vm in a K+-rich buffer at acidic pH, conditions that they encounter when they enter the parasitophorous vacuole [9]. This is differentfrommammalian cells, which are usually depolarized by either acidic or high extracellular K+ concentrations. Amastigotes, in contrast, appear to be impermeable to K+ in agreement with the high intracellular K+ environment in which they live [9]. The marked differences in the regulation of Vμ in trypomastigotes as compared to amastigotes suggest that during transformation to amastigotes, trypomastigotes undergo significant changes in their ion transport mechanisms. However, the nature of these changes and the molecular identity of K+permeable pathways are unknown.
K+ channels are members of one of the largest and most diverse families of membrane proteins, widely described from bacteria to humans [10]–[12]. Their roles include plasma membrane potential maintenance, pHi and cell volume regulation, excitability, organogenesis and cell death [13]–[16]. From the structural point of view, they can be divided into two main groups: channels containing six transmembrane domains, including in this category voltage-dependent K+ channels and calcium-activated K+ channels [17], and channels with only two transmembrane domains, such as the inward rectifier K+channels (Kir channels) [18] and the widely described bacterial channel KcsA [10]. As a general rule, a functional K+ channel is formed by interaction of four pore-forming subunits interacting through a conserved tetramerization domain. Association with other proteins, interaction with surrounding lipids and post-translational modifications generate a functional diversity that exceeds the predictions based solely on the number of identified genes [17].
High yield recombinant expression and purification of functional ion channels has been technically very difficult and restricted to prokaryotic channels until recently [19]. In this work we demonstrate the feasibility of purifying a functional cation channel from T. cruzi after recombinant expression in bacteria. We report the molecular and electrophysiological characteristics of this inwardly rectifying K+permeable channel and the changes occurring in its localization during the parasitetransformation into different developmental stages. Our results indicate that T. cruzi has the unexpected ability to change the localization of this cation channel to adapt to different environments to which it is exposed in its different developmental stages.
We searched for K+channels in the TriTryp database (http://tritrypdb.org/tritrypdb/) and found two genes encoding for putative voltage-dependent K+channels in T. cruzi (Tc00.1047053511301.140 and Tc00.1047053507213.30). The sequences showed 98% identity between them and likely correspond to alleles of the same gene (TcCat). The orthologous identified in T. brucei (Tb927.10.16170) and L. major (LmjF19.1620) shared 64% and 55% amino acid identity respectively (Fig. S1A). Structural analysis (TopPred) (Text S1)predicted two transmembrane domains between amino acids 77–97 and 169–189 and a tetramerization domain at position 5–73 (Pfam02214) that is the only region with similarity to other K+channels like Kv4.3 (Fig. S1B). The ORF predicts a 297 amino acid protein with an apparent molecular weight of 34 kDa. No significant identity was found with well-characterized bacterial channels like KcsA or with mammalian (Kir channels)and bacterial inward-rectifiers (Fig. S1C). Interestingly, no conserved K+ channel signature sequence [T-X-G-Y(F)-G] [20] was identified in TcCat, raising the question of the ion selectivity of this channel. Other important features of TcCat are the presence of longer mode 2 interacting phospho-motif for 14-3-3 proteins at positions 128–134 (RHALTIT), putative phosphorylation sites at serines 103, 190, 214 and 248 and N-glycosylation sites at positions 181 (NGTA), 228 (NFTF) and 286 (NSTR), that can be relevant for the regulation of the activity and the interaction with other proteins.
TcCat localization was analysed by indirect immunofluorescence using affinity-purified antibodies against the recombinant protein. In trypomastigotes, the channel has a clearly defined punctuate pattern along the flagellum (Fig. 1A). In epimastigotes (Fig. 1B), TcCat also has a peripheral punctuated pattern with some apparently intracellular labeling. To further evaluate whether the punctuate localization could be due to labeling of patches of plasma membrane and not intracellular vesicles we performed immunolocalization in permeabilized and non-permeabilized cells. In both trypomastigotes and epimastigotes TcCat was detected, at least in part, exposed to the cell surface (Fig. S2). In amastigotes that were spontaneously released to the supernatant of infected L6E9 myoblasts the channel showed plasma membrane localization (Fig. 1C). However, in intracellular amastigotes TcKCat seems to be confined to a spot that could be the remaining short flagellum (Fig. 1D). This change in localization is consistent with a role of TcCat in K+ uptake, which would become less important in the intracellular environment rich in K+. In agreement with a developmental regulation of TcCat expression, labeling decreased considerably in metacyclictrypomastigotes (Fig. 1E). Immunoelectron microscopy analysis confirmed the patched distribution of TcCat in tissue culture-derived trypomastigotes along the flagellar attachment zone (Figs. 1F and 1G). The association of ion channels in clusters has been described previously [21]–[23] and seems to be related with preferential targeting to specific membrane lipid microdomains or lipid rafts, which are known to be more abundant in the flagellar membrane of trypanosomes [24].
Expression of TcCat was verified by western blot analysis (Fig. 1H) confirming the presence of the channel in all three stages of the parasite. The native protein detected in the parasites has an apparent molecular mass of 43 kDa, slightly higher than that predicted by the ORF. This difference could be due to post-translational modifications as can be expected from the presence of several putative phosphorylation and N-glycosylation sites. Densitometry analysis using α-tubulin as a loading control as well as Coomassie blue staining (Fig. S3) indicated that the level of expression is similar in trypomastigotes, amastigotes and epimastigotes (Fig. 1I), as suggested by IFA.
We evaluated the change in the localization of TcCat by IFA during the differentiation in vivo. At 5 h post-infection of mammalian cells, TcCat is already detected at a single intracellular spot both in parasites with trypomastigote-like morphology (Fig. 2A, yellow arrows) and in rounded amastigote-like cells (Fig. 2B, red arrows). At 24 and 48 h post-infection (Figs. 2C and 2D) TcCat remains intracellular in the replicating amastigotes, close to the flagellar pocket. In extracellular trypomastigotes, 96 h post-infection, TcCat was always localized at the plasma membrane (Fig. 1A).
We studied the expression of TcCat during the differentiation of trypomastigotes to amastigotes in vitro. At 30 min after induction of differentiation in vitro at pH 5.0, staining with antibodies against TcCat was at the tip of the flagellum (Fig. 2E), and this single spot labeling was maintained even when the cells rounded up to transform into amastigotes (Figs. 2F–H).
Potassium uptake defective S. cerevisiae mutants (Δtrk1, Δtrk2, Δtok1) were used to investigate the K+ influx ability of TcCat (see Text S1). These mutants depend on high extracellular K+ concentration for their growth as they only have the non-specific cation uptake mechanism, termed NSC1, for growth [25]. Mutants were kept in defined medium (SC ura-) supplemented with 100 mMKCl, pH 5.8. TcCat expression was induced by switching the carbon source and the channel was rapidly detected on the yeast surface by immunofluorescence analysis with anti-TcCat antibodies (Fig. 3B). After 2 h (Fig. 3C), yeasts were collected by centrifugation and placed in standard SC ura-medium without KCl. Otherwise the presence of high K+ concentration was toxic upon induction of TcCat expression. The channel was expressed on the yeast surface for up to 72 h at high levels, although, at 24 h, some labeling could be observed in the periphery of the yeast vacuole, probably due to recycling or degradation (Fig. 3D). Control cells were transformed with empty vector pYES2. A monoclonal antibody against the 69-kDa subunit of the vacuolar H+-ATPase was used as a control of proper permeabilization (Fig. 3A–D).
TcCat expression in complemented yeasts was also verified by western blot analysis using anti-TcCat antibodies (Fig. 3E). Two bands were detected, one that corresponds to the predicted molecular weight of the protein product (about 35 kDa) and a second band of approximately 45 kDa similar to that of the native protein in the parasites (Fig. 1H), suggesting that post-translational modifications also occurred in yeasts.
Functional complementation and restoration of the normal growth phenotype was achieved when culturing the mutant yeast in serial dilutions in SC ura-galactose agar plates without addition of KCl. Under these conditions, mutants transfected with vector alone (MpYES2) were not able to grow when diluted to 10−1 or more, while mutants complemented with TcCat(MC) showed no significant difference in growth as compared with wild type yeast (WT) (Fig. 3F). The results indicate that TcCat is indeed a K+ conductive pathway able to functionally complement a heterologous system.
The activity of TcCat was detected in patches excised from cell-size giant liposomes (inside-out configuration) containing the purified recombinant protein (Text S1, Figs. S4 and S5, and Table S1).Currents from liposomes containing only asolectin were recorded as control (Fig. S6A,black squares), showing a significant lower level compared with currents from liposomes containing purified TcCat (Fig. S6A, red circles). Currents were recorded under symmetrical conditions in the absence of Mg2+, unless stated otherwise, with bath and pipette solutions containing 140 mMKCl, 10 mMHepes-K, pH 7.4. Single channel currents were observed when an increasing voltage-pulse protocol between −80 to +80 mV was applied (Fig. 4A). The current-potential relationship for the single channel was not linear in the presence of Mg2+, as expected for an inward rectifier channel. The chord conductance (γ) (Fig. 4B, open circles) calculated under symmetrical KCl in the absence of Mg2+was 77±4 pS and 59±2 pS at −80 and +80 mV, respectively (n = 14) indicating a slight intrinsic rectification. Although no significant reduction in the current was observed at positive potentials in the presence of Mg2+, a significant increase in the inward current was evident at negative potentials in the presence of 1 mM MgCl2 in the bath solution (Fig. 4B, black squares), with unitary conductances of 122±7 pS and 56±3 pS at −80 and +80 mV, respectively (n = 13). These results suggest that the mechanism of blockage by Mg2+ is different from the one described for inward rectifier K+ channels. The unitary level of current was frequently observed in clusters, as shown in Fig. 4C where at least two channels could be detected, opening and closing independently. The histograms showncorrespond to the unitary current of one or two channels at the indicated voltages (Fig. 4C). This recorded activity agrees well with the localization in patches described above. Important variations in the open probability were observed in recordings from different days. When 14 independent experiments were analyzed, the open probability was not significantly sensitive to voltage, with values of 0.26±0.04 and 0.2±0.04 at −80 and +80 mV, respectively (Fig. 4D).
The cationic nature of the TcCat conductive properties was verified applying a voltage-ramp protocol from −80 to +80 mV under symmetrical conditions(Fig. 5A, black line) or replacing the bath solution for a non-permeantcation (140 mM NMDG-Cl, 10 mMHepes-K, pH 7.4 (Fig. 5A, gray line). A shift in the reversal potential of the current (ΔVrev) was observed from 0 mV to −54±6 mV (n = 4), close to the theoretical Vrev calculated for K+ under those conditions (−70 mV). Replacement of the bath solution for buffered 340 mMKCl induced a ΔVrev of +39±1 mV (n = 5) with a theoretical calculated Vrev of +22 mV. These results suggest that TcCat preferentially permeates K+. To calculate the selectivity for cations over anions of TcCat, we applied similar voltage ramp protocols but replacing the bath solution for a non-permeant anion (140 mM K-gluconate, 10 mMHepes-K, pH 7.4). A ΔVrev of −8.7±0.4 (n = 10) was measured under asymmetrical conditions (Fig. 5A, red line). Based on the bi-ionic equation (see Equations under Text S1), the calculated permeability ratio for K+ over Cl− was 5.9±0.5 (n = 10), indicating a preferential cation permeability but with weak selectivity filter, in agreement with the sequence data.
In order to study TcCat selectivity for monovalent cations, a voltage-ramp protocol was applied under symmetrical condition for K+ (Fig. 5B, black line) or replacing the bath solution for 140 mMXCl, X being different cations. Under bi-ionic conditions, the shift in the Vrev indicates the relative permeability of X+ respect to K+ (Fig. 5B). The permeability sequence obtained was: K+>Cs+>NH4+>Rb+>Na+>Li+>NMDG+ (Fig. 5C) that corresponds to Eisenman sequence IIIa [26]. This represents a selectivity of about 2.5 for K+ over Na+, which may indicate that TcCat is a potential conductive pathway for both physiological ions.
This biophysical characterization shows that TcCat is, indeed, a channel permeable to K+ and that shows inward rectification. In the presence of Mg2+, the unitary conductance is, as expected, higher at negative than at positive potentials and the open probability is not voltage-dependent.
The effect of divalent cations was evaluated by adding controlled concentrations of Ba2+, Ca2+or Mg2+to the bath solution. Fig. 6A shows that in the presence of 10 mM BaCl2 (red line) or 10 mM CaCl2 (green line) a significant decrease in the total current can be observed compared with the control (black line). No important shift in the Vrev was recorded, indicating the low permeability for these ions. A consistent decrease in the total current was observed when a voltage-step protocol was applied (Fig. 6B) in the presence of lower concentrations of the divalent cations, with a more remarkable effect for Mg2+ (blue inverted triangles). A concentration-dependent effect was observed for Ba2+ and Ca2+ when applying a voltage-step protocol in the presence of controlled concentrations of both ions (Fig. S6 C and D). The effect of Ba2+ over the leak through asolectin vesicles was evaluated comparing the normalized current in the presence of different concentrations of the divalent ion on empty liposomes (Fig. S6B,black circles) or liposomes containing TcCat (Fig. S6B,black squares). In the presence of 1 mM BaCl2 the residual current at −80 mV (Fig. S6B,upper panel) is about 55% in empty liposomes while it is close to 30% in liposomes containing TcCat, indicating that a percentage of the current through the channel is sensitive to the presence of the divalent cation. Similar results are obtained at +80 mV (Fig. S6B,lower panel).Based on the dose-dependent blockage, the calculated inhibition constants (Ki) for Ba2+ were (in mM): 0.54±0.08 and 0.63±0.06 at −80 and +80 mV, respectively, with no significant dependency on the applied voltage (n = 3 independent experiments).
The blockage by Ca2+ required higher concentration, with calculated Kis (in mM) of 5.2±0.3 and 4.5±0.2 at −80 and +80 mV, respectively. In all cases, a residual current was observed, even at 10 mM divalent cation concentration, indicating a leaking current or a partial blockage (Fig. S6D and F).
Conventional K+ channel blockers were also tested (Fig. 6C), with no significant effect for tetraethylammmonium (TEA) up to 10 mM (n = 3) and a 50% reduction in the total current for 4-aminopyridine (4-AP) at 1 µM (n = 4). Importantly, a blockage effect was observed in the presence of the anti-TcCat antibody when added to the bath solution at a concentration of 0.12 µg/µl and at holding potentials of −80 and +80 mV (Fig. 6D). No significant effect was observed when the same concentration of pre-immune sera was applied to the preparations (Fig. 6D, pre-immune).
As mentioned before, an inward-rectifier K+ channel seems to be involved in the maintenance of plasma membrane potential, intracellular pH and osmoregulation [5], [9]. To assess the role of TcCat on some of these processes we evaluated the localization of the protein in T. cruzi epimastigotes and trypomastigotes under osmotic stress. Under isosmotic conditions, the channel is localized at the plasma membrane, in a punctuate pattern, with some intracellular staining more evident in epimastigotes (Fig. 7A and B, Iso). When epimastigotes were placed under hyperosmotic stress, maintaining the same ionic concentrations, TcCat almost completely translocated to the plasma membrane (Fig. 7A, Hyper). Remarkably, in trypomastigotes, after 30 sec under hyperosmotic stress, TcCat disappeared from the cell surface of the parasites (Fig. 7B, Hyper). No intracellular accumulation of the protein was observed suggesting that the protein is released to the extracellular medium, probably by shedding mechanisms previously described for other T. cruzi surface proteins [27]. To prove this hypothesis, the supernatant of parasites under different osmotic conditions were precipitated and evaluated by western blot analysis. In trypomastigotes under hyperosmotic stress TcCat was detected in the supernatants (Fig. S7A). That was not the case for trypomastigotes under isosmotic or hyposmotic conditions (Fig. S7A) or for epimastigotes under similar treatments (Fig. S7B). Parasites overexpressing GFP were used as a control to rule out lysis of the parasites as a mechanism of release of TcCat(Fig. S7A).
Addition of Ba2+ or 4-AP, at concentrations that inhibit TcCat activity, could prevent the change in the localization of the protein in trypomastigotes(Fig. 7C) suggesting that the mechanism of elimination is linked to the sensing of the K+ concentration.
No differences were observed in TcCat localization when the parasites were under hyposmotic stress (Fig. 7A and B, Hypo). Morphologically, the typical response toosmotic stress can be detected in the parasites, with a more accentuated change in the shape of epimastigotes compared with trypomastigotes (Fig. 7A and 7B). Although the osmotic stress response is sensitive to tubulin de-polymerizing agents [28], TcCat translocation was not blocked or modified by treatment with trifluralin (500 µM) or chloralin (10 µM) (Fig. S8).
The potential role of TcCat in osmoregulation was further evaluated in the parasites under osmotic stress. Interestingly, the shrinkage of trypomastigotes under hyperosmotic stress was significantly reduced in the presence of 1 mM BaCl2 (Fig. 7D, open triangles) or 100 µM 4-AP (Fig. 7D, black squares) as compared to the control (Fig. 7D, open diamonds) suggesting a role of K+ influx through TcCat during cell volume decrease.
In T. cruzi [9] as well as in T. brucei [29]–[31] and other protists like Toxoplasma gondii [32] and Pneumocystis carinii [33], the membrane potential is not dependent on K+ but rather is proton driven. The permeability to K+in the plasma membrane is not predominant and it varies depending on the developmental stage of the parasite.We postulated previously that an inward rectifier K+ channel was involved in K+ uptake in T. cruzi trypomastigotes, dissipating the membrane potential generated by a plasma membrane located H+-ATPase and thereby facilitating its function in controlling pHi [5]. In this work, we demonstrate that a gene encoding a protein with sequence homology to K+ channels is present in the T. cruzi genome (TcCat) and can complement yeast deficient in K+transporters, providing genetic evidence that it encodes a functional K+channel. In addition, we demonstrate by patch clamping giant liposomes containing recombinant TcCat that this channel has several distinct characteristics that differentiates it from mammalian inward rectifier K+(Kir)channels. Furthermore, this channel has the unexpected ability of translocating to different cellular compartments in response to environmental stress.
Examination of TcCat indicates that the only well conserved sequence of the protein compared with other K+ channels is its tetramerization domain, which is probably the reason why it was first annotated as a voltage-dependent K+ channel (TriTrypDB.org). The same domain is present in all voltage-dependent ion channels, including K+, Na+and Ca2+channels. BLASTP analysis of TcCat shows as only significant hits voltage-dependent K+channels from different organisms (e.g. E value: 3−08 to 6−06 for Xenopus). We need to consider that the overall sequence identity between ion channels is low. ClustalW alignments show that the identity between KcsA and H. sapiens Kv4.3 (both voltage-dependent K+channels) is only 23%. Same analysis indicates that TcCat and H. sapiens Kv4.1, and Kv4.3 are 19 and 20% identical, respectively. When inward-rectifier K+channels are compared, similar results are found, with 20% identity between H. sapiens Kir2.1 and KirBac1.1 and 23% identity between Kir2.1 and KirBac2.1. Moreover, when the comparison is established considering other evolutionary distant organisms, like C. elegans, the identity values obtained are always close to 20%. For other type of channels, like Na+and Ca2+channels, the conservation is even lower, underscoring the relevance of functional validation to establish the function of putative ion channels.
TcCat lacks the conserved K+ channel signature sequence [T-X-G-Y(F)-G] [20], which is compatible with the relatively low selectivity of the channel for K+ over Na+ and the possibility that TcCat can potentially transport both ions although the relative permeability ratio PNa/PK is lower than the values previously reported for other non-selective cation channels [34], [35]. This is in agreement with results showing that either K+ or Na+ (at high concentrations) are important for pHi control in trypomastigotes under acidic conditions [5]. Unfortunately, there are no structural data identifying amino acids that form the pore in non-selective cation channels and the sequence TLESW, recently identified as the selectivity filter for Na+channels [36], is not present in TcCat. The extracellular pore-forming region of TcCat has a short sequence (TFGADG in TcCat and TYGADG in TbCat and LmCat, Fig. S1) characterized by the presence of negatively charged amino acids, and a glycine residue, which could be involved in K+ selectivity, as occurs in HKT transporters from plants [37] and bacteria [38], and that is also conserved in the TbKHT1 K+ transporter of T. brucei [39]. This selectivity filter would be more similar to the low conserved pores described for some bacterial KirBac [40] or with the mutation in the pore of Kv3.1 in weaver mice that turns it into a non-selective cation channel [41], [42]. The presence of a distinct selectivity filter in TcCat could be important for the development of specific inhibitors with therapeutic potential against trypanosomes. In bloodstream trypomastigotes TcCat is exposed to the cell surface making it accessible to blockade by pharmacologic agents.
Previous investigators have used yeast strains carrying trk1Δ and trk2Δ deletions for complementation with inward-rectifying K+ channels from a variety of organisms [43]–[45]. However, it has been indicated that many inwardly rectifying K+ channels (as occurs with TcCat) are inhibited by high concentrations of external divalent cations and that to analyze heterologously expressed K+ channels in yeast it is desirable to reduce the concentration of external divalent cations. These conditions, however, favor the activity of the non-specific uptake system NSCI [46]. On the other hand, growth at the low pH required for mutants carrying the trk1, trk2, and tok1 deletions used in this work inactivates NSC1 [25]. The complemented mutants obtained in this work will therefore provide a versatile genetic system for further studies of the assembly and composition of TcCat.
There is no previous electrophysiological description of the biophysical properties of ion channels in trypanosomatids. There are several limitations for the characterization of ion channels in motile unicellular organisms. Small size, irregular shape and active motility represent a problem for direct recording. The presence of a strong subpellicularcytoskeleton beneath the plasma membrane makes extremely hard to excise the patch and obtain a seal of suitable quality for single-channel recordings. The alternative of a cell-attached configuration is limited by the noise that the motility of the parasite introduces. We failed in many attempts to do direct patch-clamp in the parasites. Methods that decrease the motility like low temperature or use of actin-depolymerizing agents were considered but it can also be argued that they change the physiological conditions of the cell making the results obtained subject to discussion. Based on these facts we decided to use a reconstituted system in giant liposomes for ion channel characterization that has been extensively validated [19], [35], [47]–[58].
Although the mechanism by which the proteins are inserted in the liposomes is unknown, the orientation in which this occurs is not random. Reconstitution of acetylcholine receptors [47], glycine receptors [48], glutamate receptors [49], KcsA [52], [59] and KirBac1.1 [57] indicate that the proteins are oriented “right-side out”, meaning with the intracellular side facing the bath. The direction of the rectification observed for TcCat (Fig. 4B) suggests that this channel is also oriented with the intracellular side facing the bath.
The low conservation of the structure, particularly the sequence of the selectivity filter, can explain the functional differences observed in TcCat compared with other cation channels. Characterization of K+channels, although detailed and exhaustive in some cases, is mainly limited to bacterial (KcsA) and mammalian channels, with some particular cases of model organisms like Drosophila and C. elegans. In Fasciola hepatica [60]and Dictyosteliumdiscoideum [61] the presence of K+ channels with relative permeability ratio PK/PCl of 5 and 7, respectively, have been previously reported, suggesting that the selectivity of some channels in these organisms is not as high as for bacterial or mammalian K+channels.
Electrophysiological characterization of TcCat by patch clamping of giant liposomes indicates characteristics of K+ permeable channel with inward rectification characterized by non-linear current potential relationship for the single channel conductance. Therectification is weak, similar to what has been reported for KirBac1.1 and Kir1.1 [20], [57]. TcCat unitary conductance is significantly higher at negative than at positive potentials only in the presence of Mg2+, with no important differences in the outward currents suggesting a different mechanism of TcCat blockage by Mg2+. Although the residue responsible for the rectification in mammalian Kir channels (171D) is conserved in TcCat (Fig. S1C), it is followed by a positively charged amino acid (171His), which could potentially interfere with the binding of Mg2+ to the aspartic acid residue. Site-directed mutagenesis studies are in our future plans to address this and other structural properties.
Overall, the open probability did not show significant voltage-dependency. We have to mention that important differences were observed with different preparations. This could be explained by variations in the way that channels associate into clusters when reconstituted in liposomes. This type of behavior has been previously reported when purified KcsA was recorded in giant liposomes (52).
TcCat also differs from other inward rectifier K+ channels in its low selectivity for K+ over Na+, suggesting that it can transport either cation, and in its pharmacology. Blockers most commonly used for Kir channels are Ba2+ and Cs+ while TEA and 4-AP are known as inhibitors of Kv channels but have little effect on Kir channels [20]. TcCat was much less sensitive to Ba2+ than classical Kir channels (Ki 540–630 µM as compared to 13–390 µM for Kir2.x channels, [20]) and was insensitive to Cs+, when assayed in giant liposomes. Cs+, however, was as effective as Ba2+ in decreasing pHi of intact trypomastigotes [5]. TcCat is not permeable to Ca2+, even more it can be blocked by it at similar concentration that we have previously reported as inhibitory for non-selective cation channels from T. cruzi epimastigotes membranes [35].
Although Ca2+usually does not block inward-rectifier channels, structural and functional differences observed in TcCat make it a non-canonical inward rectifier. KirBac1.1 is also a weak rectifier K+channel that has been demonstrated to show several atypical behaviors compared with mammalian Kirs like inhibition by Ba2+and Ca2+ [62], polyamine insensitivity [57] and blockage (instead of activation) by PIP2 [63]. In addition, 4-AP had inhibitory effect on TcCat total current at relatively low concentrations (1 µM).
It has been reported that although free-living prokaryotes have recognizable K+ channel genes, most but not all, parasitic prokaryotes have no K+ channel genes since they live in the K+-rich environment of their host cells [64]. The eukaryote T. cruzi, which has both intracellular and extracellular stages has solved the problem of having a K+ channel while in a K+-rich environment by sequestering it to an intracellular location in intracellular amastigotes. This sequestration starts very early during differentiation of trypomastigotes upon acid pH-stimulation. Interestingly, if the amastigotes are released as such to the extracellular medium, poor in K+, the channel reappears at the surface of the cells. Translocation of the channel to the surface also occurs in epimastigotes submitted to hyperosmotic stress suggesting a role for this channel in the recovery from this type of stress. Although a number of mechanisms are involved in the control of localization of K+channels in other cells [20], there is no precedent for the rapid translocation of TcCat that occurs when the cells are submitted to acidic pH (trypomastigotes), hyperosmotic (epimastigotes), or extracellular stress (amastigotes). Rapid phosphorylation/dephosphorylation changes could be involved in this translocation, as occurs with Kv4.2 [65] and ROMK [66]. The two-pore K+ channels K2P 3.1 and K2P 9.1 cell surface destination is also dependent on phosphorylation which regulates the interaction with 14.3.3 proteins [67]. Ιν αδδιτον, the localization of TcCat in the flagellar membrane of trypomastigotes suggests a role for this channel in the modulation of flagellar motility and sensing. In this regard, K+ channels are required to modulate the motility of ciliates and sperm cells [68] and Ca2+ channels located in the flagellar membrane of T. brucei are important for flagellar attachment and intracellular signaling [69].
In conclusion, we identified and characterized, at the molecular and biochemical levels, an novel inward-rectifier cation channel in T. cruzi with electrophysiological characteristics different from other Kir channels and that has the surprising ability to change its cellular localization when cells are exposed to different environmental stresses. In addition we have obtained yeast mutants that will provide a useful genetic system for studies of the assembly and composition of the channel and we demonstrated the feasibility of purifying a functional ion channel from T. cruzi after recombinant expression in bacteria.
The entire open reading frame of TcCat was amplified by PCR with the primers 5′-CGGGATCCATGAGAAGGCGGGCCGTC-3′ and 3′-AACTGCAGTTAATGCGCTCTCCATATGTC -5′ introducing restriction sites for BamHI and PstI (underlined). PCR product was cloned into pGEM-T easy (Promega) and verified by automated sequencing. Cloned product was digested with restriction enzymes and ligated into pQE80L (Qiagen) expression vector. Expression of the recombinant protein in E. coli pLysS strain was induced with 0.5 mM isopropyl-β-D-thiogalactopyranoside (IPTG) overnight at 37°C. His-tagged recombinant protein was purified under denaturing conditions with His-Bind cartridges (Novagen). Purified product was separated by SDS-PAGE, stained with Coomasie blue and the corresponding band was excised from the gel and used as immunogen to obtain a rabbit polyclonal antibody against TcCat at CocalicoBiologicals, Inc (Reamstown, PA).
For immunofluorescence microscopy, parasites were fixed in PBS, pH 7.4, with 4% paraformaldehyde, adhered to poly-lysine coverslips, and permeabilized for 3 min with PBS, pH 7.4, containing 0.3% Triton X-100. Permeabilized cells were treated for 30 min at room temperature with 50 mM NH4Cl and blocked overnight with 3% BSA in PBS pH 8.0. Purified polyclonal antibody against TcCat (dilution 1∶250) was incubated for 1 h at room temperature. Goat α-mouse and goat α-rabbit Alexa conjugated secondary antibodies (1∶2,000) were incubated for 1 h at room temperature. DNA-containing organelles were stained with 4′,6-diamidino-2-phenylindole (DAPI) (5 µg/ml). For TcCat immunolocalization in intracellular amastigotes, the cells were grown in coverslips and fixed at different times post-infection in cold methanol for 20 min. Immunolocalization in non-permeabilized parasites was done as described omitting the permeabilization step. A monoclonal antibody against T. brucei phosphate pyruvate dikinase (PPDK) (glycosomal marker) (a gift from Frédéric Bringaud, Université Bordeaux Segalen, France) was used as a permeabilization control. Differential interference contrast (DIC) and direct fluorescence images were obtained by using an Olympus IX-71 inverted fluorescence microscope with a PhotometrixCoolSnapHQ charge-coupled device camera driven by Delta Vision softWoRx3.5.1 (Applied Precision, Issaquah, WA). This same software was used to deconvolve and process the final images. The figures were built by using Adobe Photoshop 10.0.1 (Adobe System, Inc., San Jose, CA).
For western blot analysis, T. cruzi epimastigotes, amastigotes and trypomastigotes were collected by centrifugation at 1,600× g for 10 min, washed twice in PBS, pH 7.4, and resuspended in modified RIPA buffer (150 mM NaCl, 20 mM Tris-Cl pH 7.5, 1 mM EDTA, 1% SDS and 0.1% Triton X-100) containing protease inhibitor cocktail (2 mM EDTA, 2 mMphenylmethylsulfonyl fluoride (PMSF), 2 mMtosylphenylalanylchloromethyl ketone (TPCK), 0.1 mMtrans-epoxysuccinyl-L-leucylamido(4-guanidino)butane (E64) and Sigma P8340 protease inhibitor cocktail, 1∶250). Total homogenate of each sample were separated by SDS-PAGE. Proteins were transferred onto nitrocellulose membranes and blocked overnight with 5% nonfat dry milk in PBS-0.1% Tween 20 (PBS-T). Blotting was done with α-TcCat (1∶5,000) and goat α-rabbit horseradish peroxidase conjugated antibodies (1∶20,000) for 1 h at room temperature and developed with ECL reagent. Membranes were stripped with 62.5 mMTris-HCl, pH 6.8, 2% SDS, 1% β-mercaptoethanol at 50°C for 30 min, extensively washed in PBS-T and incubated with monoclonal α-tubulin (Sigma) and goat α-mouse horseradish peroxidase conjugated antibodies (1∶10,000) as a loading control. Densitometric analysis of 4 independent experiments was performed with Alfa-Imager software.
T. cruzi trypomastigotes and epimastigotes at log phase of growth (3 days) were collected at 1,600× g for 5 min, washed twice in PBS and resuspended in isosmotic buffer (64 mMNaCl, 4 mMKCl, 1.8 mM CaCl2, 0.53 mM MgCl2, 5.5 mM glucose, 150 mM D-mannitol, 5 mMHepes-Na, pH 7.4, 282 mosmol/L) at a cell density of 1×108/ml. Aliquots of 5×106 parasites were placed in tubes and 500 µl of either hyposmotic (64 mMNaCl, 4 mMKCl, 1.8 mM CaCl2, 0.53 mM MgCl2, 5.5 mM glucose, 50 mM D-mannitol, 5 mMHepes-Na, pH 7.4, 177 mosmol/L) or hyperosmotic buffer (64 mMNaCl, 4 mMKCl, 1.8 mM CaCl2, 0.53 mM MgCl2, 5.5 mM glucose, 500 mM D-mannitol, 5 mMHepes-Na, pH 7.4, 650 mosmol/L) were added. TcCat blockers were added at the indicated concentrations to the corresponding buffers. Cells were fixed at different times after osmotic stress by adding same volume of 8% paraformaldehyde in PBS, pH 7.4, and immunofluorescence analysis was performed as described before. Relative cell volume changes after osmotic stress were measured by light scattering method [71]. Briefly, the cells were washed twice in PBS and resuspended at a density of 4×108 cells/ml in isosmotic buffer. Aliquots of 4×108 parasites were distributed in 96 well plates and an appropriate volume of hyperosmotic buffer was added to reach a final osmolarity of 650 mosmol/L. Absorbance at 550 nm was monitored every 10 sec for 12 min. The results were normalized respect to the value of a 3 min pre-reading under isosmotic conditions.
To measure TcCat release after hyperosmotic stress, trypomastigotes and epimastigotes under osmotic stress were collected by centrifugation (1,600× g for 10 min) after 2 min of treatment. Supernatants were precipitated with 10% trichloroacetic acid for 1 h on ice. Precipitated proteins were collected by high-speed centrifugation (20,000× g for 20 min), washed and evaluated by western-blot analysis using anti-TcCat. Anti-tubulin was used as control to show that no parasite material was present in the supernatant, other than the released protein. Parasites overexpressing GFP were used as a control to rule-out lysis of the cells as a mechanism of release of TcCat (Fig. S7A).
|
10.1371/journal.pcbi.1000781 | Impact of Dendritic Size and Dendritic Topology on Burst Firing in Pyramidal Cells | Neurons display a wide range of intrinsic firing patterns. A particularly relevant pattern for neuronal signaling and synaptic plasticity is burst firing, the generation of clusters of action potentials with short interspike intervals. Besides ion-channel composition, dendritic morphology appears to be an important factor modulating firing pattern. However, the underlying mechanisms are poorly understood, and the impact of morphology on burst firing remains insufficiently known. Dendritic morphology is not fixed but can undergo significant changes in many pathological conditions. Using computational models of neocortical pyramidal cells, we here show that not only the total length of the apical dendrite but also the topological structure of its branching pattern markedly influences inter- and intraburst spike intervals and even determines whether or not a cell exhibits burst firing. We found that there is only a range of dendritic sizes that supports burst firing, and that this range is modulated by dendritic topology. Either reducing or enlarging the dendritic tree, or merely modifying its topological structure without changing total dendritic length, can transform a cell's firing pattern from bursting to tonic firing. Interestingly, the results are largely independent of whether the cells are stimulated by current injection at the soma or by synapses distributed over the dendritic tree. By means of a novel measure called mean electrotonic path length, we show that the influence of dendritic morphology on burst firing is attributable to the effect both dendritic size and dendritic topology have, not on somatic input conductance, but on the average spatial extent of the dendritic tree and the spatiotemporal dynamics of the dendritic membrane potential. Our results suggest that alterations in size or topology of pyramidal cell morphology, such as observed in Alzheimer's disease, mental retardation, epilepsy, and chronic stress, could change neuronal burst firing and thus ultimately affect information processing and cognition.
| Neurons possess highly branched extensions, called dendrites, which form characteristic tree-like structures. The morphology of these dendritic arborizations can undergo significant changes in many pathological conditions. It is still poorly known, however, how alterations in dendritic morphology affect neuronal activity. Using computational models of pyramidal cells, we study the influence of dendritic tree size and branching structure on burst firing. Burst firing is the generation of two or more action potentials in close succession, a form of neuronal activity that is critically involved in neuronal signaling and synaptic plasticity. We found that there is only a range of dendritic tree sizes that supports burst firing, and that this range is modulated by the branching structure of the tree. We show that shortening as well as lengthening the dendritic tree, or even just modifying the pattern in which the branches in the tree are connected, can shift the cell's firing pattern from bursting to tonic firing, as a consequence of changes in the spatiotemporal dynamics of the dendritic membrane potential. Our results suggest that alterations in pyramidal cell morphology could, via their effect on burst firing, ultimately affect cognition.
| Neurons exhibit a wide range of intrinsic firing patterns with respect to both spike frequency and spike pattern [1]–[3]. A distinct type of firing pattern that is critically involved in neuronal signaling and synaptic plasticity is burst firing, the generation of clusters of spikes with short interspike intervals [4]. Bursts can improve the signal-to-noise ratio of neuronal responses [5] and may convey specific stimulus-related information [6]. Bursts of spikes can be more effective than single spikes in inducing synaptic long-term potentiation (LTP) [7], [8], or can even determine whether LTP or LTD (long-term depression) occurs [9]. In synapses with short-term facilitation, bursts can be transmitted more reliably than isolated spikes [10].
Electrophysiology, in combination with computational modeling, has elucidated the ionic mechanisms underlying intrinsic neuronal burst firing. Two main classes of mechanisms have been distinguished [4]. In so-called dendrite-independent mechanisms—responsible for bursting in thalamic relay neurons [11], for example—the fast, spike-generating conductances and the slow, burst-controlling conductances are co-localized in the soma. Conversely, in dendrite-dependent mechanisms—involved in pyramidal cell burst firing—these conductances are distributed across the soma and dendrites, with the interaction between somatic and dendritic conductances playing an essential role in burst generation. Dendritic voltage-gated Na+ and K+ channels, which promote propagation of action potentials from the soma into the dendrites, cause the dendrites to be depolarized when, at the end of a somatic spike, the soma is hyperpolarized, leading to a return current from dendrites to soma. The return current gives rise to a depolarizing afterpotential at the soma, which, if strong enough, produces another somatic spike [12], [13]. This whole process was described by Wang [13] as ‘ping-pong’ interaction between soma and dendrites.
Although ion channels play a pivotal role in burst firing, dendritic morphology also appears to be an important factor. In many cell types, including neocortical and hippocampal pyramidal cells [14]–[17], neuronal firing patterns and the occurrence of bursts are correlated with dendritic morphology. Results from modeling studies also suggest a relationship between dendritic morphology and firing pattern [18]–[21]. However, these studies are mainly correlative [21], focus on morphologically very distinct cell classes [18], use only the physiologically less appropriate stimulation protocol of somatic current injection, and do not investigate the impact of topological structure of dendritic arborizations. Consequently, the effects of dendritic size and dendritic topology on burst firing, and the underlying mechanisms, remain poorly known.
Considering that dendritic morphology can undergo significant changes in many pathological conditions, such as Alzheimer [22], [23], mental retardation [24], [25], epilepsy [26], and chronic stress [27]–[29], it is important to examine the implications of altered morphology for neuronal firing. Using computational models of neocortical pyramidal neurons, we here explore in a systematic and rigorous way the impact of a cell's dendritic morphology on the ping-pong mechanism of burst firing, under either somatic current injection or synaptic stimulation of the apical dendritic tree. Importantly, we thereby distinguish between the effects of size and topology of the apical dendrite. Furthermore, we identify the underlying mechanism by which morphology affects burst firing in the model.
We use a morphologically and biophysically realistic model of a bursting layer 5 pyramidal cell from cat visual cortex (Fig. 1A) that is based on [18] (which in turn builds upon [30]). The model is implemented in NEURON [31] and captures the general features of bursting in pyramidal cells, particularly the interaction between soma and dendrites in burst generation [12], [13]. In the soma, the voltage-dependent currents and associated maximal conductances (in pS µm−2) are as follows: a fast sodium current, ; a slow voltage-dependent non-inactivating potassium current, ; a fast non-inactivating potassium current, ; a slow calcium-activated potassium current, ; and a high voltage-activated calcium current, . In the axon hillock, and In the apical dendrite, the conductances are as in the soma, except that . In both the soma and the dendrites, the membrane capacitance µF cm−2, the axial resistance 80 Ω cm, and . Internal calcium concentration is computed using entry via the high-voltage activated calcium current and removal by a first order pump, where the baseline calcium concentration is 0.1 µM and the time constant of calcium removal is 200 ms [18], [32]. The reversal potentials (in mV) are , , , and . All currents are calculated using conventional Hodgkin-Huxley-style kinetics. For the specific rate functions for each current, we refer to [18].
The pyramidal cell is activated by either somatic or dendritic stimulation. For somatic stimulation, the cell is continuously stimulated with a fixed current injection of 0.2 nA at the soma. For dendritic stimulation, the cell is stimulated by synapses that are regularly distributed across the apical dendrite, with a density of 1 synapse per 20 µm2. For this synaptic density, the total input current, based on the current transfer at a single synapse, is approximately the same as with somatic stimulation.
The excitatory synaptic input is mediated by AMPA receptors. The time course of conductance changes follows an alpha function , where , with the peak time ms, the peak conductance , nS, and the reversal potential mV [33]. Each synapse is randomly activated, whereby the time intervals between the activations of a synapse are drawn from a negative exponential distribution , where is the mean of the distribution. Over the time period of a complete simulation, this results in a Poisson distribution of synaptic activation times. For each synapse, the mean activation frequency is set to 1 Hz, and each synapse is activated independently of the other synapses.
The firing patterns were recorded from the soma. Each simulation lasted 10000 ms, of which the first 1000 ms were discarded in the analysis in order to remove possible transient firing patterns.
In studying the influence of pyramidal cell morphology on burst firing, we distinguish between dendritic size and dendritic topology. The size of a dendritic tree is the total length of all its dendritic segments. The segment between the soma and the first branch point is called the root segment (see Fig. 2B). Dendritic segments between two branch points are intermediate segments, and segments between a branch point and a terminal tip are terminal segments. The topology of a dendritic tree is the way in which the dendritic segments are connected to each other. For example, a tree with a given number of terminal segments can be connected in a fully asymmetrical or a fully symmetrical way (see Fig. 2B, dendritic trees 1 and 23, respectively).
To investigate how the dendritic size of the pyramidal cell influences burst firing, we varied the total length of the cell's apical dendrite according to two methods. In the first method, we successively pruned terminal segments from the apical dendritic tree. Starting with the full pyramidal cell morphology, in each round of pruning we randomly removed a number of terminal segments from the apical dendritic tree. Each terminal segment had a chance of 0.3 to be removed. From the reduced dendritic tree, we again randomly cut terminal segments, and so on, until in principle the whole apical dendrite was eliminated. This whole procedure was repeated 20 times. The density of synapses was kept constant during pruning, so with dendritic stimulation pruning also changed the total input to the cell. With somatic simulation, the total input to the cell did not change when the apical dendrite was pruned.
In the second method, we kept the dendritic arborization intact and changed the size of the apical dendrite by multiplying the lengths of all its segments by the same factor. Thus in this way the entire apical dendritic tree was compressed or expanded. For dendritic stimulation, we kept the total synaptic input to the cell constant by adapting the density of the synapses. So, both with somatic and dendritic stimulation, the total input to the cell did not change when the size of the apical dendrite was modified.
To examine the impact of the cell's dendritic branching structure on burst firing, we varied the topology of the apical dendritic tree by swapping branches within the tree. The apical dendritic trees that were generated in this way all have exactly the same total dendritic length and other metrical properties such as total dendritic surface area and differ only in their topological structure. The total input to the cell, both with somatic and dendritic stimulation, did not change when the topological structure was altered.
To facilitate a systematic analysis of the role of dendritic size and dendritic topology in shaping burst firing, we also use a set of morphologically simplified neurons. The neurophysiological complexity of these neurons is similar to that of the full pyramidal cell model. For a systematic study, one must use trees with a relatively small number of terminal segments, because otherwise the number of topologically different trees becomes so large that simulating all of them becomes impossible. For a tree with only 12 terminal segments, for example, there already exist as many as 451 different tree topologies [34]. Here, we use a set of 23 neurons consisting of all the topologically different trees with 8 terminal segments (Fig. 2). The trees may also be thought of as representing the backbones of potentially much larger dendritic arborizations.
All segments in the tree (intermediate and terminal segments; see Fig. 2B) have the same length, so that the different tree topologies do not differ in total dendritic length. In almost all types of neurons, including neocortical pyramidal cells, the diameters of dendritic segments decrease at each branch point, with terminal segments having the smallest diameter [35], [36]. In the trees we use, the diameter of a parent segment, , is related to the diameters of its daughter segments, and , as , where the branch power e is equal to 1.5 (Rall's power law) [37]. For pyramidal cells, values of the branch power were found to range between 1.5 and 2 [35], [36]. Since terminal segment diameters show only a narrow range of values [38], all terminal segments were given the same diameter (0.7 µm) [35], [38], while the diameters of intermediate segments were calculated using Rall's power law. This implies that asymmetrical topologies will have a higher total dendritic surface area than symmetrical topologies. We therefore also considered the case in which all segments in the tree have the same diameter (3 µm), so that the different tree topologies do not differ in dendritic surface area. All neurons also have a soma compartment, with a diameter and length of 14 µm [39].
The ion channel types and densities are based on Mainen and Sejnowski's [18] reduced model. In the soma, there is a fast sodium current, , and a fast non-inactivating potassium current, (maximal conductances, in pS µm−2). The dendrites contain a fast sodium current, ; a slow voltage-dependent non-inactivating potassium current, ; a slow calcium-activated potassium current, ; a high voltage-activated calcium current, ; and a leak current, . Internal calcium concentration is computed using entry via the high-voltage activated calcium current and removal by a first order pump, where the baseline calcium concentration is 0.1 µM and the time constant of calcium removal is 200 ms [18], [32]. For both the soma and the dendrites, the membrane capacitance µF cm−2 and the axial resistance 80 Ω cm. The reversal potentials (in mV) are , , , and . All currents are calculated using conventional Hodgkin-Huxley-style kinetics. For the specific rate functions for each current, we refer to [18].
As in the pyramidal cell model with full morphological complexity, the neurons are activated by either somatic or dendritic stimulation. All the tree topologies receive the same input. For somatic stimulation, the neurons are continuously stimulated with a fixed current injection of 0.03 nA (0.1 nA for the non-Rall neurons, in which segment diameter is equal throughout the dendritic tree). For dendritic stimulation, the cells are stimulated by 600 synapses, with on each terminal or intermediate segment (in total 15 segments for a tree with 8 terminal segments) 40 uniformly distributed synapses. With this number of synapses, the total input current, based on the current transfer at a single synapse, is approximately the same as with somatic stimulation. The synaptic input is mediated by AMPA receptors, with the same parameters as in the full pyramidal cell model. Also as in the full pyramidal cell model, each synapse is randomly activated according to a Poisson process, with a mean activation frequency of 1 Hz.
The simulations were performed in NEURON [31]. The firing patterns were recorded from the axosomatic compartment. Each simulation lasted 10000 ms, of which the first 1000 ms were discarded in the analysis in order to remove possible transient firing patterns.
To examine how the size of the dendritic tree influences firing pattern, we changed the total dendritic length of a given tree topology by multiplying the lengths of all its segments by the same factor. For dendritic stimulation, the number of synapses on the tree was thereby kept constant. So, both with somatic and dendritic stimulation, the total input to the cell did not change when the size of the dendritic tree was modified.
For presenting the firing patterns from the different tree topologies, we ordered the trees according to the degree of symmetry of their branching structure. To do this, we used a variant of the ranking scheme proposed by Harding [34], [40]. A binary tree can be described by denoting at each node, i.e., a branch point or the root, the sizes of the subtrees (in number of terminal segments) it carries. For example, a tree of size 1 is simply denoted as 1. A tree of size 2 is denoted as 2(1,1). For trees of size 3 (see Fig. 2B), there are two possibilities, 3(1,2(1,1)) and 3(2(1,1),1), but they do not represent topologically different trees, because the only difference is the order of the two subtrees. For trees with four terminal segments, there are two topologically different trees, 4(2(1,1),2(1,1)) and 4(3(2(1,1),1),1). Note that the last tree can also be written in several other ways, e.g., 4(3(1,2(1,1)),1).
To obtain a unique notation, we applied the following two rules. First, if the subtrees at a particular node have a different size, the largest subtree is put left of the comma. So we write 3(2(1,1),1) instead of 3(1,2(1,1)). Second, to order (sub)trees of equal size, we consider to be larger the tree that has the highest number at the first figure in which the tree descriptions differ. So of the following two trees, 4(2(1,1),2(1,1)) and 4(3(2,1),1), the second one is considered to be larger (since 3>2). Thus, in a description of an 8-terminal tree of which these two trees are the subtrees, the second subtree is put first, i.e., 8(4(3(2,1),1),4(2(1,1),2(1,1))).
Once all tree topologies were written in a unique form, they were ordered according to their size (in the extended sense, as described above), whereby the largest one was put first in the list. Since two trees can now be ordered simply by looking at the first figure in which their descriptions differ, this ordering is called a reverse lexicographical ordering. Applied to trees of the same size, it puts trees in order of symmetry, with the most asymmetrical tree first and the most symmetrical tree last. Fig. 2A shows the ordering of the 23 topologically different trees of 8 terminal segments. Note that the ordering is only used for presentation purposes and does in no way affect the results.
Bursting is defined as the occurrence of two or more successive spikes with short interspike intervals followed by a relatively long interspike interval. To quantify bursting, we used the burst measure developed in [41]. This measure is based solely on spike times and detects the correlated occurrence of one or more short (intraburst) interspike intervals followed by a long (interburst) interspike interval. It quantifies the extent of bursting in the whole spike train; it does not try to identify individual bursts, as some other approaches do [42]. The burst measure is based on the following notion (see also Text S1). If a spike train consists of spikes with independent successive interspike intervals, the variance of the sum of two successive interspike intervals [, where is the time of the ith spike in the spike train] is twice the variance of the single interspike intervals []. If bursting occurs, successive intervals are no longer independent, and this relation is violated. Thus, the difference between the two variances is a measure for bursting. If we divide this difference by the squared average interspike interval, we obtain a normalized burst measure (called B) that is sensitive only to the relative sizes of the interspike intervals and not to the average interval size:(1)where means , and stands for taking the expectation or average value of the interspike intervals between two successive spikes in the spike train. We used eqn (1) to quantify the extent of bursting in a spike train, thus taking into account all interspike intervals and to calculate the average interspike intervals and their variances. If a spike train consists of spikes with independent successive interspike intervals (i.e., no bursting),, and . Although B is a continuous measure assessing the degree of bursting and not classifying spike trains as either bursting or non-bursting, we will for practical purposes consider spike trains with a value of B below 0.15 as non-bursting (see also Fig. S8).
In Text S1, we derive that, for a periodic spike train with two-spike bursts, the expected value of B is(2)where is the average interspike interval between two consecutive bursts and is the average interspike interval within a burst. If , there is no bursting, and . The higher the ratio of inter- to intraburst interspike intervals, the stronger the bursting and the higher the value of . In the limiting case if goes to infinity, B goes to 1. The burst measure in eqn (1) is a general measure for detecting bursting in a spike train, and in Text S2 we show that it is equally valid for spike trains containing bursts that consist of more than two spikes.
The input conductance of a pyramidal cell with a given dendritic morphology was determined by applying a static, subthreshold current injection at the soma. The ratio of the magnitude of the injected current to the resulting change in membrane potential at the soma is defined as the input conductance of the cell [43]. The input conductance is the reverse of the input resistance.
To quantify the electrotonic extent of a dendritic tree, we introduce a new measure called mean electrotonic path length (MEP). For a given terminal segment (see Fig. 2B), the electrotonic path length is the length (normalized to the electrotonic length constant) of the path from the tip of the segment to the soma. This electrotonic path length is determined for each terminal segment, and the sum of all electrotonic path lengths is divided by the total number of terminal segments to obtain the MEP of the dendritic tree. More precisely, to obtain the MEP of a dendritic tree, we first normalize the length of each terminal, intermediate or root segment i (see Fig. 2B) with respect to its electrotonic length constants , yielding a dimensionless electrotonic length [44], in which is defined as [43](3)where is the radius of dendritic segment i, and and are constants denoting the specific membrane resistance and the intracellular resistivity, respectively. The MEP of a dendritic tree with terminal segments is then given by(4)where is the sum of the electrotonic lengths of all the dendritic segments in the path from the tip of terminal segment to the soma.
The analysis and model code for this paper including a tool for NEURON parameter scanning is available from ModelDB at http://senselab.med.yale.edu/modeldb via accession number 114359.
Employing a standard model of a bursting pyramidal cell [18], we investigated how dendritic morphology influence burst firing by varying either the size or the topology of the apical dendrite. The neurons were activated either at the soma with a fixed current injection or along the dendritic tree with random synaptic input. To facilitate a more comprehensive analysis, we also examined a set of morphologically simplified cells with systematic differences in dendritic topology.
To facilitate a better understanding of our findings obtained with the pyramidal cell model and to analyse more precisely the role of dendritic morphology in shaping burst firing, we also investigated a set of 23 morphologically simplified neurons consisting of all the topologically different trees with 8 terminal segments. Because the cells have relatively few terminal segments, the impact of dendritic topology on burst firing can be studied in a systematic way.
Given the crucial role of bursts of action potentials in synaptic plasticity and neuronal signaling, it is important to determine what factors influence their generation. Using a standard compartmental model of a reconstructed pyramidal cell [18], we have investigated how size and topology of dendritic morphology affect intrinsic neuronal burst firing.
We have shown that either shortening or lengthening the apical dendrite tree beyond a certain range can transform a bursting pyramidal cell into a tonically firing cell. Remarkably, altering only the topology of the dendritic tree, whereby the total length of the tree remains unchanged, can likewise shift the firing pattern from bursting to non-bursting or vice versa. Moreover, both dendritic size and dendritic topology not only influence whether a cell is bursting or not, but also affect the number of spikes per burst and the interspike intervals between and within bursts.
The influence of dendritic morphology on burst firing is attributable to the effect dendritic length and dendritic topology have, not on input conductance, but on the spatial extent of the dendritic tree, as measured by the mean electrotonic path length between soma and distal dendrites. For the spatiotemporal dynamics of dendritic membrane potential to generate burst firing, this electrotonic distance should be neither too small nor too large. Because the degree of symmetry of the dendritic tree also determines mean electrotonic path length, with asymmetrical trees having larger mean path lengths than symmetrical trees, dendritic topology as well as dendritic size affects the occurrence of burst firing.
In Mainen and Sejnowski's [18] two-compartment model for explaining the role of dendritic morphology in shaping firing pattern, the spatial dimension of morphology was completely reduced away. Although the model is able to reproduce a wide range of firing patterns, it does not capture the essential influence of dendritic morphology on burst firing, in which, as we have shown here, the spatial extent of the dendritic tree and the resulting spatiotemporal dynamics of the dendritic membrane potential are crucially involved.
The effect of dendritic size and topology on burst firing and the correlation of burst firing region with mean electrotonic path length are robust to changes in model properties, including morphology, strength of input stimulus, ion channel densities, and keeping the number of ion channels constant as morphology is changed. The specific range of dendritic sizes that supports burst firing, as well as the impact of dendritic topology, does not strongly depend on the strength of the input stimulus, especially with somatic stimulation (Figs. S1 and S2). More importantly, the overall way in which dendritic morphology influences burst firing is independent of stimulus strength. Likewise, the impact of dendritic morphology is qualitatively insensitive to the value of the branch power used in the morphologically simplified cells: even in dendritic trees in which the segment diameters are uniform we observe the same effect of dendritic length and topology (Fig. 8).
In changing dendritic size or topology, we held the density of ion channels constant (i.e., the conductances were fixed), which implies that the total number of ion channels also changed when dendritic morphology was varied. Keeping the conductances fixed seems biologically the most appropriate choice, since removing membrane to shrink the dendritic tree (as well as adding membrane to enlarge it) will include the membrane's ion channels and is therefore not expected to affect ion channel density. But even if we hold the number of ion channels constant, by adjusting the values of the conductances as the surface area of the dendritic tree is changed when dendritic topology or total length is varied, we obtain surprisingly similar results (Fig. S3). Although the precise values of mean electrotonic path length that support burst firing are slightly different, the overall effect of dendritic size and topology on burst firing and the correlation of burst firing region with mean electrotonic path length remain the same.
Since recent studies have shown that the same firing patterns can be produced by different combinations of conductances [47], [48] and even by different combinations of conductances and morphological properties [49], it is important to ensure that our results are not specific for the particular choice of conductance values in the Mainen and Sejnowski model [18]. Provided the model supports the ping-pong mechanisms of burst firing, our main findings are indeed robust to considerable changes in ion channel densities, both under somatic and under dendritic stimulation. Although the range of tree sizes that supports burst firing may be different for different ion channel densities, Figs. S4, S5, S6, S7 show that the general impact of dendritic size and topology on burst firing, as well as the correlation of burst firing with mean electrotonic path length, is maintained for a wide range of dendritic ion channel densities. Interestingly, the value of the mean electrotonic path length where burst firing commences, going from small to large trees, is not affected by ion channel density, as opposed to the value of the mean electrotonic path length where burst firing stops. This suggests that when the dendritic tree is reduced in size so that the cell no longer exhibits burst firing, compensatory changes in ion channel conductances [49] may not be able to bring back the cell to a bursting mode. In contrast, when the dendritic tree is enlarged beyond the range where the cell bursts, compensatory changes in ion channel conductances (e.g., increased dendritic ) may be able to restore burst firing.
Since it has experimentally been shown that removal of the apical dendrite abolishes bursting in layer 5 pyramidal cells [50] and pathological conditions often affect the apical dendrite, but not the basal dendrites [29], we focused in this study on the morphology of the apical dendritic tree. The simulations with the morphologically simplified cells, which do not have basal dendrites, show that basal dendrites are not essential for burst firing. In the pyramidal cell and the simplified cells, burst firing is similarly affected by dendritic morphology, which again emphasizes the robustness of our findings.
Compared with other modeling studies investigating the relationship between dendritic morphology and firing pattern [18]–[21], [51], our study is unique in that it focuses on burst firing, adopts a systematic approach, investigates morphological changes within a cell type, considers not only somatic stimulation but also physiologically more appropriate dendritic stimulation, and especially examines the impact of topological structure of dendritic arborizations. Moreover, our study is not just correlative but provides insight into the mechanisms underlying the influence of morphology on firing pattern.
We stimulated the cells either by a current injection at the soma, as is done in most experimental and modeling studies [18], or by synapses distributed over the dendritic tree, which is physiologically more relevant. Importantly, we found that the influence of dendritic morphology on burst firing is essentially the same under both stimulation regimes.
Our study confirms a suggestion by Krichmar et al. [21] that dendritic branching structure might have a direct influence on neuronal firing activity. In simulation studies, they found that although dendritic size could account for much of the differences in neuronal firing behavior between CA3 pyramidal cells, it did not provide a complete explanation for the observed electrophysiological variability.
Our results are in accord with empirical observations suggesting that pyramidal cells should have reached a minimal size to be capable of burst firing. In weakly electric fish, the tendency of pyramidal cells to fire bursts is positively correlated with the size of the cell's apical dendritic tree [52]. In rat prefrontal cortex [16] and visual cortex [14], [53], the classes of pyramidal cells that exhibit burst firing have a greater total dendritic length than the other classes.
In addition, the developmental time course of bursting shows similarities with that of dendritic morphology. In rat sensorimotor cortex, the proportion of bursting pyramidal cells progressively increases from postnatal day 7 onwards, while at the same time the dendritic arborizations become more complex [54]. In pyramidal cells from rat prefrontal cortex, the total lengths of apical and basal dendrites increase dramatically between postnatal days 3 and 21, with neurons capable of burst firing appearing only from postnatal day 18 onwards [55], [56].
Direct experimental testing of the influence of dendritic morphology on burst firing could be done by physically manipulating the shape or size of the dendritic tree, e.g., by using techniques developed by Bekkers and Häusser [50]. They showed that dendrotomy of the apical dendrite indeed abolished bursting in layer 5 pyramidal cells.
Dendritic morphology can undergo significant alterations in many pathological conditions, including chronic stress [27]–[29], [57], epilepsy [26], hypoxic ischemia [58], Alzheimer [22], [23], and disorders associated with mental retardation [24], [25]. Functional consequences of these morphological changes are usually interpreted in terms of loss or formation of synaptic connections as a result of a diminished or expanded postsynaptic surface area. Our modeling results indicate that alterations in dendritic morphology can directly modify neuronal firing, irrespective of changes in total synaptic input.
Chronic stress, as well as daily administration of corticosterone, induces extensive regression of pyramidal apical dendrites in hippocampus [27], [57], [59] and prefrontal cortex [28], [29]. As a result of a decrease in the number and length of terminal branches, the total apical dendritic length can reduce by as much as 32% [29], while basal dendrites are not affected. Similarly large alterations have been observed in response to mild, short-term stress [60]. Our results predict that stress and the accompanying reduction in apical dendritic length could turn a bursting neuron into a non-bursting one. Indeed, Okuhara and Beck [61] found that two weeks of high corticosterone treatment caused a decrease in the relative number of intrinsically bursting CA3 pyramidal cells. Since burst firing of CA3 pyramidal cells is critically involved in LTP [62], this could have profound functional consequences for hippocampal information processing [63].
With regard to epilepsy, a significant decrease in total dendritic length and number of branches has been found in pyramidal cells following neocortical kindling [26]. In line with our results, Valentine et al. [64] reported that activity of single cells recorded from the primary auditory cortex of kindled cats showed a reduction in the amount of burst firing and a decrease in the number of spikes per burst.
In Alzheimer's disease, various aberrations in dendritic morphology have been observed— including a reduction in total dendritic length and number of dendritic branches [22], [23] and alterations in the pattern of dendritic arborization [65]—which may contribute to the abnormal neurophysiological properties of Alzheimer pyramidal cells [66]. The anomalies in morphology could influence the cells' ability to produce burst, and, because of the role of burst firing in LTP and LTD [8], [9], ultimately affect cognition. In disorders related with mental retardation, the observed alterations in dendritic length and pattern of dendritic branching [24], e.g., changes in the degree of symmetry of the apical dendrite [67], may likewise be hypothesized to contribute to impaired cognition. Importantly, our results indicate that dendritic sprouting—which too has been observed in Alzheimer [68], [69]—may also be able to change neuronal burst firing.
Since firing patterns characteristic of different classes of neurons may in part be determined by total dendritic length, we expect on the basis of our results that a neuron may try to keep its dendritic size within a restricted range in order to maintain functional performance. Indeed, Samsonovich and Ascoli [70] have shown that total dendritic size appears to be under intrinsic homeostatic control. Statistically analyzing a large collection of pyramidal cells from hippocampus and prefrontal cortex, they found that, for a given morphological class and anatomical location, fluctuations in dendritic size in one part of a cell tend to be counterbalanced by changes in other parts of the same cell, so that the total dendritic size of each cell is conserved.
We predict that dendritic topology may similarly be protected from large variations. In fact, there could be a trade-off between dendritic size and dendritic topology. In a set of bursting pyramidal cells, we expect that apical dendritic trees with a lower degree of symmetry are shorter in terms of total dendritic length or have thicker dendrites to reduce electrotonic length than those with a higher degree of symmetry.
Although changes in dendrite morphology of pyramidal cells have been observed in response to environmental enrichment [71] and learning [72], recent long-term in vivo imaging studies have demonstrated remarkable stability of dendrites in adult animals [73], [74]. Like the homeostatic control of dendritic size, this stability may point to the functional relevance of dendritic topology.
An intriguing possibility is that firing pattern and dendritic morphology could mutually tune each other during development, as a result of a reciprocal influence between dendritic growth and neuronal activity. Dendritic morphology affects firing pattern, and neuronal activity in turn is known to modulate dendritic growth and branching [75], with, interestingly, firing frequency and firing pattern having distinct effects on outgrowth [76].
As our study underscores, differences in neuronal firing properties may not necessarily reflect differences in ion channel composition. In some cases, variability in dendritic morphology may even have a relatively bigger effect on firing pattern than variability in membrane conductances [49], [77]. Our results show that alterations in either the size or the topology of dendritic arborizations, as have been observed in many pathological conditions, could have a marked impact on pyramidal cell burst firing and, because of the critical role of bursting in neuronal signaling and synaptic plasticity, ultimately affect cognition.
|
10.1371/journal.ppat.1003600 | Kinetics of Myeloid Dendritic Cell Trafficking and Activation: Impact on Progressive, Nonprogressive and Controlled SIV Infections | We assessed the role of myeloid dendritic cells (mDCs) in the outcome of SIV infection by comparing and contrasting their frequency, mobilization, phenotype, cytokine production and apoptosis in pathogenic (pigtailed macaques, PTMs), nonpathogenic (African green monkeys, AGMs) and controlled (rhesus macaques, RMs) SIVagmSab infection. Through the identification of recently replicating cells, we demonstrated that mDC mobilization from the bone marrow occurred in all species postinfection, being most prominent in RMs. Circulating mDCs were depleted with disease progression in PTMs, recovered to baseline values after the viral peak in AGMs, and significantly increased at the time of virus control in RMs. Rapid disease progression in PTMs was associated with low baseline levels and incomplete recovery of circulating mDCs during chronic infection. mDC recruitment to the intestine occurred in all pathogenic scenarios, but loss of mucosal mDCs was associated only with progressive infection. Sustained mDC immune activation occurred throughout infection in PTMs and was associated with increased bystander apoptosis in blood and intestine. Conversely, mDC activation occurred only during acute infection in nonprogressive and controlled infections. Postinfection, circulating mDCs rapidly became unresponsive to TLR7/8 stimulation in all species. Yet, stimulation with LPS, a bacterial product translocated in circulation only in SIV-infected PTMs, induced mDC hyperactivation, apoptosis and excessive production of proinflammatory cytokines. After infection, spontaneous production of proinflammatory cytokines by mucosal mDCs increased only in progressor PTMs. We thus propose that mDCs promote tolerance to SIV in the biological systems that lack intestinal dysfunction. In progressive infections, mDC loss and excessive activation of residual mDCs by SIV and additional stimuli, such as translocated microbial products, enhance generalized immune activation and inflammation. Our results thus provide a mechanistic basis for the role of mDCs in the pathogenesis of AIDS and elucidate the causes of mDC loss during progressive HIV/SIV infections.
| Myeloid dendritic cells (mDCs) are potent antigen-presenting cells that regulate both innate and adaptive immune responses and act as “watch-dogs”, sensing and controlling aberrant immune activation; as such, they may significantly impact the outcome of HIV/SIV infection. By comparing and contrasting the frequency, function, migration to tissues and levels of activation and apoptosis in progressive, nonprogressive and elite-controlled SIV infections, we investigated the mechanisms responsible for mDC loss in HIV/SIV infection and their role in driving progression to AIDS. We report that progression to AIDS is associated with low mDC preinfection levels and depletion throughout infection, due to massive migration of these cells to mucosal sites and excessive cell death by apoptosis. We also show that residual mDCs from blood and intestine have a high capacity to produce proinflammatory cytokines, thus contributing to the increased immune activation and inflammation characteristic of progressive infections.
| Myeloid dendritic cells (mDCs) are potent antigen-presenting cells which are responsible for initiating both innate and adaptive immune responses. mDCs stimulate NK, B and T cells [1], [2], but can also act as “watch-dogs” to sense and regulate aberrant immune activation, induce tolerance and thus prevent autoimmune diseases [2]. The ability of mDCs to act as immune sentinels is conferred by their capacity to migrate into lymphoid organs where they secrete cytokines and initiate immune responses [2]. Altogether these characteristics suggest that mDCs may have a critical role in the pathogenesis of human and simian immunodeficiency virus (HIV/SIV) infections, in which progression to AIDS is driven by excessive generalized immune activation and inflammation [3].
A considerable amount of correlative data reporting changes in mDC counts during pathogenic HIV/SIVmac infections has been published [4]–[14], showing that mDC depletion from circulation occurs during acute HIV/SIV infection, at the time of peak viremia, and persists throughout the chronic infection and progression to AIDS [4], [7], [11], [14]. The exact mechanism of mDC depletion is not yet understood. In SIV-infected nonhuman primates (NHPs), mDC loss is inversely correlated with virus loads (VLs) [7], the NHPs that rapidly progress to AIDS having high VLs and a profound loss of mDCs [3]. Conversely, SIVmac-infected rhesus macaques (RMs) with normal or delayed disease progression have lower VLs and increased numbers of circulating mDCs [7]. NHP studies corroborate reports in HIV-infected patients showing that antiretroviral therapy restores mDCs and suggesting that direct killing of mDCs by the virus is a potential mechanism. mDCs can indeed be infected by HIV-1, the virus burden in mDCs isolated from chronically HIV-1-infected patients being considerable [15]. However, mDC depletion from circulation was also reported for HIV-2–infected patients and RMs infected with SIVmac (which is closely related to HIV-2), albeit mDCs appear to be less susceptible to HIV-2 infection [16], suggesting that additional factors may be responsible for mDC depletion. Other studies have suggested that mDC loss from the circulation may occur through either increased apoptosis [17] or mDC recruitment to the lymph nodes (LNs) [7], [18]–[20].
While mDCs are pivotal in shaping the mucosal microenvironment, no study to date has been carried out in either humans or animal models of HIV infection to explore whether or not the loss of circulating mDCs is due to recruitment to mucosal tissues such as the intestine, which is the main target of viral replication and where a high degree of inflammation occurs [21]. It was previously shown that a loss of the chemokine balance in the LN environment occurs in SIV-infected macaques and results in recruitment of mDCs to LNs [7], . Similar changes may also occur in the mucosal tissues, leading to recruitment of immune cells, including activated mDCs, to these sites [24].
Thus, the exact role of mDCs in the pathogenesis of AIDS is unknown, yet may be essential for designing better vaccines and therapeutic interventions. Loss of mDCs during pathogenic HIV/SIV infections prevents a clear understanding of their precise role in HIV/SIV pathogenesis. One option towards realizing this goal may rely on comparing and contrasting the dynamics, trafficking and function of mDCs in lentiviral infections with variable pathogenic outcomes.
There are three pathogenic outcomes of lentiviral infections: (a) The persistent progressive infection occurs in the majority of cases of HIV infection and SIVmac/smm infection of macaque species and is characterized by (i) massive, continuous viral replication [25]–[27], with VL setpoint being predictive for the time of progression to AIDS [28]–[30]; (ii) continuous depletion of CD4+ T cells from the peripheral blood [31], [32] that is more pronounced at mucosal sites [21], [33]–[35]; and (iii) high levels of T cell immune activation [36], [37], the magnitude of which has been reported to be predictive of disease progression [36], [37]. The interaction between these factors cripples the immune system and eventually results in severe immunodeficiency and death [31], [32], [38]. (b) The persistent nonprogressive infection is observed in African NHPs that are natural hosts of SIV, such as African green monkeys (AGMs), sooty mangabeys (SMs) and mandrills (reviewed in [39]), and is characterized by (i) active viral replication, with setpoint VLs similar to or even higher than those reported in pathogenic infections [40], [41]; (ii) transient depletion of peripheral CD4+ T cells during acute infection that rebound to preinfection levels during chronic infection [39], [40]; (iii) significant CD4+ T cell depletion in the intestine that can be partially restored during the chronic infection, regardless of significant viral replication [41]–[43]; (iv) low levels of CD4+ T cells expressing the CCR5 coreceptor in blood and tissues [44], [45], and (v) transient and moderate increases in immune activation and T cell proliferation during acute infection, which is resolved to near baseline levels with the transition to chronic infection [42], [46], [47]. Altogether, the action of these factors results in an active, persistent SIV infection, which only rarely progresses to AIDS in natural hosts [48]. (c) The controlled infection occurs in a minority (1–5%) of HIV-infected individuals, which are defined as long-term nonprogressors (LTNPs). LTNPs that have undetectable VLs for at least 1 year are referred to as elite controllers (ECs) (reviewed in [49]). The main characteristics of LTNP infections are (i) infection for more than 7 years; (ii) stable CD4+ T cell counts greater than 600 cells/µl; (iii) low/undetectable levels of HIV in the peripheral blood; (iv) no symptoms of HIV-induced disease; and (v) presence of a vigorous immune response against HIV, with multifunctional, persistent CD4 and CD8 responses. ECs that control immune activation in addition to VLs are referred to as superelite controllers. Similarly, SIV infection in NHPs may result in a controlled infection: in a fraction of Indian RMs infected with SIVmac and carrying restrictive MHC genotypes (MaMu A*01, B*17, B*08), in Chinese macaques, or upon cross-species transmission of SIVs [50]–[55]. By exposing RMs to SIVagmSab, we have recently developed an animal model in which superelite control/functional cure of SIV infection occurs in 100% of cases [56].
To date, all information on mDCs in HIV/SIV infections is derived from studies performed on persistent progressive (pathogenic) infections (reviewed in [15]). Very few of these studies focused on mDCs at mucosal sites [57]–[60]. Furthermore, to our knowledge, the role of mDCs in the pathogenesis of either persistent nonprogressive or controlled lentiviral infections has not yet been established.
Hence, to test the hypothesis that distinct mDC profiles are associated with particular pathogenic outcomes, we dissected the kinetics of mDCs in blood, LNs and at mucosal sites in persistent progressive infections in PTMs, nonprogressive persistent infections in AGMs and controlled infections in RMs after inoculation with the same virus strain (SIVagmSab). We report that the three pathological outcomes are associated with different dynamics of mDCs. To explain these outcomes, we examined the mobilization, migration, activation and apoptosis of mDCs in three animal models. We corroborated this descriptive data with mDC functional studies using cells isolated from blood and intestine from SIVagmSab-infected PTMs, AGMs and RMs. Altogether, our results indicate that mDCs play a significant role in driving the outcome of SIV infection.
Five PTMs, four AGMs, and five RMs were intravenously infected with the 300 TCID50 of SIVagmSab92018 [61]. All of the PTMs developed significant lymphadenopathy during the acute infection and exhibited signs of disease progression during the follow-up: two experienced weight loss during acute infection, followed by rapid progression to AIDS (78 and 104 days postinfection, dpi); two additional PTMs progressed to AIDS within a year postinfection; the remaining PTM did not progress to AIDS during the follow-up, but showed profound CD4+ T cell depletion and persistently high VLs, which are indicative of progressive, pathogenic infection. The PTMs with AIDS presented with anorexia/cachexia, diarrhea, LN and spleen atrophy, neurological disease, thrombotic microangiopathy (TMA), glomerulonephropathy and myocarditis. RMs had marked lymphadenopathy during acute infection. No clinical sign of infection was observed during acute infection in AGMs or during chronic infection in AGMs and RMs.
Acute VLs peaked at day 8–14 dpi (Figure 1a, b and c) and were similarly high in all three species. However, there were slight species-specific variations, with PTMs having the highest VLs, followed by RMs and AGMs; the different peak levels correlated with different levels of CD4+ T cells expressing the CCR5 molecule (the target cells of SIVagmSab), which are the highest in PTMs, intermediate in RMs and the lowest in AGMs [44]. In stark contrast to this uniform viral replication during the acute infection, chronic viral replication patterns were completely divergent between the three models (Figure 1): VLs remained high in PTMs and showed two patterns of chronic replication: noncontrolled in the two rapid progressors (chronic VLs of 108 SIVagmSab RNA copies/ml, i.e., a reduction of less than 1.5 logs from the peak VLs); in the remaining three PTMs, a relative control was observed (chronic VLs of 105–106 SIVagmSab RNA copies/ml, i.e., a contraction of 3 logs from the peak VLs) (Figure 1a). Conversely, in AGMs VLs reached a setpoint level of 4×104 copies/ml by 42 dpi, which was maintained throughout the follow-up (Figure 1b), in agreement with our previous studies showing that this setpoint can be maintained for decades in AGMs [62]. Finally, SIVagmSab-infected RMs gradually controlled VLs, which became undetectable from 60 dpi on and remained so throughout the follow-up (Figure 1c), in agreement with our previous studies showing that viral control can extend beyond 5–6 years [56].
As a first step of dissecting the role of mDCs in driving different outcomes of SIV infection, we assessed their characteristics in blood, LN and intestine in the three species of NHPs prior to infection. We first characterized mDCs in PTMs, AGMs and RMs by applying the strategy previously reported to identify mDCs in RMs [6], [7] and showed that selection based on Lineageneg, HLA-DR+ and CD11c+ identified mDCs with similar phenotypic characteristics in all three species in blood (Figure 2a), LNs (Figure 2b) and intestine (Figure 2c). Characterization of mDCs from blood, LNs and intestinal mucosa revealed, however, different levels of HLA-DR expression between the three sites (Figure 2): low in peripheral blood, high in the LN and intestine. This phenotypic variation is probably due to differences in mDC maturation stages: immature mDCs with low HLA-DR expression are present in circulation, while more mature mDCs expressing higher levels of HLA-DR are found in the LNs and mucosal tissues.
Comparison of mDC baseline levels between PTMs showed that the two rapid progressor PTMs had lower peripheral baseline mDC counts compared to the normal progressors. As the number of rapid progressor PTMs included in the study group precluded statistical analysis of the data, we compared the baseline levels of mDCs between rapid and normal progressors in a larger cohort of PTMs from previous studies. This analysis confirmed an inverse correlation between the low numbers of circulating mDCs prior to infection and the rapid disease progression of SIVagmSab-infected PTMs (Figure S1), thus suggesting a protective role of mDCs during the course of HIV/SIV infection.
Using a random effects model to analyze the dynamics of mDC during infection in all animals, we found that there were significant differences in the changes over time across the 3 infection models (p = 0.0003). We then analyzed in more detail the changes in circulating mDCs at key time points of SIVagm infection in the three species to determine whether or not differences in the dynamics of peripheral mDCs could explain the different pathogenic outcomes. Although dynamics of acute virus replication were similar in all three infections, the number of mDCs significantly dropped only in PTMs (Figure 3a) and AGMs (Figure 3b), but not in RMs (Figure 3c). Furthermore, the duration and timing of acute mDC depletion were different in PTMs and AGMs: rapid (from 1 dpi) and maintained throughout the acute infection in PTMs (Figure 3a), while very transient, reaching significance only around the peak of viral replication in AGMs (Figure 3b). mDC depletion from circulation prior to detectable viremia is not surprising, as in the mouse model the virus is already present in the LNs and that DCs are migrating to the LNs carrying the virus 30 minutes post intravenous HIV inoculation [63].
Major differences in the dynamics of mDCs became discernible between the three models during chronic SIV infection. In PTMs, progression to AIDS was characterized by a significant decline of mDCs (Figures 3a). To confirm that mDC depletion is specifically associated with disease progression, we compared mDC counts prior to SIV infection and during chronic SIV infection in animals included in other studies and showed that mDCs are indeed lost with disease progression (Figure S2). Our results thus corroborate previous reports documenting loss of peripheral mDCs during pathogenic HIV/SIV infections [7].
In AGMs, mDC levels returned to near baseline levels with the transition to chronic infection, then were maintained throughout follow-up (Figure 3b).
Finally, in RMs, a 300% increase from the baseline mDC counts occurred during early chronic infection, coincident with the control of viral replication (Figure 3c). Once the VLs were controlled, the circulating mDCs declined in RMs but remained significantly increased compared to baseline values long after the viremia was completely controlled (Figure 3c).
Our results show that the degree of mDC recovery during chronic infection rather than their acute depletion predicts disease progression in SIV-infected NHPs. These results corroborate previous studies showing that higher peripheral mDC counts are associated with slower disease progression in SIVmac-infected RMs [7].
We further investigated the mechanisms responsible for the distinct dynamics of peripheral mDCs observed in progressors vs nonprogressors vs controllers.
To determine if mDC depletion is due to direct virus killing, we first correlated the absolute mDC counts with the VLs in each species between 8 and 120 dpi. The use of the same virus to infect these three distinct NHP species eliminates the variables related to the ability of different SIV/HIV to infect and thus directly kill mDCs. A strong negative correlation (using mixed-effects models) was found in AGMs between higher VLs and lower mDC counts (p = 0.006) (Figure 3e), yet we could not establish any significant association of the same parameters in PTMs (p = 0.43) (Figure 3d) and RMs (p = 0.15) (Figure 3f) infected with SIVagm.
To assess the role of virus killing in inducing mDC loss, we further sorted mDCs from the LNs collected from all three species and showed that mDCs from PTMs, AGMs and RMs do not harbor detectable levels of SIVagmSab DNA (data not shown). Conversely, a cellular fraction containing Lineage+ cells (and representing a mixture of CD3+ T cells and B lymphocytes) harbored high amounts of virus in all the three species included (data not shown). This result is not surprising, as the recent description of SAMHD-1 restriction factor suggested that dendritic cells cannot be infected by primate lentiviruses that do not contain a vpx gene, as is the case of SIVagmSab92018 [64].
We next investigated if differences in mDC mobilization from the bone marrow in SIVagmSab-infected PTMs, AGMs and RMs can explain the variation in numbers of circulating mDCs in these three species.
Since mDCs rarely undergo division in the periphery and tissues [2] and Ki-67 antigen is not expressed by cells in the resting phase (G0), Ki-67 expression by circulating mDCs is indicative of active or recent cell division, i.e., production and mobilization from bone marrow, the primary site where mDCs undergo cell division and differentiation [65]. Overall, there was a significant difference in the dynamics of Ki67+ mDC in circulation among the three species (p = 0.0054).
Ki-67 expression by peripheral mDCs increased in the PTMs throughout SIVagmSab infection (Figure 4a), thus excluding the possibility that loss of mDC in this species is due to their lack of mobilization from the bone marrow. Even rapid progressor PTMs that had lower mDC counts at baseline showed a clear tendency to restore these cells in the periphery through increased mDC mobilization from bone marrow (Figure S3). This shows that the massive mDC loss is not due to a defect of bone marrow specific to the rapid progressors, but rather to other factors.
Increased mDC mobilization from BM was also observed in AGMs (Figure 4b) and RMs (Figure 4C) in the early stages of SIVagm infection and was maintained throughout the follow-up. Interestingly, the highest mDC mobilization from bone marrow was observed in the RM controllers (Figure 4C), which may be the mechanism behind the increased mDC counts observed in SIVagm-infected RMs.
However, with the exception chronic RM infection, in which significant increases in Ki-67 expression by mDC are accompanied by significant increases in mDC counts (Figure 3c), there were no other direct positive correlation between the degree of mDC mobilization from bone marrow and the peripheral mDC counts in the other species.
Our results indicate that mDC mobilization from bone marrow occurs very early postinfection in all pathogenic scenarios and is maintained unaltered even during progression to AIDS. Yet, this increased mDC mobilization alone is not sufficient to preserve healthy counts of peripheral mDC in SIV-infected NHPs, and most likely other factors play a significant accessory role in the preservation or loss of these cells from the circulation.
To determine whether or not mDC recruitment to LNs and intestine is responsible for their loss from circulation and to establish if this outcome drives disease progression, we first used flow cytometry to assess mDC dynamics in the LNs and intestine during acute and chronic SIV infection in the three types of SIV infection. We found differences in the changes in mDC frequency in LN over time across the species (p = 0.0162), as well as trends toward a difference in the intestine (p = 0.0645). Assessment of the mDC frequency in the LNs and intestine showed that they decreased during acute infection in PTMs in both compartments (Figure 5a and 5d) and were only partially restored during chronic infection. While mDC depletion in the intestine did not reach statistical significance during chronic infection of PTMs included in this study (Figure 5d) or in a larger cohort including PTMs from our previous studies (Figure S4a), a subset of mDCs expressing CD103+ were preferentially lost at the mucosal sites in chronically-infected PTMs (Figure S4b). This result corroborates data previously reported for other pathogenic SIV infections [60]. No significant changes in mDC counts occurred in the LNs in AGMs and RMs (Figures 5b and c), while mDC percentages were significantly increased in the intestine in these two species at different time points of SIVagmSab infection (Figures 5e and f).
These immunophenotypic results were confirmed by immunohistochemistry (IHC) for CD11c (Figures S5, S6, S7 and S8). In general, there was a good match between IHC and flow cytometry data, indicating acute mDC depletion in the LN (Figure S5 a–c) and intestine (Figure S6) in the progressive infection of the PTMs, increased during chronic infection in the gut of nonprogressor AGMs (Figure S7 d–i) and increased during acute infection in RM controllers (Figure S8 d–i).
IHC showed that, in the intestine, the majority of the CD11c+ cells were located in the Peyer's Patches (Figure S7 d–f) and in the lymphoid aggregates (Figure S8 d–f), with only a small fraction of mDCs being present in the lamina propria (Figures S6a–c, S7g–I and S8d–i). This points to potential differences in CD11c dynamics in different gut segments (i.e., inductive vs effector sites). Similarly, IHC results revealed differences in the dynamics of CD11c expression between superficial and mesenteric LNs in the pathogenic infection of PTMs (Figure S5).
Our data showing increases in the mDC population in the intestinal compartment in persistent nonprogressive SIVagmSab infection in AGMs and in controlled SIVagmSab infection in RMs, suggest that mDC recruitment to these sites is not deleterious for SIV/HIV pathogenesis. In contrast, loss of mDCs from mucosal sites was correlated with disease progression in SIVagmSab-infected PTMs.
The decreased numbers of mDCs in the LNs and intestine of PTMs may result from a lack of recruitment to these tissues. Yet, increased apoptosis of mDCs from circulation and LNs was previously reported in progressive HIV/SIVmac infections [7]. Consequently, we could not rule out mDC loss from lymphoid and mucosal tissues through a similar mechanism, and we assessed additional markers of mDC migration and recruitment to lymphoid sites.
We measured Ki-67 expression on mDCs isolated from LN and intestine as a marker of their recent mobilization at these sites. We found significant differences over time across the three species, both in LN (p = 0.0001) and in the intestine (p = 0.0319). Ki-67 expression rapidly increased on mDCs in the LNs and was maintained at high levels throughout infection in all three species, the highest increase being observed in AGMs (Figures 6a, b, c). Slight, but consistent increases of Ki-67 expression by mucosal mDCs were observed in AGMs and RMs (Figures 6e, f), especially during chronic SIV infection. Increased mDC mobilization to the intestine may explain the increased numbers of mDCs observed at this site in nonprogressive infections. Surprisingly, however, the most prominent increase in Ki-67 expression occurred in the intestine of SIVagmSab-infected PTMs, starting from the acute infection throughout the follow-up (Figure 6d), suggesting that mDCs are also recruited to mucosal site in progressive infections. This observation is also supported by the observation of significant increases in α4β7 integrin expression on circulating mDCs isolated from both AGMs and PTMs (Figure S9), confirming their recruitment to the gut in the progressive infection [24]. Low levels of mDCs in the gut despite their documented massive recruitment indicate that additional factors are responsible for mDC depletion in the intestine of SIVagmSab-infected PTMs.
To complete the characterization of mDC migration to either LNs or mucosal tissues, we examined the expression of chemokine receptors on circulating mDCs. We reasoned that, if mDC recruitment to LNs and tissues is driven by specific chemokine signaling, circulating mDCs that migrated to certain effector sites should express the necessary chemokine receptors to respond to chemokines from tissues or LNs. We thus measured the levels of CCR7, a receptor for two chemokines (CCL19 and CCL 21) usually expressed in the LNs, [66]–[68] and CCR5, a receptor for chemokines (CCL3, CCL4 and CCL5) usually elevated in inflamed mucosal tissues [69], to assess mDC migration to either LNs or mucosal tissues.
In pathogenic SIV infection of PTMs, expression of both CCR5 and CCR7 chemokine receptors increased on peripheral mDCs, with a more prominent increase occurring for CCR5 (Figures 6g and j).
In the nonpathogenic SIV infection of AGMs, very high CCR7 expression, but lower CCR5 expression was observed in the peripheral mDCs (Figures 6h and k). The decreased CCR5 expression on circulating mDCs from SIVagmSab-infected AGMs may suggest their lower recruitment to intestine in AGMs vs PTMs. However, the increases observed for all the other mobilization/recruitment markers and the observation that mDCs are increased in the intestine support recruitment of mDCs to the gut in AGMs. Finally, in the controlled SIVagmSab infection of RMs, circulating mDCs showed transient but significant increases of both CCR7 and CCR5 expression (Figure 6i and l).
Altogether, these results strongly support recruitment of mDCs to LNs and intestine in all the pathogenic scenarios.
Two lines of evidence suggest that significant mDC apoptosis occurs in the pathogenic SIV infection of PTMs: (a) increased mDC mobilization from bone marrow (increased Ki-67), yet decreased levels of circulating mDCs; and (b) migration of activated mDCs to the intestine (increased CCR5 and α4β7 expression on circulating mDCs and increased Ki-67 expression on intestinal mDCs), without consistent increase of mDCs in the intestine. As a result, we hypothesized that while mDCs are mobilized to mucosal tissues, once they arrive at this particular compartment they undergo apoptosis, which offsets their influx. To test this hypothesis, we examined the expression of activated Caspase-3 by mDCs from both intestine and peripheral blood. We gated on live mDCs that expressed activated Caspase-3 to specifically identify the cells that undergo apoptosis (Figure 7a). Circulating mDCs from PTMs expressed increased levels of activated Caspase-3 during SIV infection (Figure 7b). Conversely, activated Caspase-3 expression of mDCs did not significantly change at any time during SIV infection in AGMs or RMs (Figure 7b). Similar kinetics of apoptotic mDCs were observed in the intestine, with activated Caspase-3 expression being increased during acute and chronic infection in PTMs, while remaining unchanged throughout infection in AGMs and RMs (Figure 7c). In conclusion, our data clearly show that mDC loss in progressive infections results through bystander apoptosis, which is particularly severe at mucosal sites.
Our further goal was to determine if mDC loss observed during progressive SIV infection in PTMs is due to immune activation-induced bystander apoptosis. We therefore assessed expression of CD80 and CD95 receptors on mDCs isolated from blood and tissues from PTMs, AGMs and RMs. Both CD80 and CD95 have been reported to increase during activation of mDCs [70], [71]. Meanwhile, mDCs that express CD95 can potentially undergo apoptosis [72].
We found that immune activation of mDCs is global and sustained in PTMs, with both CD80 and CD95 being significantly increased throughout follow-up (Figures 8a and b). mDC activation was less significant in AGMs, and limited to increases in CD80 expression during acute and, to a lesser extent, chronic infection, while increases in CD95 expression were not significant in this species (Figures 8a and b). Finally, in SIVagmSab-infected RMs CD80 expression increased on circulating mDCs throughout acute infection and returned to baseline levels with the passage to chronic infection. Similarly, increased CD95 expression on circulating mDCs paralleled that of CD80 in RMs, being limited to the acute infection and returning to baseline during chronic infection (Figure 8b). Interestingly, normalization of mDC activation occurred before complete control of SIVagmSab infection in RMs (Figures 8a and b), in contrast to CD4+ T cell activation that normalizes long after the control of viral replication [56].
Collectively, our results identified distinct patterns of mDC immune activation in progressive, nonprogressive and controlled SIV infections that generally parallel the degree of mDC apoptosis. This point to a direct association between the degree of immune activation of innate effectors and their death. This mechanism has also been reported to account for at least a fraction of the CD4+ T cell loss during progression to AIDS in pathogenic HIV/SIV infection [37].
While massive mDC recruitment to intestinal sites occurred in all SIVagmSab disease models, apoptosis associated with increased immune activation of mDCs in both gut and periphery was detected only in the progressive model. We hypothesized that the mDC overstimulation and death unique to SIVagmSab-infected PTMs relies on the severe gut dysfunction, which is characteristic of this progressive infection model. It was reported that translocation of microbial products in the lamina propria and in the general circulation occurs in PTMs [73], where they may enhance mDC stimulation in conjunction with the virus. In contrast, the integrity of the gut is maintained in nonprogressive and controlled infections [73], [74], thus exposing mDCs only to SIV stimulation. We addressed this hypothesis by comparing the ability of LPS (as a surrogate for microbial products) and R848, a TLR7/8 agonist that stimulates mDCs through TLR8, similar to SIV/HIV [75], [76] (as a surrogate for the virus), to induce immune activation and apoptosis in mDCs isolated from PTM blood prior to SIVagmSab infection. This experiment showed that increased immune activation (documented as increased CD80 expression) (Figures 9a and c) and apoptosis [measured as increased expression of Annexin V (AnnV) on the surface of live mDCs] (Figures 9b and c) occurred after stimulation with both LPS and R848. LPS appeared to be a more potent inducer of mDC apoptosis, as shown by the higher levels of AnnV expression on mDCs stimulated with this TLR4 ligand (Figure 9b and c). Altogether, our findings suggest that the combined action of SIV and translocated microbial products may result in excessive activation and apoptosis of circulating and mucosal mDCs in progressive SIV infections.
To assess the impact of virus stimulation on mDC function, we compared the cytokine production of mDCs prior to and at critical time points during progressive, nonprogressive and controlled SIV infections. mDCs were isolated and stimulated with R848 [77]. This experiment showed that, within the same species, mDCs from acutely and chronically SIVagmSab-infected monkeys responded to TLR7/8 stimulation by decreasing IL-6 (Figure 10a) and TNF-α production (Figure 10b). The decreased production of proinflammatory cytokines in response to viral stimulation occurred in all three NHP species, independent of the pathogenic outcome of SIVagm infection. Note however, that AGMs exhibited the lowest production of proinflammatory cytokines both before and after SIV infection (Figure 10a and b). These results are in agreement with those previously reported in HIV infection [60], [78] and suggest that SIV infection per se does not trigger proinflammatory cytokine production by mDCs.
Since microbial translocation (MT) significantly increases only in pathogenic SIVagmSab infection of PTMs and not in nonprogressive SIVagmSab infection of AGMs and RMs [42], [56], we reasoned that mDCs might be stimulated to produce proinflammatory cytokines by the microbial products translocated from the gut during pathogenic SIV infections. We assessed TNF-α production in response to LPS stimulation of peripheral mDCs isolated from PTMs prior to SIV infection and during chronic infection. Our measurements revealed that TNF production is indeed increased in LPS-stimulated mDCs isolated from chronically SIV-infected PTMs (Figure 11), indicating that mDCs may contribute to the increased immune activation and inflammation observed during progressive SIV/HIV infection.
To determine if mDCs contribute to the gut dysfunction described in progressive HIV/SIV infections, we compared the spontaneous production of proinflammatory cytokines (TNF-α and IL-6) by unstimulated mDC isolated from the gut in normal vs. acutely and chronically SIVagmSab-infected PTMs. We also compared and contrasted these results with those obtained from assessment of the function of unstimulated intestinal mDCs isolated from normal, acutely and chronically SIVagmSab-infected AGMs and RMs which lack intestinal dysfunction [56], [73]. No significant difference in cytokine production was observed in any of the three models between acutely infected and uninfected animals (Figure 12).
TNF-α and IL-6 production by intestinal mDCs isolated from chronically SIVagmSab-infected PTMs was significantly greater than that of intestinal mDCs from uninfected PTMs (Figure 12a and d). No modifications in mDC TNF-α and IL-6 production could be detected in chronically infected AGMs (Figure 12b and e) and RMs (Figure 12c and f).
Our results suggest that intestinal mDCs play a significant role in the etiology of the gut dysfunction characteristic of progressive infection. Furthermore, increased mDC production of TNF-α in the intestine is not beneficial and is associated with a poor clinical prognosis, in agreement with previous reports [79]. Finally, these data support our hypothesis that mDC overstimulation and increased proinflammatory cytokine production occurs only in the animal model in which microbial products are translocated into the lamina propria.
In this study, we assessed the role of mDCs in the pathogenesis of HIV/SIV infection and AIDS. The rationale for focusing on this cell subset is that mDCs are major regulators of immune activation and inflammation [2], [3], two major determinants of HIV disease progression [80]. A limitation of previous mDC studies is that they have exclusively focused on pathogenic HIV/SIV infections in which a major depletion of circulating mDCs occurs, thus precluding the assignment of a major role to this subset in driving or containing disease progression. Conversely, comparing and contrasting the fate of mDCs in pathogenic, nonpathogenic and controlled SIV infections allowed us to dissect the role of mDCs in SIV pathogenesis.
For these comparative studies, we used three different NHP models developed in our lab, in which the same viral strain, SIVagmSab, induces different pathogenic outcomes: persistent progressive (pathogenic) infection in PTMs [81], [82], persistent nonprogressive (nonpathogenic) infection in AGMs [61], [83] and completely controlled infection in RMs [56]. The use of the same SIVagmSab strain to produce these different pathogenic outcomes eliminated the interference of viral factors on pathogenesis while permitting the targeting of host responses with significant impact on the outcome of infection. Thus, our biological model allows us to complement the existing information regarding the fate of mDCs during pathogenic infection with data from nonpathogenic and elite-controlled infections and to address two fundamental questions: (i) what is the mechanism of mDC depletion in HIV/SIV infection? and (ii) what is the role of mDCs in the pathogenesis of SIV infection? We performed a stepwise approach to address these key issues:
We report that, in agreement with previous reports [7], a significant drop in peripheral mDC counts occurs during the pathogenic SIV infection of PTMs. However, our comparative analysis also identified a similar depletion of circulating mDCs in the nonpathogenic infection of AGMs. As such, our study suggests that the magnitude of the acute depletion of circulating mDCs is not predictive of SIV disease progression. Similarly, while it was previously reported that mDC counts at the setpoint are predictive of the outcome of infection [84], our results could not establish such a correlation.
In our study, mDC levels during chronic infection appear to be associated with the control of disease progression. Thus, in RMs, mDC levels increased over baseline levels coincident with the control of viral replication. Altogether, these features point to mDC restoration as one of the mechanisms through which control of SIV infection is achieved.
Our results suggest that mDC depletion from circulation is not due to direct virus killing: (i) The three species exhibited similar high levels of viral replication during acute infection (Figure 1), yet mDCs were depleted only in PTMs and AGMs, but not in RMs. (ii) The kinetics of acute mDC depletion from circulation significantly diverged between PTMs and AGMs, in spite of similar dynamics of viral replication. (iii) mDC depletion occured prior to detectable viremia in PTMs. (iv) SIVagmSab does not have a vpx gene and therefore cannot infect dendritic cells because of the SAMHD1 restriction [64], as documented in our study that failed to identify viral DNA in sorted mDCs from SIVagmSab-infected PTMs, AGMs and RMs. By documenting mDC depletion in a biological system that uses a virus that does not infect mDCs, our study demonstrates that factors other than direct virus killing are responsible for mDC depletion during SIV/HIV infection.
We assessed mDC mobilization from the bone marrow by measuring Ki-67 expression. For the same purpose, other groups have used thymidine analogues such as Bromodeoxyuridine (BrdU), which becomes incorporated into dividing mDCs. BrdU assesses mDC mobilization from the bone marrow and tracks the cells to tissues. However, BrdU induces mutations, limiting its use in long-term studies in large animal models [85]. Furthermore, studies reported that BrdU incorporation and Ki-67 expression in DCs have a similar value [86]. Therefore, we reasoned that measuring the expression of cell proliferation markers by mDCs is appropriate to assess trafficking in long-term studies and postulated that Ki-67 expression in tissue mDCs is due to rapid recruitment of cells that were recently mobilized into circulation from the bone marrow.
We report that mDC mobilization from the bone marrow occurs in pathogenic, nonpathogenic and controlled SIV infections. Increased mobilization was observed throughout the follow-up and did not appear disrupted during the terminal stages of infection in the animals that progressed to AIDS. Furthermore, rapid progression to AIDS was not associated with a defect in mDC mobilization from the bone marrow. As such, our study failed to document any evidence that insufficient mobilization of mDCs from the bone marrow is a factor driving the drop of circulating mDCs.
We took a series of steps to carefully assess mDC migration to the LNs and the gut. By immunophenotyping the mDCs from the LNs and intestine, we showed that increases of mDCs in the LNs and intestine occur in persistent nonprogressive SIVagm infection in AGMs and to a lesser extent in controlled SIVagm infection in RMs, suggesting mDC recruitment to these sites. Conversely, transient loss of mDCs from LNs and intestinal mucosal sites only occurred in SIVagm-infected PTMs. To explain this discrepancy between progressive and nonprogressive infections, we further assessed Ki-67 expression by mDCs at the LNs and mucosal sites, as a sign of their recent mobilization, and we showed that Ki-67 expression increased in all models throughout infection, suggesting massive mDC influx at these sites. Next, we assessed mDC surface expression of the chemokine receptors involved in cell mobilization to LNs [66]–[68] and mucosal [69] tissues to further document mDC mobilization to these sites in the three types of infections. Increased levels of CCR5 and CCR7 documented a massive influx of mDCs to both LNs and mucosal tissues in the pathogenic infection of PTMs. Conversely, in the nonpathogenic infection of AGMs, a low expression of CCR5 by the mDCs recapitulates previous reports in this species documenting a similar low CCR5 expression by CD4+ T cells [44], [45] and suggests a reduced mDC migration to the intestine in this animal model. Meanwhile, transient increases of both CCR5 and CCR7 pointed to a brief mDC mobilization to both LNs and intestine in SIVagm-infected RMs and showed that mDC trafficking to these tissues returns to baseline when viral replication is controlled. We also assessed the expression of the intestine homing marker α4β7 integrin on mDCs and documented their migration to the gut in PTMs and AGMs. Finally, the flow cytometry results were confirmed by IHC. Altogether, our data demonstrate here for the first time that mDC mobilization to the LNs and mucosal tissues occurs in progressive, nonprogressive and controlled SIVagmSab infections and is a major factor responsible for their depletion from circulation.
mDCs were either depleted or not significantly increased from the intestine during pathogenic infection, in spite of continuous recruitment to this site. As such, we assessed the expression of apoptosis markers by mDCs in progressive SIV infection of PTMs and demonstrated that mDC depletion is indeed due to increased programmed cell death. By comparing the results of in vitro stimulation of mDCs isolated from uninfected PTMs with R848 (to mimic virus stimulation) and LPS (a surrogate for microbial products that are abundantly present in the lamina propria and general circulation in SIV-infected PTMs), we found that mDC hyperactivation and death is due to microbial products rather than the virus. These data corroborate our findings that a lower degree of mDC activation and no significant increase in apoptosis are associated with lack of MT, rather than low viremia, in the nonprogressive infections.
While it is difficult to extrapolate in vivo the in vitro data and conclusively establish which is the major cause of mDC hyperactivation and death during SIV infection, our data supports a view in which a synergistic SIV and microbial mDC hyperstimulation are responsible for the depletion of this immune cell subset in pathogenic HIV/SIV infections. Conversely, in nonprogressive and controlled infections, mDC apoptosis may be kept at bay because, in these models, there is no MT associated with SIVagmSab infection and mDC stimulation occurs predominantly through direct virus action.
This is a key question addressed by our study. Our complex approach allowed us to establish four lines of evidence in support of a positive role for mDCs in SIV infection: (i) We established a direct correlation between mDC counts and prognosis of SIV infection: animals with the lowest mDC levels prior to infection were rapid progressors; mDC depletion from blood and intestine was associated with disease progression; animals that completely recovered mDCs during chronic infection experienced delayed or no disease progression. (ii) While mDC mobilization from the bone marrow occurred in all the NHPs included in our study, increased mobilization was associated with a lack of disease progression during the follow up (Figure 4), suggesting possible mDC involvement in virus control. (iii) We documented mDC mobilization to LNs and intestine in all three species in response to SIV infection and showed that increased levels of mDCs in the intestine and LNs are associated with lack of disease progression/control of infection, while mDC loss at these sites is associated with progression to AIDS. (iv) The comparison between progressive, nonprogressive and controlled infections revealed that mDCs are not inducing immune activation and inflammation in the biological systems in which SIV replication is not associated with MT. Thus, mDC accumulation in the intestine in nonprogressive and controlled SIVagmSab infection of AGMs and RMs, respectively, is not associated with increases in immune activation and inflammation in these species. This is supported by our findings that peripheral mDC production of proinflammatory cytokines in response to TLR7/8 ligands after SIV infection is decreased in these species. Adequate preservation of mDCs may impact immune activation and inflammation in SIVagmSab-infected AGMs and RMs through additional pathways. It was recently reported that subsets of mDCs may play a significant role in inducing differentiation of Tregs, which may in turn control immune activation and promote tolerance to SIV infection [87]. Altogether, these features strongly support a role of mDCs in reducing the levels of immune activation and inflammation, maintaining gut integrity and possibly inducing tolerance to SIV in nonprogressive infections.
Due to their depletion from both circulation and the intestine, it is virtually impossible to precisely define the role of mDCs in progressive infection. While our studies showed that mDCs isolated from PTMs show a similar hyporesponsiveness postinfection to TLR7/8 stimulation compared to nonprogressive infections, mDC activation was significantly higher in PTMs. Our results show that mDC hyperactivation and production of proinflammatory cytokines may be enhanced in progressive SIV infection by microbial products translocated to the circulation as a result of the severe SIV-associated gut dysfunction in these species. Finally, we documented increased TNF-α and IL-6 production by intestinal mDCs isolated from PTMs (Figure 12), suggesting an association between increased inflammatory cytokine production and poor clinical prognosis. Altogether, our results show that during progressive SIV/HIV infections, the beneficial role of mDCs may be obscured by either their excessive loss and/or by hyperfunction reflected in an overproduction of proinflammatory cytokines by residual mDCs in response to microbial products or opportunistic pathogens concurrent with HIV/SIV infection.
In conclusion, we report that mDCs have different kinetics, immune activation/apoptotic patterns and functions in pathogenic, nonpathogenic and controlled SIV infection. Our results provide a mechanistic basis of the role of mDCs in the pathogenesis of AIDS. These results are informative for designing both therapeutic interventions and vaccinations aimed at controlling HIV infection.
All animals were housed and maintained at the University of Pittsburgh according to the standards of the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC), and experiments were approved by the University of Pittsburgh Institutional Animal Care and Use Committee (IACUC). These studies were covered by two IACUC protocols: 0907039/12080831 Animal Model for SIV Infection Control (Approved in 2009 and renewed in 2012); 0911844/12121250 Pathogenesis of SIV in African green monkeys (Approved in 2009 and renewed in 2012). The animals were fed and housed according to regulations set forth by the Guide for the Care and Use of Laboratory Animals and the Animal Welfare Act.
Four AGMs, five RMs and five PTMs were included in the study. All animals were infected with plasma equivalent to 300 tissue culture infectious doses (TCID50) of SIVagm.sab92018 [61] and followed for up to one year or until progression to AIDS. Animals were clinically monitored throughout the follow-up. Plasma VLs were quantified by real time-PCR as previously described [61].
Blood was collected from all animals at the following time points: prior to the infection (0 dpi), during acute SIVagm infection to closely monitor the innate responses (1, 3, 4, 5, 8, 14, 21 and 28 dpi), around the viral setpoint (35 and 42 dpi), and during chronic infection (60, 72, and 120 dpi).
LN biopsies were performed prior to infection, at the peak of viral load (during acute infection), and at necropsy (during chronic infection). Intestinal resections (5–10 cm) were surgically performed prior to infection, during acute infection and during chronic infection, as previously described [42], [62]. Additional intestinal samples were collected at the necropsy.
Within one hour after blood collection, plasma was harvested and peripheral blood mononuclear cells (PBMCs) were separated from the blood using Ficoll density gradient centrifugation. Lymphocytes from the intestine and LNs were isolated and stained for flow-cytometry, as previously described [42], [61], [62]. Lymphocytes were isolated from the axillary or inguinal LNs by gently mincing and pressing tissues through nylon mesh screens. Intestinal resections were processed as described previously to obtain an enriched mononuclear cell suspension [42], [61], [62], [88]. Briefly, intestinal resections were minced mechanically, washed with EDTA and subjected to collagenase digestion followed by Percoll density gradient centrifugation.
PBMCs, LNs and intestinal cells from PTMs, AGMs and RMs infected through the same route and with the same dose of SIVagm.sab as in previous protocols were also included to match infection time points when samples from the present study were not available, or to increase the number of animals and improve statistical significance.
Whole peripheral blood, LN cell suspensions and intestinal mononuclear cell suspensions were stained with fluorescently-labeled antibodies to Lineage markers CD3 (clone SP34-2; all antibodies from BD Bioscience, San Jose, CA, USA unless otherwise noted), CD14 (M5E2), CD163 (GH1/6, only in LN and intestine) and CD20 (2H7)]; HLA-DR (G46-6), CD11c (S-HCL-3), CD103 (2G5) CD45 (D058-1283), CD80 (L307.4), CD86 (FUN-1), CD95 (DX2), CCR5 (3A9), CCR7 (3D12, eBioscience, San Diego, CA, USA) and α4β7 (FIB504). An amine-reactive fixable dead-cell dye (Invitrogen, Grand Island, NY, USA) was used to distinguish live from dead cells. For intracellular staining, cells were stained to identify live mDCs as described above, then fixed, permeabilized and stained for activated Caspase-3 (Cas-3, C92-605) and Ki-67 (B56). Flow cytometry acquisitions were performed on an LSR II flow cytometer and analyzed with FlowJo software (Treestar, Ashland, OR, USA).
mDCs were defined as CD45+ (only in peripheral blood) Lineage negative (Lineageneg) HLA-DR+ cells expressing CD11c (Figure 2). In the LNs, a broad Lineageneg HLA-DR+/++ gate was used to include all mDCs, as described previously (Figure 2). mDC populations positive for CD80, CD86, CD95, CCR5, CCR7, α4β7 and Ki-67 were identified by staining with an isotype control antibody. Cas-3+ mDCs were identified as a distinctly separate population.
A two-step TruCount technique was used to enumerate mDCs in blood, as previously reported [89]. The number of blood CD45+ cells was quantified using a precise volume of blood stained with antibodies in TruCount tubes (BD Biosciences) that contained a defined number of fluorescent beads to provide internal calibration. The number of blood mDCs was then calculated based on the ratio of mDCs to CD45 cells in whole blood at the same time point.
Cryopreserved single-cell preparations of superficial lymph nodes collected at time of necropsy from various protocols were thawed and stained with CD3/C20-PE, CD14-FITC, HLA-DR-APC-Cy7, CD11c-APC, CD123-PE-Cy7, Live/Dead-Yellow Dye L-34959 from Molecular Probes (Eugene, OR) for 30 minutes at 4°C; then washed once with and resuspended in PBS supplemented with 0.5% BSA and 2 mM EDTA. mDCs (Lineageneg CD14neg HLA-DR+ CD123neg CD11c+) along with Lineage+ cells, were sorted on a 3-laser FACSAria instrument using FACSDiva 6 software (BD Biosciences) and collected in polystyrene tubes containing RPMI+20% heat-inactivated FBS. The sorted cells were pelleted and frozen, and DNA was extracted with the DNeasy Blood and Tissue Kit (Qiagen, Germantown, MD). Viral copy number was determined by qPCR of extracted DNA, as previously described [56].
IHC was performed on formalin-fixed, paraffin-embedded tissue samples. Four µm thick sections were deparaffinized, rehydrated, and rinsed. For antigen retrieval, the sections were microwaved in Vector Unmasking Solution and treated with 3% hydrogen peroxide. Sections were incubated with CD11c monoclonal primary antibody (Novacastra, USA). Secondary antibodies were from Vector Vectastain ABC Elite Kit. For visualization, sections were treated with DAB (Dako) and counterstained with hematoxylin.
mDC activation and apoptosis was measured after stimulation of total PBMCs for 24 hours with 10 µM of the TLR7/8 ligand R848, (Invivogen, San Diego, CA, USA) or with 100 ng/ml Escherichia coli lipopolysaccharide (LPS), (Invivogen, San Diego, CA, USA). Cells were harvested and stained for dendritic cell markers, CD80 and AnnV. An amine-reactive fixable dead-cell dye (Invitrogen, Grand Island, NY, USA) was used to discriminate live from dead cells. Cells cultured in the absence of R848 and LPS were used as background control.
Intracellular cytokine production by isolated mononuclear cells from blood and intestine was measured as described previously, with slight variation [7]. Briefly, cells were cultured for seven hours with 10 µM of the TLR7/8 ligand R848 (Invivogen, San Diego, CA, USA) or with Escherichia coli lipopolysaccharide (LPS; 100 ng/ml; Invivogen, San Diego, CA, USA) with and without the addition of 10 µg/mL brefeldin A (Sigma, St. Louis, MO, USA) after two hours. Cells were stained with surface-labeling antibodies as above then fixed and permeabilized prior to incubation with antibodies to TNF-α (MAb11), IL-6 (MQ2-6A3) and IL-12 (C8.6, eBioscience) and analyzed by flow cytometry. Cells (mDCs) cultured with R848 or LPS stimulation but without the addition of brefeldin A were used as background control.
In each species, postinfection time point values for each parameter were compared with pre-infection values using the Mann-Whitney U test. GraphPad Prism 5 (GraphPad Software) was used for statistical analysis. Correlations were determined using the non-parametric Spearman rank test. Differences in temporal dynamics were analyzed using mixed-effects models, with macaque as the grouping factor to account for the repeated measurements made in each animal. For these analysis we used the nlme package [90] of R (http://cran.r-project.org/). All P<0.05 values were considered to be significant.
|
10.1371/journal.pcbi.1003168 | Finding Associations among Histone Modifications Using Sparse Partial Correlation Networks | Histone modifications are known to play an important role in the regulation of transcription. While individual modifications have received much attention in genome-wide analyses, little is known about their relationships. Some authors have built Bayesian networks of modifications, however most often they have used discretized data, and relied on unrealistic assumptions such as the absence of feedback mechanisms or hidden confounding factors. Here, we propose to infer undirected networks based on partial correlations between histone modifications. Within the partial correlation framework, correlations among two variables are controlled for associations induced by the other variables. Partial correlation networks thus focus on direct associations of histone modifications. We apply this methodology to data in CD4+ cells. The resulting network is well supported by common knowledge. When pairs of modifications show a large difference between their correlation and their partial correlation, a potential confounding factor is identified and provided as explanation. Data from different cell types (IMR90, H1) is also exploited in the analysis to assess the stability of the networks. The results are remarkably similar across cell types. Based on this observation, the networks from the three cell types are integrated into a consensus network to increase robustness. The data and the results discussed in the manuscript can be found, together with code, on http://spcn.molgen.mpg.de/index.html.
| Nucleosomes are protein complexes around which the DNA is wrapped for compactness. They are made of histone proteins that can be post-translationally modified and these histone modifications can affect the expression of surrounding genes. In the past decade, scientists have developed a strong interest in the part of gene regulation provided by epigenetics, i.e. those heritable characteristics that are not based on the DNA sequence and that can therefore be cell-type-specific, such as histone modifications. Striking patterns about the co-occurrence of modifications have been discovered, leading to the hypothesis that different combinations of modifications lead to different effects. Different histone modifications could act jointly to recruit certain proteins, or be required sequentially, which is reflected in statistical dependencies in measured data. The focus of this article is on building a network that represents the global dependencies by extracting direct associations of histone modifications. We find that, although histone modifications patterns are cell-type specific (modifications may not necessarily appear at the same loci), the dependencies are to a large degree cell-type independent, which is supported by a large overlap of the inferred associations in the networks built for different cell types. We are able to find meaningful associations, both known and novel.
| The study of gene regulation is traditionally based on DNA sequence analysis, gene interactions and transcription factor binding events. It has however over the past decade been revolutionized by genome-wide maps of epigenetic marks, specifically DNA methylation and histone modifications. Histone modifications are post-translational modifications of the histone proteins which form nucleosomes by wrapping about 147 base pairs of DNA. These modifications can have effects on biological processes including transcription, DNA repair, splicing, dosage compensation and more [1], [2], either by altering the chromatin structure or by recruiting key proteins [1]. The observation of different histone modifications co-occurring in different contexts has raised the possibility of combinatorial effects and has led to the histone code hypothesis [3], whereby combinations of histone modifications have a biological meaning and lead to distinct downstream effects.
In particular, there has been much evidence for a strong role of histone modifications in the regulation of gene expression [4], [5], not only at promoters and enhancers, but also along the gene body. Many authors have contributed genome-wide pattern analyses of modifications around regulatory regions [6]–[10]. For example, it has been found that acetylation marks generally co-occur with active genes, whereas methylation marks can be associated with active genes or repressed genes, depending on the modified residue. Histone modifications can be clustered according to their average level around promoters into two groups, one group containing active marks and the other repressive marks [7]. Ernst et al. [9] used hidden Markov models to extract genome-wide epigenetic states, many of which can be thought of as characterizing the transcriptional process at various positions along the gene body, or different kinds of enhancers, or splicing or heterochromatin, etc. Although it is still unclear whether they are causes or effects of transcription, these observations clearly demonstrate a connection between different combinations of histone marks and different transcription states. For instance, it is well established that promoters carry H3K4me3 and/or H3K27me3 and that actively transcribed genes carry H3K36me3 [11], whereas enhancers are marked by H3K4me1 and H3K27ac [11], [12]. Histone modifications have even been successfully used to determine the presence of regulatory elements such as promoters or enhancers [11], [13]–[17]. Beyond these qualitative findings, a remarkable quantitative relationship with mRNA expression levels has been demonstrated in [18]. However, so far all of these studies deal with co-occurrence but do not provide insights about associations between histone modifications.
In this article, we are interested in building networks of histone modifications. This is a problem that benefits from relatively few variables (histone modifications) and many samples (genomic regions of interest), allowing the use of rigorous statistical methods. In such networks, nodes represent histone modifications, and edges connections between them. The nature of these connections depends on the construction method used to obtain the network. Other authors, again particularly in the context of promoters, could capture associations using Bayesian networks (BNs) of histone modifications [19]–[23]. They aimed at establishing causal links: which modifications are required for the presence of another one. However claims about causality in BNs are controversial [24]–[27], especially in the presence of hidden confounding factors, which occur quite frequently in biological systems. Additionally, BNs do not allow cycles or feedback mechanisms, which seems unrealistic in biological systems.
The ChIP-seq data currently available represents a summary of the epigenome, averaged over many cells. For each histone modification, the read counts represent the average frequency at which it is found in the population of cells. This has three main implications for the interpretation of the edges. Firstly, it is very hard to make any claims about causality, as temporal information is missing. Secondly, discretization of the read counts is less plausible. Even if a histone modification is either present or absent at a specific region in a specific cell, the read counts represent the average over many cells, and discretizing these averages is no longer meaningful. Thirdly, given that only an average picture is available, it can safely be assumed that various states will be represented in the data and will appear in the network. Being in one particular state will mean highlighting relevant associations and downplaying others, but all associations will be present in the same network. In a way, we expect to infer the wiring of the circuit as opposed to the flow in the circuit, i.e. the statics as opposed to the dynamics. Edges can reflect co-occurence, mutual exclusivity, or they can mean that two modifications occur sequentially as part of the same pathway. We cannot distinguish between these scenarios with the data at hand.
An observed correlation between two variables may either reflect a direct association or an induced association that may be due to a mutual association with a third variable. For example, if the lack of sports generates both a drop in fitness and a bad mood, a correlation between the variables and will be observed when, actually, they are only connected through the variable and do not interact otherwise. The third variable (here ) is often referred to as confounding factor. Confounding factors, which can be accountable for part of the associations between other variables, are often presented as a nuisance - experimental techniques for instance may lead to biases that are undesirable confounding factors - however they need not be. For example, expression level is a confounding factor of great interest. In any case, looking at how apparent associations may be explained away can be very insightful.
Let us suppose we have two variables of interest and . The correlation coefficient is a powerful tool but it cannot distinguish direct associations from those due to confounding factors. The partial correlation coefficient was designed to remedy that very problem [28]. The idea is to subtract from and the information contained in a control group of variables by linearly regressing (resp. ) against , and to keep the residuals (resp. ). We then compute the correlation between and . This correlation is called a partial correlation, written and is a measure of the correlation between and that remains after the explanatory power of is taken out.
Let us assume we have a set of variables , and we compute the correlation matrix such that . Let denote the partial correlation matrix (PCM) that contains the pairwise partial correlations, each using as control the remaining variables, i.e. the matrix such that . Note that, in this framework, each variable in turn is treated as a confounding factor, regardless of its expected biological relevance. A property of partial correlations is that may be obtained by simply inverting, normalizing and negating the correlation matrix [29]–[32]. This procedure, that we will use throughout the study, is a very fast alternative to the linear regressions. It also shows the involvement of all variables in the computation of through the inversion step, as opposed to that is only computed on and .
It is common practice to recover the undirected network connecting these variables by simply building a fully connected network and by removing all edges for which [29]–[32]. This rests on the theoretical grounds that the variables are normally distributed and are linearly related, therefore having is equivalent to having independence between and conditioned on the other variables [29]–[32], which is exactly the requirement for the absence of edge in an undirected network. Such networks are therefore referred to as graphical Gaussian models (GGMs) [29]–[32]. In case the true network is Bayesian (i.e. directed and acyclic) then the GGM will contain the original edges and will connect the parents of a same child. GGMs provide a simple and efficient method, whereby networks can be built in just a few seconds. They have been successfully applied to infer gene regulatory networks, even in the presence of small sample size, and a short review of these applications can be found in [33].
In this study, we propose to focus on edges that represent direct dependencies. We want to draw edges between histone modifications that are directly linked in a pathway or that act together, i.e. whose association cannot solely be explained by confounding factors. We build on GGMs, and put forward a robust method to compute sparse partial correlation networks (SPCNs). To the best of our knowledge, PCNs have not yet been applied to histone modifications. In contrast to gene regulatory networks, here the sample size is very large and the variables are few. Formally, partial correlations require normal distributions. In our work this need is overcome and outliers accounted for by rank-transforming the input data. Sparseness is achieved via a cross-validation scheme. Our SPCNs reveal edges that are symptomatic of direct associations, mutual exclusivities, direct edges in a pathway, indirect edges where the intermediate variable(s) are not available, or collaborative work to produce a third variable.
Zhao's group was one of the first to produce genome-wide profiles for a large number of histone modifications, they did so in CD4+ cells [6], [7]. In the meantime, several other groups have contributed to the Roadmap Epigenomics project [34], a database that now contains data for varying numbers of histone modifications in different cell types. Based on this data, the cell types with the largest number of histone modifications were chosen: CD4+, IMR90 and H1. CD4+ cells are lymphocytes (white blood cells), they are part of our immune system. IMR90 cells are fibroblasts (cells involved in the synthesis of tissues' external structure) in the lung, and H1 cells are embryonic stem cells. 21 histone modifications are available for all three cell types, we keep only those. Histone modification data is obtained via ChIP-seq experiments, so openness of the chromatin is a potential confounding factor to include in the analysis via DNaseIHS, which marks the hypersensitivity of the DNA to the enzyme DNaseI. The relationship of histone modifications to mRNA levels is of particular interest because of the role of histone modifications in transcription, so mRNA data is included. We look at the amounts of ChIP-seq reads for these 23 variables in the [−2000,+2000] around the transcription start sites (TSSs) of known genes, and at the amounts of RNA-seq reads in the exons of those genes. Antibodies can also play a role as confounding factors (because of their cross-reactivity), and may also vary from experiment to experiment. Antibodies are an interesting case because, although they are not semantically “hidden” (we know which ones are used and we know they can cross-react and act as confounding factors), they are technically hidden since we do not know how they cross-react as no data is available. However, we can build a table of cross-reactions and look it up as a possible source of explanation for links between histone modifications. Details about data collection and antibody can be found in Materials and Methods.
We modify GGMs in two respects: first by rank-transforming the input data, and second by enforcing sparseness via a cross-validation scheme. A global view of the algorithm is shown in Figure 1. Precision is favored over completeness: an edge is only found in a network if it is strongly supported by the data. Therefore interpreting edges is favored over interpreting the lack thereof. Details about the computation of the PCMs, the p-values and the q-values can be found in Materials and Methods.
“Explaining away” in machine learning is “a common pattern of reasoning in which the confirmation of one cause of an observed event reduces the need to invoke alternative causes” [37]. We take over this concept and translate it into our own context. A connection between and is explained away by when is negligible compared with , because we assume that was the main cause of the apparent connection between and and that therefore the need to find further causes is alleviated.
When controlling for confounding factors, the partial correlation coefficient is substituted to the correlation coefficients and the difference can be very large. is generally smaller (in terms of absolute value) as it is explained away by the control variables, but it can also be greater as control variables tie and together. For example, if and are independent co-parents of such that , they become dependent upon conditioning on , such that may be different from 0. We would like to know which variables are responsible for most of the change from to . Running an exhaustive search on combinations of about 20 variables is neither possible nor desirable. Instead we condition on a single variable . We repeat the operation for every possible in the dataset and identify the that leads to the biggest discrepancy between and , i.e. the control variable that has the highest impact on the correlation. The impact of all variables is shown for some pairs in
It needs to be established that networks remain stable upon using input data from different experiments or from different cell types. To this end, we define an index of overlap between PCMs, based on the ranking of the entries which represent the associations between pairs of variables. For each PCM (), the pairs of variables are ranked by increasing q-values and the first pairs () are stored in a list. The number of pairs that occur in all lists divided by is a measure of the similarity between all the when pairs are considered. Results are presented in plots where varies from 1 to . The overlap expected at random depends on the number of matrices being compared and on the number of pairs being examined . It is easily computed, as seen in Materials and Methods. For , it follows a hypergeometric distribution, and therefore p-values are directly available.
We now turn to a detailed analysis of the CD4+ network. Note that, the data containing 23 variables, the SPCN has edges maximum. The resulting network is shown in Figure 2a, all the partial correlation coefficients, their q-values and the mask are given in Text S1 Section 7.
Looking at edges around mRNA, we find it is negatively connected to H3K27me3 (a mark of repression) and positively to H3K27ac (a mark of activation), H3K79me2 and H4K20me1 (marks of elongation), which have been, with the exception of H3K27me3, found to be important in predicting expression in CD4+ cells [18]. Interestingly, H3K36me3 has no link to mRNA, in line with [18]. The scatter plots in Text S1 Section 9.1 confirm the lack of relationship. Note that there is no standard correlation either. The data for H3K36me3 is not abundant, very few reads map to the regions of interest. This could come from H3K36me3's preference for exons [39]. Indeed exons are only a small part of the studied region, as shown in Text S1 Section 3, so the lack of connection to expression could be due to poor data, it is hard to tell.
Expected connections are numerous, such as the negative link between H3K27ac and H3K27me3. These two histone modifications are by nature mutually exclusive, and therefore need not be explained by any other histone modification. The strong connections between the various methylation states of H3K4, with H3K4me2 in between, are explained by the fact that these different methylation states are coupled by bidirectional links from H3K4me1 to H3K4me2 and to H3K4me3. Alternatively, it can be explained by antibody cross-reactivity, but it may not be explained by any other histone modification. Connections between DNaseIHS and H3K4me3 and H4K20me1 reflect the need for open chromatin to have transcription.
Finding expected associations is a requirement, however it is more interesting to find unexpected connections. H3K27me3 and H3K9me3 are positively associated (see scatter plots in Text S1 Section 9.2). They have been thought to be mutually exclusive, H3K9me3 encoding constitutive heterochromatin, H3K27me3 facultative heterochromatin. Both would act as repressors but as part of two different processes (involving the PRC1/2 complex for H3K27me3 and the HP1 proteins for H3K9me3), that have been assumed mutually exclusive [40]. Clearly it is not the case here. It has been found that SUZ12, which is part of PRC2 and involved in setting H3K27me3, promotes H3K9 methylation [41], giving a straightforward explanation for our finding. The negative edge between H3K79me2 and H3K4me1 is puzzling given that they are two marks associated with transcription, and that the trend is mostly tue in active genes (see scatter plots in Text S1 Section 9.3). However a possible explanation is that H2BK120ub1, which is required both for the production of H3K4me2/3 and of H3K79me1/2 [42], acts as hidden confounding factor.
Some expected edges exist albeit with an unexpected sign. In particular, H3K4me3 and H3K36me3, associated with initiation and elongation, are positively linked to the repressive mark H3K27me3 (see scatter plots in Text S1 Section 9.4). In fact, for high levels of H3K27me3, this trend already exists in the raw data. This may indicate that some promoters cycle between the repressed H3K27me3 state and the active H3K4me3/H3K36me3 state. The cycling idea of epigenetic states is not without precedent. It has been shown that the estrogen receptor target TFF1 is cyclically methylated and demethylated [43], [44]. In some cells promoters are active (H3K4me3), in some cells they are repressed (H3K27me3), and in some cells they may be bivalent (H3K4me3 AND H3K27me3). All we measure is the population average. If these fluctuations are stochastic, we expect no correlation. However if promoters can move from being active (H3K4me3) to being inactive (H3K27me3) in a regulated manner, then we expect a positive correlation. This could be due to the cell cycle, e.g. promoters get active during S-phase and are rendered inactive thereafter [45]. When looking at the scatter plots in Text S1 Section 9.4, the correlation seems to come from repressed genes, and a little bit from bivalent genes, supporting this hypothesis.
Another example is the negative link between H4K20me1 and H4K5ac (see scatter plots in Text S1 Section 9.5), which seems at first glance counter-intuitive because H4K20me1 is positively linked to expression and acetylations are generally thought to be associated with transcription. This apparent paradox can be resolved by the following reasoning: H4K20me1 is mainly associated with transcription elongation, while acetylations are heavily enriched around the promoter. It has been shown in Drosophila that H4K20me1 recruits the factor RPD3/HDAC1, leading to the deacetylation of H4K [46]. Thus it seems that H4K20me1 helps to prevent cryptic initiation in the transcribed gene body.
Since mechanisms are to a large degree cell-type-independent, the precision and robustness of the results can be increased by integrating information from all available cell types. A SPCN is created for each cell type. Figure 2bc shows the consensus network which contains only those edges that are found in at least two cell-type-specific SPCNs. Light blue edges show negative associations that are found in two cell types, blue edges negative associations found in all three cell types. Pink edges show positive associations that are found in two cell types, red edges positive associations found in all three cell types. It looks very similar to the CD4+ SPCN in Figure 2a. Important associations such as mRNA-H3K27me3, mRNA-H3K79me2, DNaseIHS-H3K4me3, DNaseIHS-H4K20me1 and H3K27ac-H3K27me3 are conserved across cell types. Surprising connection such as H3K27me3-H3K9me3 and H4K20me1-H4K5ac are also stable. The strong connection between H3K4me1 and H4K20me1 is only found in CD4+.
Some of the edges that are common to all networks (marked in bright red and blue) are of particular interest. The antibody table in Text S1 Section 2 (see Materials and Methods) shows that there is antibody cross-reactivity for H3K4's various methylations and for H3K79me1/2. The edges may reflect biologically meaningful associations but may (also or instead) be due to cross-reactions. H3K23ac's antibody reacts with H3K14ac, H3K18ac's with H4K5ac, and H3K27ac's with H3K9ac, which explains partially these three connections. The group H2BK12/20/120ac remains unexplained, however it is plausible that it may be the result of unreported antibody cross-reactions. Other edges that may be explained by antibody cross-reactivity are H4K5ac-H3K27ac and H4K5ac-H3K18ac as well as H3K14ac-H3K18ac.
The explaining away procedure was applied. Text S1 Section 10 shows some of the plots that are obtained for all the edges of interest. Figure 4 summarizes the critical information into one matrix. The colors give the magnitude of the differences between and . If zooming in is available, the numbers on the lower part of the diagonal give the actual difference, and the text on the upper part of the diagonal gives the histone modification that has the most incidence on .
Partial correlations work in such a way that, in order to explain the correlation between and , it is sufficient that a control variable explain . The variable with the most impact then says something about regardless of . Symptomatic of this scenario, the first explanatory variable is then often the same along the column of the matrix corresponding to . For example, in the column associated with H3K27me3, H3K27ac is very often the most influential variable. It can be assumed that H3K27ac explains H3K27me3 and therefore leads to the loss of correlation between H3K27me3 and other variables. H4K5ac seems to explain H3K14ac. This may be due to antibody cross-reactivity, as H4K5ac is often seen in H3K23's column, and H3K14ac's and H3K23ac's antibodies are known to cross-react.
An interesting example that shows how well this procedure works is the pair H3K4me1 and H3K4me3. After glancing at Text S1 Section 8.1 or after zooming into Figure 4, it can be seen that the variable most responsible for the correlation is H3K4me2. This makes a lot of sense biologically, as H3K4me2 is an intermediate state of methylation. Another example is the correlation between mRNA and H3K4me3, which seems to be largely explained by H3K27ac. This maybe due to the fact that H3K4me3 recruits the SAGA complex required for acetylation [47] which puts H3K27ac, which in turn is predictive of mRNA levels, as was seen in [18]. The relationship between H3K4me3 and H4K20me1 is fully explained by DNaseIHS. One possible reason for this is that chromatin openness favors transcription, thereby explaining H3K4me3. The role of H4K20me1 in HDAC recruitment has been demonstrated in the context of chromatin reassembly [46]. Thus it seems that transcription may lead to higher histone turnover, which results in higher levels of H4K20me1.
Similarly to the networks, a consensus effect matrix is shown in Text S1 Section 8.4. It is surprising to see how well the effect of partial correlation and the explanatory variables are conserved across cell types. Indeed, out of 21 possible variables that are all correlated, in most cases the same one comes out in at least two cell types.
We put forward SPCNs, a fast and robust tool, to construct undirected networks of histone modifications. By definition SPCNs can handle continuous data. Moreover they contain all relevant links, and allow for cycles and symmetric relationships. Edges in a SPCN may be seen as controlled associations, where the link between two variables is only established after controlling for potential confounding factors (the other variables at hand). We believe they are the perfect tool for our purposes. The algorithm is designed to maintain a high precision level in the reconstruction of the networks. To be present, an edge must appear in 7 out of 10 sub SPCMs, i.e. be highly supported by the data. Some edges may be missed, and the lack of edges must be carefully interpreted, however given that only 10% of the maximal drop in performance is allowed, we believe that most contributing edges are recovered, and that the lack of edges mainly corresponds to the lack of relevant associations.
We used the availability of data from different experiments and different cell types to our advantage and quantified the variability that could be expected. Firstly, it is interesting to note that the variability across experiments, for the same cell type, is not low. This tends to show that biological data is difficult to reproduce, that results should be interpreted with care, and that evidence may not be overwhelming even though a phenomenon is true. Here, the cell type is the same so it is true that the mechanisms should be the same, yet the evidence is not as high as one might have expected. Secondly, the variability across cell types is marginally higher than the one across experiments, showing that the networks are stable across cell types, and that the variability is mostly due to experimental noise. This last observation is a significant result. Histone-modifications-related mechanisms are often assumed to be the same in all cell types, but it is not systematically checked. Our simulations show that meachanisms are strikingly similar across cell types, almost as similar as two different experiments in the same cell type.
Gathering information on antibody cross-reactivity was difficult but it proved insightful as it revealed important biases in the data. In particular, different methylation states, such as H3K4me1/H3K4me2/H3K4me3 or H3K79me1/H3K79me2, are difficult to distinguish. The edges between such histone modifications may be biologically relevant or/and due to antibodies' lack of specificity, probably both, it is impossible to tell with the data at hand. A similar phenomenon was observed for acetylations. This ought to be a warning for the community. Antibodies are too trusted in many ChIP-seq studies. Instead cross-reactivities should be documented and biases reported when appropriate. In fact, cross-reaction studies are missing for many antibodies, and biases may be more important than we think.
The SPCN gives a global view of the associations between histone modifications, however this view assumes a closed environment containing only the variables in the network. This is an intrinsic limitation of the method. If the set of variables is increased, the new network will not necessarily contain the previous one, all edges might be affected. How much they might be affected depends on the relevance of the variables that are introduced, and on the number of these variables. This makes the network very hard to test experimentally, as the presence of other variables in the cell will make the network by definition obsolete. However such assumptions are not new in biology, where subsets of variables are often chosen, and consequently studied as if they were isolated from the rest of the world.
The effect matrix on the other hand gives a detailed view of what partial correlation does. It shows the difference between the correlation and the partial correlation conditioned on all other variables. In particular, it allows to see which variable causes the highest difference between and . This is of high biological interest, not only because it identifies potential hidden interactions, but also because such effects can be in principle verified experimentally.
Associations of histone modifications are interesting as a first step to understanding their relations. However their connections are not physical and therefore remain abstract. Edges in a SPCN are as direct as possible given the variables at hand, but they can most probably be explained away by enzymes or proteins that float around and provide a physical interface for histone modifications, in particular chromatin modifiers. The next step is therefore to include data for such proteins. Ram et al. have now produced data for chromatin regulators [48]. Including them in the network and particularly in the effect matrix would allow to gain much deeper insight into the physical mechanisms. Further steps should also include transcription factors, and various genomic regions, such as proximal promoters and enhancers.
With two lists of selected pairs from a pool of pairs, the number of common pairs follows a hypergeometric distribution with equal number of white balls and drawn balls () and with a total number of balls of , and a hypergeometric test is appropriate to compute p-values. The probability for pairs to appear in the two lists is obtained through the hypergeometric distribution with successes (white balls) in draws from a finite population of size containing successes (white balls), so . The expected number of same pairs in the two lists is therefore , so the expected proportion is , i.e. a straight line. The p-value is then given by the hypergeometric test: . The appropriate call in R is .
With three lists, things are more complicated. The probability for a pair to appear in the three lists is obtained through a Binomial distribution with number of trials 3 and probability , so . The expected number of pairs common to the three lists is , the expected proportion is therefore , i.e. a quadratic curve. For an observation , the p-value is computed by simulating intersections between three lists containing pairs sampled randomly from with replacement, and by counting the proportion of times the length of these intersections was at least as high as . If the result is 0, is reported as upper bound.
|
10.1371/journal.ppat.0040029 | A Mouse Model for Chikungunya: Young Age and Inefficient Type-I Interferon Signaling Are Risk Factors for Severe Disease | Chikungunya virus (CHIKV) is a re-emerging arbovirus responsible for a massive outbreak currently afflicting the Indian Ocean region and India. Infection from CHIKV typically induces a mild disease in humans, characterized by fever, myalgia, arthralgia, and rash. Cases of severe CHIKV infection involving the central nervous system (CNS) have recently been described in neonates as well as in adults with underlying conditions. The pathophysiology of CHIKV infection and the basis for disease severity are unknown. To address these critical issues, we have developed an animal model of CHIKV infection. We show here that whereas wild type (WT) adult mice are resistant to CHIKV infection, WT mouse neonates are susceptible and neonatal disease severity is age-dependent. Adult mice with a partially (IFN-α/βR+/−) or totally (IFN-α/βR−/−) abrogated type-I IFN pathway develop a mild or severe infection, respectively. In mice with a mild infection, after a burst of viral replication in the liver, CHIKV primarily targets muscle, joint, and skin fibroblasts, a cell and tissue tropism similar to that observed in biopsy samples of CHIKV-infected humans. In case of severe infections, CHIKV also disseminates to other tissues including the CNS, where it specifically targets the choroid plexuses and the leptomeninges. Together, these data indicate that CHIKV-associated symptoms match viral tissue and cell tropisms, and demonstrate that the fibroblast is a predominant target cell of CHIKV. These data also identify the neonatal phase and inefficient type-I IFN signaling as risk factors for severe CHIKV-associated disease. The development of a permissive small animal model will expedite the testing of future vaccines and therapeutic candidates.
| Chikungunya virus (CHIKV) is transmitted by mosquito bites. CHIKV has recently re-emerged and is responsible for a massive outbreak in the Indian Ocean region and India. It has also reached Italy, indicating that CHIKV has a great potential to spread globally. Infection from CHIKV typically induces a mild disease in humans, characterized by a flu-like syndrome associated with muscle and joint pain and rash. Cases of severe infection involving the central nervous system (CNS) have recently been described, notably in neonates. We have developed the first animal model for CHIKV infection and studied the pathophysiology of the resulting disease. We show here that mouse neonates are susceptible to CHIKV and neonatal disease severity is age-dependent. Adult mice with a partial or complete defect in type-I interferon pathway develop a mild or severe infection, respectively. In mice with a mild infection, CHIKV primarily targets muscle, joint and skin fibroblasts, a cell and tissue tropism similar to that observed in biopsy samples of CHIKV-infected humans. In case of severe infections, CHIKV also disseminates to the CNS. Our work indicates that CHIKV-associated symptoms perfectly match viral tissue and cell tropisms, and demonstrate that the fibroblast is a prominent target cell of CHIKV. It also identifies the neonatal phase and inefficient type-I interferon signaling as risk factors for severe CHIKV-associated disease. The development of a permissive small animal model will expedite the testing of future vaccines and therapeutic candidates.
| Chikungunya virus (CHIKV) was first isolated in Tanzania in 1953 [1], and has recently emerged in islands of the Indian Ocean in 2005, and engendered the largest Chikungunya fever epidemic on record [2]. The most affected region was the island of La Réunion, where CHIKV infected approximately a third of the island's inhabitants (i.e., ∼300,000) [3–5]. The outbreak, which now also involves India with an estimated 1.3 million cases [6–8], has a significant potential to spread globally given the wide distribution of its arthropod vector [9,10].
CHIKV is a member of the genus Alphavirus in the family of Togaviridae. Alphaviruses are small, enveloped viruses with a message-sense RNA genome that encodes four non-structural proteins (nsP1–4) and three structural proteins (C, E1–2). This arbovirus is maintained in nature by uninterrupted cycles of transmission between mosquitoes and vertebrate hosts such as macaques [11–13]. Several alphaviruses cause disease in humans, primarily as a result of epizootic infections. These include the American encephalitic alphaviruses and several species in the Semliki Forest Virus group, principally the Afro-Asian CHIKV, the African O'Nyong-Nyong virus, as well as the Australasian Barmah Forest virus and Ross River virus [14]. CHIKV infection is characterized by fever, arthralgia, myalgia, rash and headache. During the La Réunion Island outbreak, previously unreported severe forms of Chikungunya infection were observed in adults, complicated by encephalopathy and hemorrhagic fever. These severe cases almost exclusively occurred in adults with underlying conditions such as diabetes, alcoholic hepatopathy or impaired renal function [3,15]. Moreover, while CHIKV-associated fatalities had not been reported prior to this outbreak, at least 213 persons infected with CHIKV died in La Réunion Island [16,17]. Finally, although never described before, per-partum mother-to-child CHIKV transmission has been observed and is associated with severe neonatal disease characterized in more than half of the cases by encephalopathy [18,19].
The pathophysiology of human CHIKV infection has so far remained essentially unknown, in part because of the lack of a permissive small animal model. In order to gain a better understanding of CHIKV-associated pathophysiology, we have developed a mouse model of infection. Using this model system and comparing it to human samples, we have uncovered the following pathophysiological features of CHIKV infection: (i) viral dissemination and disease severity are strongly increased during the neonatal period, (ii) type-I IFN signalling plays a critical role in the control of the infection and is associated with severe infection when deficient, (iii) symptomatic organs are those infected by CHIKV, and (iv) fibroblasts of connective tissues are prominent cell targets in vivo, in permissive mice as well as in humans. Together, this study offers the first in depth in vivo analysis of CHIKV cellular tropism and offers a validated small animal model that may prove useful for the development and testing of novel vaccine and therapeutic strategies.
In order to study the pathophysiology of CHIKV infection in the adult and neonatal hosts, we aimed to develop a small animal model of infection. We first inoculated a series of classical laboratory mouse strains: outbred OF1 mice and inbred C57BL/6 and 129s/v mice. Intra-dermal (ID) injection of 106 PFU of CHIKV-21, isolated from an individual from La Reunion with central nervous system (CNS) symptoms [4], showed that WT adult OF1 C57BL/6 and 129s/v mice are resistant to CHIKV infection. Neither morbidity nor mortality was observed following infection and no infectious virus could be recovered from tissues (unpublished data). In contrast, neonatal C57BL/6 mice exhibited an age-dependent lethality to CHIKV infection (Figure 1A): six-day-old and 9-day-old mice all developed flaccid paralysis on D6 or D7 post infection (pi), and all 6-day-old mice died before D12 pi, whereas more than half of 9-day-old infected mice recovered. Strikingly, by day 12 of life, C57BL/6 mice showed neither morbidity nor mortality following infection. We investigated the kinetic of virus replication in tissues of 9-day-old mice at D3 and D5 pi and at the onset of symptoms (D7 pi). Infectious virus was detected at low level at D3 pi in serum and at D5 and D7 pi in liver (Figure 1B). Strikingly, at all time points analyzed, infectious CHIKV was detected very abundantly in muscle, joint and skin and to a lower extent in the brain (Figure 1B).
Together, these results establish that, as observed in humans, CHIKV pathogenicity is strongly age-dependent in mice, and that in less-than 12 day-old mouse neonates, CHIKV induces a severe disease. Of note, in the infected neonatal mice, CHIKV is abundantly detected in the same organs than those symptomatic in humans, particularly infected neonates and babies under the age of one year [18].
That neonatal mice but not adult mice were permissive to CHIKV infection ruled out the existence of an intractable species barrier in the mouse. It also questioned the nature of mouse adult non-permissiveness. Guided by the well-established ability of CHIKV and other alphaviruses to trigger type-I IFN synthesis and their sensitivity to type-I IFN responses [20–24], we tested the permissiveness of adult IFN-α/βR knockout (IFN-α/βR−/−) mice towards CHIKV.
In contrast to WT adult mice, all infected IFN-α/βR−/− adult mice developed a severe disease characterized by muscle weakness of the limbs (i.e., loss of muscle tone) and lethargy and died at D3 pi (Figure 2A). Whereas no mortality was observed in WT adult mice inoculated with 106 PFU of CHIKV-21, the lethal dose 50 (LD50) was of 3 PFU in adult IFN-α/βR−/− adult mice, with an average survival of 3 ± 0.2 days. Similar results were obtained when infecting IFN-α/βR−/− adult mice with two other isolates from La Réunion (CHIKV-27 and −115) and one African isolate from Congo (CHIKV-117) (unpublished data). Together, these results indicated that the basis for the resistance of WT adult mice to CHIKV is linked to type-I IFN signalling, which thus stands as a key player in the control of CHIKV infection.
We next investigated CHIKV replication in tissues of WT and IFN-α/βR−/− adult mice inoculated via the ID route. As IFNAR-α/βR−/− adult mice are highly susceptible to CHIKV infection, we inoculated them with about 10 LD50, i.e., 20 PFU, whereas we inoculated WT mice with 106 PFU, and determined the viral load in mouse tissues at different time points pi. CHIKV was not isolated from tissues of WT mice at H3, H16, H32, D3 pi, or D6 pi (Figure 2B and unpublished data). In IFN-α/βR−/− mice at H16 pi, infectious virus was detected only in the liver. At D3 pi, it was abundantly detected in muscles, joints, skin, and brain with high viral titers also present in serum, liver and spleen (Figure 2B). Similar results were obtained with the Congolese isolate CHIKV-117 (unpublished data).
In search for a model of non-lethal CHIKV infection that would reflect the mild disease predominantly observed in human adults, we evaluated disease pathogenesis in IFN-α/βR+/− adult mice. In IFN-α/βR+/− adult mice –which express IFN-α/βR to a similar level found in WT animals, as assessed by FACS analysis (unpublished data)–, no mortality nor lethargy was observed upon infection with 106 PFU of CHIK-21 (Figure 2A). However, in contrast to what was observed in WT adult mice, infectious virus was recovered from liver as early as H16, at titers similar to that of IFN-α/βR−/− infected mice (Figure 2B). Strikingly, at D3 pi, infectious virus was confined to muscles and joints (Figure 2B) after which the virus was cleared and reached undetectable levels by D6 pi (unpublished data). These data suggest that a dose effect of type-I IFN receptor gene controls the level of permissiveness of adult mice towards CHIKV, with IFN-α/βR+/− and IFN-α/βR−/− adult mice developing a mild and severe infection, respectively. As observed in human clinical cases, the mild form of the disease corresponds to a peripheral infection targeting predominantly skeletal muscles and joints, whereas the severe form is also associated with viral dissemination to other organs, including the CNS.
We next investigated the cell tropism of CHIKV in liver at H16 pi and in peripheral infected tissues, namely, liver, skeletal muscles, joints and skin at D3 pi. In liver of infected adult IFN-α/βR−/− mice at H16 pi, a weak labelling of CHIKV antigens was detected in sinusoidal capillary endothelial cells (Figure S1A–S1C), as well as in some F4/80 labelled mature macrophages (Figure S1D–S1F), whereas at D3 pi, consistent with a sharp increase in liver and serum viral loads, CHIKV immunolabeling was more intense and diffuse, and co-localized within and around sinusoid capillaries (Figure S1G–S1I). Confirmatory findings were obtained by transmission electron microscopy analysis of sections from the same infected livers. These images reveal numerous CHIKV particles budding at the surface of sinusoid capillary endothelial cells (Figure S2A). A lower number of viral particles were also found associated with (arrow) or budding at the surface (arrowhead) of Kupffer cells (Figure S2B). In order to further investigate viral replication in this cell type, we isolated primary liver macrophages and infected them in vitro. Although only low levels of viral replication were observed, we do conclude that liver macrophages are capable of being infected by CHIKV (Figure S1J). In contrast, brain-derived microglial cells were refractory to infection suggesting that macrophages are not a general target of infection (Figure S1J). In spleen, the immunolabelling for viral antigen was exclusively observed in the red pulp, notably in F4/80-positive cells (unpublished data).
In skeletal muscles of infected adult IFN-α/βR−/− mice, and to a lower intensity of IFN-α/βR+/− mice, CHIKV immunolabelling was predominantly observed in connective tissue, particularly in the external region, the epimysium (also called muscle fascia), and to a lower extent in the perimysium and endomysium (Figure 3A–3C). Consistent with these observations, viral load was higher in the epimysium than in the perimysium/endomysium (unpublished data). The main target cells in muscles were fibroblasts, as shown by the morphology of the cells and their co-immunolabelling with anti-vimentin and anti-CHIKV antibodies (Figure S3), and the absence of basal lamina surrounding the immunolabeled cells (Figure 3A–3C). Few immunolabeled F4/80-positive macrophages were also observed in the epimysium and to a lesser extent in the perimysium where they predominated around middle size arteries and veins in a close contact with fibroblasts (unpublished data). Consistent with a recent study performed on human material [25], rare satellite cells were also immunolabeled (arrowhead in Figure 3B), readily recognizable as small mononucleated cells located beneath the muscle fiber basal lamina. In skeletal muscles, myofibers were not immunolabelled with anti-CHIKV antibodies.
In IFN-α/βR−/− adult mice at D3 pi, fibroblasts of the joint connective tissue located beneath the synovial wall were also infected (Figure 3F–3H), but both deep articular and osseous tissues (i.e., chondrocytes and osteocytes and osteoblasts) were uninfected. In the skin, viral antigens were also observed in fibroblasts of the deep dermis (Figure 3D, 3E). In IFN-α/βR+/− adult mice at D3 pi, viral antigens were also detected in fibroblasts of the connective tissue of joints (Figure 3I) and skin (unpublished data) but to a lower level than in IFN-α/βR−/− mice.
Viral cell tropism in infected peripheral tissues of neonates, including muscle, joint and skin was similar to that of adult mice, with a pronounced tropism for fibroblasts (Figure 4A–4C). A notable difference was the presence of severe necrotic myositis consistent with severe myofiber necrosis and inflammation manifested by the infiltration of lymphocytes and monocytes/macrophages (Figure 4D, 4E). Importantly, our in vitro experiments with primary mouse and human muscle fibroblasts confirm the high permissiveness of this cell type towards CHIKV (unpublished data).
To investigate whether blood leukocytes were a target for CHIKV in vivo, peripheral blood mononucleated cells from infected IFN-α/βR−/− mice obtained in the course of the infection (D1 to D3 pi) were double-stained with an anti-CHIKV and a pan-murine haematopoietic cell marker (CD45.2 antibody) and analyzed by flow cytometry. No infected leukocyte was detectable in the blood of infected mice, indicating that blood leukocytes do not represent a significant cell target for CHIKV in vivo (unpublished data), as also reported in the in vitro context [26].
Together, these data show that in infected peripheral tissues of adult IFN-α/βR+/− and IFN-α/βR−/− mice, as well as in WT neonates, fibroblasts constitute a prominent target cell of CHIKV.
We next investigated the histopathology and CHIKV infection of the CNS. The only histopathological finding in IFN-α/βR−/− mice at the CNS level was a severe vacuolization of choroid plexus epithelial cells and often of the adjacent ependymocytes (Figure 5A). Choroid plexuses, ependymal wall, and lepto-meningeal cells, including external cells in the Virchow-Robin spaces, were strongly stained for CHIKV, whereas the brain parenchyma did not show significant labelling (Figure 6). We could observe no CHIKV immunolabelling in microglial cells and astrocytes, including those forming the glia limitans (unpublished data). The choroid plexuses, which form the blood-cerebrospinal fluid (CSF) barrier, were infected (Figure 6D). In contrast, microvascular endothelial cells that constitute the blood-brain barrier (BBB) were not (Figure 6A, arrowheads). Viral titer in the meninges of infected IFN-α/βR−/− was 5-fold higher than in the total brain (Figure 5B). In the CNS of infected WT mouse neonates, CHIKV infection was also detected at the leptomeningeal level (Figure 4F), but here again, no infection was detected in the brain parenchyma.
To determine whether CHIKV infection altered the permeability of the BBB, we administrated intravenously horseradish peroxydase (HRP), which does not diffuse to the brain parenchyma when the BBB is intact, but leaks into the brain parenchyma in case of BBB disruption [27]. HRP in brains of infected IFN-α/βR−/− adult mice, was confined to the lumen of brain microvessels, as observed in the brains of mock-infected mice (Figure 5C). Thus, despite a strong infection of the meninges and of the Virchow-Robin spaces, the barrier function of the brain microvessels was preserved upon infection.
These in vivo findings were confirmed in in vitro BBB systems [28,29]. Primary choroid plexus epithelial cells were highly susceptible to infection via the apical route (Figure 7A) and to a lesser extent via the basal route (Figure 7B), suggesting that CHIKV accesses the cerebrospinal fluid through the choroid plexuses, and may also secondarily infect choroid plexus epithelial cells via their apical surface, thus amplifying viral titers in the CSF. In sharp contrast, primary brain microvessel endothelial cells were fully resistant to CHIKV infection (Figure 7C). Together, these findings suggest that CHIKV gets access to the CNS via the choroid plexuses, and exhibit a marked tropism for the meninges, whereas it does not infect the brain microvessels and parenchyma and does not induce tissue alteration at the brain parenchyma level.
Given the observations made during the La Réunion outbreak that CHIKV can be vertically transmitted from viremic mothers to their newborns, we investigated maternal-fetal transmission of CHIKV in pregnant IFN-α/βR−/− mice infected with CHIKV-21 via the ID route at D16–18 of gestation. At D2 pi, animals were sacrificed and viral titers determined in maternal serum as well as in placentas and fetuses (Figure 8A). As expected, viral load in maternal serum was elevated. In contrast, placenta viral titers were at least 2 orders of magnitude lower and fetuses were uninfected (Figure 8A). Moreover, no CHIKV immunolabeling could be observed in these placentas (unpublished data). The non-permissiveness of the placental barrier towards CHIKV was confirmed in vitro, by the observation that the human syncytiototrophoblastic cell line BeWo is refractory to infection (Figure 8B).
In order to test the relevance for human of our in vivo and in vitro studies, we investigated CHIKV cell and tissue tropisms in biopsy samples of CHIKV-infected humans with acute CHIKV infection. We developed a sensitive and specific immunohistochemistry assay (see Materials and Methods) to detect CHIKV antigens in available human tissue samples from a fatal neonatal case. In the tissues that are the classical sites of symptoms in the human disease, namely the skeletal muscles, joints and skin, CHIKV antigens were detected, and viral infection appeared to be confined to fibroblasts of the joint capsule, of skeletal muscle fascia and of the dermis (Figure 9A–9C). As brain human samples were not available, we could not investigate CHIKV dissemination to the CNS. However, studies in experimentally infected Cynomolgus monkey, who develop a severe CHIKV infection, indicate that CHIKV disseminates to the CNS, where it targets the choroid plexus and the leptomeninges, but not the brain microvessels and parenchyma (Roques et al. personal communication).
Together, these results demonstrate that in humans, CHIKV is present in symptomatic organs, and that the fibroblast is the privileged cell target in these organs. Moreover, and in agreement with the frequent positivity of CHIKV RT-PCR in the CSF of humans with CNS symptoms [30], CHIKV is able to reach the CNS via the choroid plexuses and preferentially target the leptomeninges in Cynomolgous monkey.
Here we have combined in vivo, in vitro, and histopathology approaches to gain a better insight in Chikungunya disease pathophysiology. Using the mouse as a model, we show that in the neonatal host, as well as in adult mice harboring one or two copies of IFN-α/βR null allele, CHIKV exhibits a marked tropism for skeletal muscles, joints and skin, which constitute the classical symptomatic organs in the human disease. This shows that, in contrast to other acute viral infections in which symptoms may predominantly reflect the systemic immune response rather than viral organ dissemination (e.g., influenza), classical symptoms of Chikungunya disease closely reflect CHIKV tissue tropism. Indeed, our study provides direct evidence that in the mouse adult and neonate models, as well as in humans, muscles, joints and skin are privileged CHIKV targets.
We demonstrate here that CHIKV infection severity is critically dependent on two host factors: age and functionality of type-I IFN signaling, thus underlining similarities between CHIKV and the prototypic alphavirus Sindbis [23,31]. In the neonatal host as well as in the adult mouse with a totally abrogated type-I IFN signaling, CHIKV-associated disease is particularly severe, and this severity correlates with higher viral loads and dissemination to the CNS. Importantly, similar findings have been reported in human neonates and adults with severe disease [32]. The reasons why the neonatal status and a defect in type-I IFN signaling favor severe CHIKV infection may partly overlap, but specific neonatal factors may also be involved. Indeed, a number of physiological variables differentiate the neonatal and adult hosts, including the relative proportion of tissue fibroblasts, the rate of cell division, and the maturity and effectiveness of the innate immune system [33]. Future work will have to focus on the similarities and differences between the neonatal and adult hosts with respect to type-I IFN triggering, signaling and responses. Nevertheless, the basic characteristics of CHIKV cell and tissue tropisms are conserved in these two complementary models, and their similarity with what observed in humans strongly argues in favor of their pathophysiological relevance.
With the IFN-α/βR+/− mice, we also provide a model for the benign CHIKV human infection. This animal model should prove very helpful in the development of future vaccine and therapeutic strategies. Importantly, that the gene copy number of IFN-α/βR strictly influences the viral load and tissue distribution as well as the severity of the disease is a strong indication that the strength of type-I IFN signaling likely plays a critical role in the control of CHIKV replication.
The significance of CHIKV specific tissue tropism is emphasized by the observation that tissue fibroblasts constitute the principal CHIKV cell target in all these infected peripheral organs. This in vivo finding is consistent with the in vitro observation that primary mouse muscle fibroblasts are susceptible to CHIKV infection (unpublished data) as well the recent finding by Sourrisseau and colleagues that cultured human lung and mouse skin fibroblasts are permissive to CHIKV [26]. The molecular basis for this prominent in vivo tropism for fibroblasts is unknown and may indicate that fibroblasts could be, relative to other cell types, either (i) in a hyper-permissive status towards CHIKV entry/replication, and/or (ii) in a hypo-sensitive status to type-I IFN-mediated viral interference, making them a target of choice for CHIKV. Interestingly, it is proposed that fibroblasts of connective tissue of dermis, joint capsules [34–36], and muscles have in common the property to form a reticular network of cells interconnected by gap junctions. Whether this characteristic contributes to the selective in vivo hyper-susceptibility of connective fibroblasts to CHIKV infection, and if it plays a role in viral cell-to-cell dissemination deserves future investigations.
Of note, similarly to CHIKV in mouse neonates, Ross River virus, an alphavirus closely related to CHIKV and associated with muscle and joint pathology, has been shown to induce myositis in adult mice [23,37,38]. The absence of myositis in CHIKV-infected IFN-α/βR−/− adult mice could be linked to their early lethality, while it could be linked to the rapid recovery of CHIKV-infected IFN-α/βR+/− adult mice.
Before reaching its target organs, CHIKV undergoes an early burst of viral replication in the liver in IFN-α/βR+/− and IFN-α/βR−/− mice. Indeed, in these mice, the liver is the first and only detectably infected tissue until H32 pi, and CHIKV antigens are primarily detected in sinusoidal capillary endothelial cells and to a lesser extent in Kupffer cells. By D3 pi, and only in IFN-α/βR−/− mice, there is a sharp increase in viremia, with CHIKV antigen detectable in the red pulp of the spleen. This is in contrast with what has been observed with Sindbis virus infection, in which a high level of infection is detected in the spleen of IFN-α/βR−/− mice as early as D1 pi with a subsequent dissemination to the liver at D2 pi [23,37,38]. In the spleen as in the liver, macrophage-dendritic like cells are thought to be the main target cells of Sindbis virus, although by D3 pi in the liver, “sinusoid-lining cells” considered as Kupffer cells and/or endothelial cells are also infected [23]. These in vivo differences between Sindbis (and other alphaviruses such as SFV, EEV or Ross River virus) and CHIKV could reflect the non-permissiveness of dendritic cells to CHIKV (unpublished data and [26]). We also found that, in contrast to mouse liver macrophages, CHIKV is not detected in mouse blood leukocytes in vivo. These data are consistent with previous experiments that showed that human primary monocytes are not permissive to CHIKV, whereas human primary macrophages are [26].
Earlier studies on Sindbis and Ross River also identified the connective tissues of joints and skeletal muscles as sites of viral replication, although the cell type targeted by these viruses in these tissues has not been formally identified [23,37,38]. We show here that this connective tissue tropism also extends to CHIKV. It is thus likely that these shared tropisms and symptoms highlight a common viral pathogenic property, the understanding of which should provide critical clues to the pathophysiologic properties of alphaviruses responsible for arthralgia. Interestingly, both joint and muscle connective tissues contain a high amount of nociceptive nerve-endings [39] that may account for the muscle and joint pain characterizing disease caused by alphaviruses associated with muscle and joint pathology.
In the case of severe CHIKV infection, we found that CHIKV disseminates to the CNS, as also observed in human [30] and non-human primates (Roques et al. personal communication). It is noteworthy that all CHIKV isolates tested exhibited a similar ability to reach the CNS. CHIKV dissemination to the CNS does not correspond to non-specific spreading due to an overwhelming viral multiplication. Indeed, CHIKV is not detected at the brain microvessel and parenchyma levels, but gets access the CNS exclusively via the choroid plexus route, and undergoes a step of viral amplification at the ependyma and leptomeningeal levels. In agreement with these findings, CHIKV is detected in the CSF in humans with severe human Chikungunya disease associated with CNS symptoms [30]. Thus, in contrast to what has been observed for American encephalitic alphaviruses, CHIKV does not appear to be intrinsically encephalitogenic and is associated with reversible CNS symptoms in humans, in line with a virus that does not invade the brain parenchyma nor infect neurons. That CHIKV does not target the brain endothelium and is not detectable at the brain parenchyma level also contrasts with what has been observed with the more closely related Semliki Forest virus, which targets the brain microvessels and also infects neurons [40,41]. Leptomeningeal tissues are, like fibroblasts, of mesenchymatous origin, and exhibits common features with peripheral fibroblastic connective tissue capsules, as they also play a common “envelop” function and form an interconnected multicellular network that acts as a regulatory interface between cerebrospinal fluid and the surface of the brain and between arterioles within the brain and the surrounding neural tissue (Virchow-Robin spaces) [42]. These common tissue organizations and functions may play a role in CHIKV dissemination.
Given its public health and pathophysiological importance, we also investigated CHIKV infection in the pregnant host. By use of the most permissive model we have developed, the IFN-α/βR−/− mice, we show that the placenta does not constitute a privileged target for CHIKV. Indeed, no infected cells can be detected when observing placental tissue sections from infected mice. This is in line with our investigations carried out in human placentas obtained from viremic mothers, in which no infected cell could be detected by mean of immunohistochemistry either [18]. This suggests that viral titers detected in mouse and human placentas rather correspond to a contamination from remaining maternal serum than to an actual placental infection. This is supported by our in vitro finding that human syncytiotrophoblast is refractory to infection, and by the observation that ante-partum fetal contamination is exceptional in humans [18]. This provides an explanation for why all cases of vertical transmission of CHIKV in the recent outbreak in La Réunion were observed in per-partum, at a time when highly viremic maternal blood can get in contact with the fetal circulation, particularly in the setting of the uterine contractions of the labor, which are known to induce placental barrier breaches. These observations appear to contrast with what observed with Ross River virus, for which actual placental infection and transplacental dissemination have been described both in mice and humans [43,44].
In conclusion, we have developed a mouse animal model for CHIKV infection, allowing us to uncover the viral tissue and cell tropism of this re-emerging alphavirus. We have shown that, in this model, as well as in humans, the fibroblast is the cell type chiefly targeted by CHIKV, and that this accounts for its tropism for muscles, joint and skin connective tissues. The molecular basis for this tropism is currently unknown but may combine specific virus-host cell and tissue interactions as well as an intrinsic relatively lower ability of this cell type to control CHIKV infection. We have also identified two critical factors influencing viral replication, which are the neonatal status, and a defective type-I IFN signaling. Whereas it is clear that an increased neonatal susceptibility is also observed in humans, the relevance of type-I IFN defect as a basis for severe infection in humans remains to be demonstrated. However, the fact that severe infections in humans are exclusively observed in individuals with underlying conditions renders this hypothesis attractive [3,15]. This may indicate that type-I IFN could be of interest to prevent severe disease in adults, as well as in exposed neonates. In addition, the use of neutralizing antibodies also appears interesting, given the strong correlation between viral load and disease severity. A better understanding of the pathophysiology of CHIKV infection and the ensuing development of therapeutic strategies are both critical in the context of a possible globalization of the current CHIKV epidemic.
CHIKV isolates were obtained from individuals during the 2005–06 CHIKV outbreak in La Réunion Island and amplified on mosquito C6/36 cell as described [4]. CHIKV-21 was isolated from the serum of a newborn male with CHIKV-associated encephalopathy; CHIKV-27 was isolated from the CSF of another new-born male with encephalopathy; CHIKV-115 from the serum of a 24-year old female with classical CHIK symptoms. CHIKV-117 was isolated at the Institut de Médecine Tropicale du Service de Santé des Armées (IMTSSA), Marseille, France during the 2000 CHIKV outbreak in Democratic Republic of the Congo from the serum of a person presenting classical CHIK symptoms. Titers of virus stocks were determined by standard Vero cell plaque assay and are expressed as PFU per ml.
Primary choroid plexuses and brain microvascular endothelial cells were obtained, purified and cultured as described [28,29]. The Bewo cell line was obtained from the ATCC. Cells were infected with CHIKV at a multiplicity of infection (MOI) of 10.
Human tissue samples were obtained from biopsy specimens collected in the course of the clinical care of people with CHIKV infection in La Réunion.
Outbred OF1 mice, and inbred C57BL/6 and 129s/v mice were obtained from Charles River laboratories (France). IFN-α/βR−/− 129s/v mice were given by F. Tangy with permission from M. Aguet [45]. Mice were bred according to the Institut Pasteur guidelines for animal husbandries and were kept in level-3 isolators. Mice were inoculated by ID in the ventral thorax with 50 μl of a viral suspension diluted with PBS for adult mice and with 30 μl for neonates. Mock-infected mice received PBS alone. Mice were anesthetized with isoflurane (Forene, Abbott Laboratories Ltd, United-Kingdom). Blood was collected by cardiac puncture after which each mouse was perfused via the intracardiac route with 40 ml of PBS at 4 °C before harvesting of organs. Tissues were homogenized, and virus titers of each tissue sample determined on Vero cells by tissue cytopathic infectious dose 50 (TCID50), and viral titers in tissues and in serum were expressed as TCID50/g or TCID50/ml, respectively. The principles of good laboratory animal care were followed all through the experimental process. Pregnant female IFN-α/βR−/− mice at 16–18 days of gestation were infected with 20 PFU of CHIKV via the ID route. At D2 pi, maternal serum, placentas and fetuses were harvested, and viral titers determined in each sample. Mortality studies were performed on groups of six mice and viral titers in tissues from four mice at each time point.
Mouse organs were snap frozen in isopentane cooled by liquid nitrogen for cryosectionning or fixed in para-formaldehyde for paraffin embedding. Paraffin-embedded tissues were processed for histological staining (Hematoxylin and eosin). For immunofluorescence, cryosections were fixed for 10 min in ice-cold methanol before incubation for 12 h at 4 °C with the primary antibodies followed by incubation for 1 h with the secondary antibodies. Slides were counterstaining with Hoechst (Vector Lab). The following antibodies were used: polyclonal rabbit anti-collagen IV (Chemicon, Temecula CA, 1:200), polyclonal chicken anti-vimentin (Abcam, Cambridge, UK, 1:200), monoclonal mouse anti-GFAP (BD pharmingen 1:1,000 or 1:5,000 to only see the glia limitans), monoclonal rat anti-macrophage antigen F4/80 (Abcam, 1:100), polyclonal rabbit anti-PECAM1/CD31 (Abcam, 1:400), human serum anti-CHIKV were obtained and characterized by the Centre National de Référence des Arbovirus as positive for anti-CHIKV IgM and IgG. The marker specificities were systematically confirmed by examining sections in which the primary antibody was replaced by control isotype or immunoglobulins at the same concentration and by immunostaining of non-infected tissues from the same animal strain. Slides were examined with a Zeiss AxioPlan 2 microscope equipped with an ApoTome system in order to obtain 0.7 μm thick optical sections. Pictures and Z-stacks were obtained using the AxioVision 4.5 software. When necessary, images were processed using the image J software (http://rsb.info.nih.gov/ij/).
Ultra thins sections were observed at 80 kV accelerating voltage using a JEOL JEM 1010 Electron Microscope.
Chromogenic immunohistochemistry was performed on human tissues. After antigen unmasking (97 °C, 20 min), a monoclonal antibody against CHIKV (provided by bioMérieux and the IMTSSA, France) was incubated overnight and revealed with a polymer detection kit (SuperPicTure™ Zymed, http://www.invitrogen.com/) and the DAB chromogen (Vector Lab, http://www.vectorlabs.com/).
BBB permeability was studied using I.V. injection of type VI HRP (Sigma) (30 mg/ml dissolved in Evans blue 2% in 0.9% saline to provide 100 mg/kg) as previously described [27,46]. Five brains were studied for CHIKV-infected mice as well as for mock-infected mice.
Mouse splenocytes were isolated, labeled with biotinylated goat anti-mouse IFN-α/βR1 antibody (BFA3039, R&D systems), incubated with streptavidin-APC and analyzed by flow cytometry. |
10.1371/journal.pgen.1004243 | The Membrane-Associated Transcription Factor NAC089 Controls ER-Stress-Induced Programmed Cell Death in Plants | The unfolded protein response (UPR) is activated to sustain cell survival by reducing misfolded protein accumulation in the endoplasmic reticulum (ER). The UPR also promotes programmed cell death (PCD) when the ER stress is severe; however, the underlying molecular mechanisms are less understood, especially in plants. Previously, two membrane-associated transcriptions factors (MTFs), bZIP28 and bZIP60, were identified as the key regulators for cell survival in the plant ER stress response. Here, we report the identification of another MTF, NAC089, as an important PCD regulator in Arabidopsis (Arabidopsis thaliana) plants. NAC089 relocates from the ER membrane to the nucleus under ER stress conditions. Inducible expression of a truncated form of NAC089, in which the transmembrane domain is deleted, induces PCD with increased caspase 3/7-like activity and DNA fragmentation. Knock-down NAC089 in Arabidopsis confers ER stress tolerance and impairs ER-stress-induced caspase-like activity. Transcriptional regulation analysis and ChIP-qPCR reveal that NAC089 plays important role in regulating downstream genes involved in PCD, such as NAC094, MC5 and BAG6. Furthermore, NAC089 is up-regulated by ER stress, which is directly controlled by bZIP28 and bZIP60. These results show that nuclear relocation of NAC089 promotes ER-stress-induced PCD, and both pro-survival and pro-death signals are elicited by bZIP28 and bZIP60 during plant ER stress response.
| Protein folding is fundamentally important for development and responses to environmental stresses in eukaryotes. When excess misfolded proteins are accumulated in the endoplasmic reticulum (ER), the unfolded protein response (UPR) is triggered to promote cell survival through optimizing protein folding, and also promote programmed cell death (PCD) when the stress is severe. However, the link from ER-stress-sensing to PCD is largely unknown. Here, we report the identification of one membrane-associated transcription factor NAC089 as an important regulator of ER stress-induced PCD in plants. We have established a previously unrecognized molecular connection between ER stress sensors and PCD regulators. We have shown that organelle-to-organelle translocation of a transcription factor is important for its function in transcriptional regulation. Our results have provided novel insights into the molecular mechanisms of PCD in plants, especially under ER stress conditions.
| In eukaryotic cells, ER is a major site for the production of secreted, plasma membrane and organelle proteins. Cells have evolved a sophisticated quality control system to ensure the accuracy of protein folding through optimizing the protein-folding machinery and ER-associated degradation (ERAD) [1], [2], [3]. To coordinate protein-folding capacity with protein-folding demand, a collection of phylogenetically conserved signaling pathways, termed the UPR, senses the accumulation of misfolded proteins in the ER and sustains homeostatic balance according to the protein folding needs which change constantly depending on different developmental programs and/or environmental conditions [1], [4], [5].
Three arms of UPR signaling pathways, namely inositol requiring enzyme 1 (IRE1), double-stranded RNA-activated protein kinase (PKR) like ER kinase (PERK), and activating transcription factor 6 (ATF6), were identified in mammalian cells that have the ability to promote cell survival by reducing misfolded protein accumulation in the ER. IRE1 is a key component in the most conserved branch, which acts by splicing messenger RNA encoding transcription factor Hac1p in yeast or XBP1 in mammalian cell, respectively [6], [7], [8]. Recently, the equivalent pathways were discovered in plants (e.g. the IRE1-bZIP60 pathway in Arabidopsis), which also play important roles in heat stress response, as well as in plant immune response [9], [10], [11], [12], [13], [14]. PERK is an ER-localized kinase and its activation upon ER stress leads to the attenuation of bulk protein translation in metazoan cells [15]. ATF6 is an ER membrane-associated bZIP transcription factor; its activation requires ER-to-Golgi translocation and regulated intramembrane proteolysis (RIP) [16]. Although the plant PERK ortholog has not yet been reported, the ER membrane-associated Arabidopsis bZIP28 was found to be the functional homolog of mammalian ATF6, which is activated in a manner similar to ATF6 [17], [18], [19], [20], [21].
Severe or chronic ER stress can also lead to PCD, a process that kills unwanted cells under ER stress conditions to protect other cells [22]. In contrast to what is known about how UPR protects cells, less is known about the mechanisms that link UPR to PCD, especially in plants [23]. In mammalian cells, IRE1 can trigger PCD by activating the Jun amino-terminal kinase (JNK) pathway [24]. Phosphorylation of JNK leads to the activation of pro-death protein BIM and inhibition of anti-death protein BCL-2 [25]. Mammalian IRE1 also binds to BAX and BAK, two cell-death-inducing proteins involved in the mitochondrial cell death pathway [26]. The activation of mammalian IRE1 is able to cause rapid decay of selected microRNAs (miRs -17, -34a, -96, and -125b) that normally repress translation of caspase-2 mRNA, and thus sharply elevates protein levels of this initiator protease in the mitochondrial cell death pathway [27]. The ER stress-induced mammalian bZIP transcription factor CHOP is one of the major players that induces PCD, most probably through suppression of the pro-survival protein BCL-2 and up-regulation of ERO1α to further perturb the cellular redox state [28]. CHOP is a downstream target of all three aforementioned UPR signaling pathways in mammals [4]. Recently, transcriptional induction through ATF4 and CHOP was shown to increase protein synthesis leading to oxidative stress and PCD [29]. Orthologs of mammalian JNK, BAX, BAK, CHOP and ATF4 are not found in the Arabidopsis genome [30]. However, ER stress-induced PCD is reported in plants with the hallmark of DNA segmentation, and the conserved BAX inhibitor-1 (BI-1) plays important roles in suppression of ER stress-induced PCD in Arabidopsis plant [31], [32], [33], [34], [35], [36], [37], [38]. When animal cells are subjected to severe ER stress, IRE1 loses its specificity and begins to degrade mRNAs in a process called regulated IRE1-dependent decay (RIDD) [39]. IRE1 in Arabidopsis also has similar function in the RIDD process in the UPR for degradation of mRNA encoding proteins in the secretory pathway to decrease the amount of proteins entering the ER [40]. Different from the animal system, knock-outs of both IRE1s in Arabidopsis impairs UPR and enhances PCD upon ER stress, indicating that RIDD may play a negative role in PCD in plants [40]. Despite the emerging evidence on ER stress-mediated PCD in plants, the underlying molecular mechanisms of PCD in plant UPR is still largely unknown. In soybean plants, prolonged ER stress and osmotic stress synergistically activate N-rich proteins (NRPs) to induce the expression of NAC6/NAC30 to regulate PCD together with NAC081 [33], [41], [42], however, the link from ER stress-sensing machinery to these NRPs is still missing.
Here we show that NAC089 plays important roles in regulating ER-stress-induced PCD in the model plant Arabidopsis. NAC089 relocates from the ER membrane to the nucleus during ER stress response. Inducible expression of a truncated form of NAC089 promotes PCD with increased caspase 3/7-like activity and DNA fragmentation. Down-regulation of NAC089 confers ER stress tolerance and impairs ER-stress-induced caspase 3/7-like activity. Several UPR downstream genes including the PCD regulators MC5, BAG6 and NAC094 are shown to be regulated by NAC089 under ER stress condition. NAC089 itself is also up-regulated by ER stress, which is directly controlled by bZIP28 and bZIP60. Therefore, NAC089 is an important PCD regulator in plant UPR, linking ER stress-sensing to downstream PCD regulators during ER stress response in plants.
An ER-stress-related NAC (for NAM, ATAF, and CUC) transcription factor NAC089 (also known as ANAC089) was identified from our previous microarray analysis [43]. Its expression was up-regulated rapidly by ER stress inducers tunicamycin (TM) and dithiothreitol (DTT) (Figure 1A and Figure S1). Knock-out of either bZIP28 or bZIP60 partially suppressed, while knock-outs of both bZIP28 and bZIP60 in zip28zip60 double mutant completely abolished the up-regulation of NAC089 under ER stress condition (Figure 1B). Previously two ER stress responsive cis-elements UPRE and ERSE-I were identified as the binding sites of bZIP28 and bZIP60 [21], [43], [44]. We searched the NAC089 promoter region and found that one copy of UPRE and one copy of ERSE-I like (one mismatch) cis-elements are present over the segment [−95, −49] relative to the TSS site of NAC089. To assess the activation of NAC089 promoter by bZIP28 and bZIP60, an effector-reporter dual-luciferase transient assay was set up. The NAC089 promoter fragment containing the aforementioned cis-elements was fused to the firefly luciferase reporter and tested in Arabidopsis leaf protoplasts. As expected, the reporter was activated by either TM or DTT treatment (Figure 1C). Using this assay system, co-expression of either bZIP28D or bZIP60S dramatically enhanced the firefly luciferase reporter activity (Figure 1D). To demonstrate the direct binding of bZIP28 or bZIP60 to the NAC089 promoter, electrophoretic mobility shift assays (EMSAs) were performed with the biotin-labeled NAC089 promoter DNA. When the truncated form of either bZIP28 or bZIP60 was incubated with the biotin-labeled DNA probe, a band shift was observed reflecting the formation of the respective complex. To show the binding specificity, excess un-labeled probe was added and shown to be an effective competitor for the formation of each complex. On the contrary, the un-labeled mutated UPRE probe could not compete with the binding (Figure 1E–F). Through further mutation analysis, it was found that neither bZIP28 nor bZIP60 binds to the ERSE-I like cis-element presented in the probe (Figure S2). Thus, the expression of NAC089 is up-regulated by ER stress, which is directly controlled by both bZIP28 and bZIP60 through the UPRE cis-element.
NAC089 is predicted to be a membrane-associated transcription factor [45] with the N-terminal DNA-binding domain facing the cytoplasm (Figure 2A). It has transcriptional activation activity and forms homodimers [46] (Figure S3). To confirm the membrane association of NAC089 and also to investigate the possible membrane-to-nucleus translocation of NAC089 in response to ER stress, 4X MYC tag was fused to NAC089 at the N-terminus and the fusion protein was expressed in Arabidopsis plants. Total proteins were extracted from transgenic seedlings and MYC-NAC089 was detected with the western blotting analysis. Without TM or DTT treatment, one prominent band reacted with the anti-MYC antibody. After TM or DTT treatment for 6 hr, one excess band with smaller molecular weight was induced, with a similar migration rate to the truncated form NAC089D, in which the C-terminal 24 amino acids of NAC089 were replaced with the 4X MYC tag (Figure 2B). To track the movement of NAC089 in response to ER stress, mGFP-NAC089 was expressed in Arabidopsis and observed under confocal laser scanning microscopy. In the mock (H2O) treatment, most of the mGFP-NAC089 signals were observed in the ER (Figure 2C–D); after either TM or DTT treatment for 6 hr, the fluorescence signals were largely found in the nuclei of the Arabidopsis leaf protoplasts and root cells (Figure 2C–D). The nuclear relocation of mGFP-NAC089 was not observed in the root cells when the transgenic plants were treated with TM (5 µg/ml) for short period of time (e.g. 2 hr). The ER-to-nucleus movement was also confirmed in the protein fractionation studies (Figure S4). Taken together, NAC089 is an ER membrane-associated protein and it relocates from the ER membrane to the nucleus in response to ER stress.
To investigate the biological function of NAC089 in the ER stress response, we created partial loss-of-function mutants by RNA interference (RNAi) and chimeric repressor silencing technology (CRES-T). NAC089 knock-down plants (RNAi089) grew as well as the wild-type (wt) control under normal growth condition, but they were more tolerant to ER stress than the wt (Figure 3A–B, Figure S5). More greenish big (G-B) plants and less yellowish small (Y-S) plants were observed in the RNAi089 plants than that in the wt under ER stress conditions (Figure 3B). In CRES-T system, fusion of an EAR-motif repression domain to a transcription factor converts an activator into a repressor, which results in partial loss function of the transcription factor [47]. We replaced the C-terminal hydrophobic tail of NAC089 with the EAR-motif, and expressed the chimerical fusion protein NAC089D-EAR in Arabidopsis with NAC089's native promoter. NAC089D-EAR expression did not affect seedling development under normal growth condition, but also conferred ER stress tolerance in plants (Figure S6A–D). ER stress should be built-up in plants which had been grown on solid growth medium with low concentrations of tunicamycin for a long period of time, as ascertained by the up-regulation of UPR marker genes in the wild-type plants (Figure S6E). All together, we concluded that partial loss-of-function of NAC089 in Arabidopsis increases chronic ER stress tolerance.
To gain insight into mechanisms by which NAC089 operates, we conditionally expressed a MYC-tagged truncated form of NAC089 (NAC089D-MYC) with the beta-estradiol (BE) inducible system [48] (Figure S7). This truncated form co-migrates with the ER stress-induced nuclear form of MYC-NAC089, has the transcriptional activation activity and localizes in the nucleus as mentioned above. The wt control and NAC089D-MYC expressing plants (XVE089D) were transferred to growth medium supplemented with or without BE. There was no obvious difference between the wt and XVE089D plants on the growth medium without BE (Figure 4A). However, when BE was included in the growth medium, root growth of the XVE089D plants was inhibited, and chlorotic leaves were observed, representing a typical PCD phenotype (Figure 4B–C). Cysteine-dependent aspartate-directed proteases (caspases) are the key regulators of PCD in animals, of which caspase-3 is the crucial executioner of PCD and recognize tetra-peptide motif DEVD [49]. Although the ortholog of animal caspase-3 is absent in plants, caspase 3/7-like activity has been reported in many examples involved in plant development and adaptation to environmental stresses [50]. To investigate whether the NAC089D-MYC-induced PCD was associated with the caspase-like activity, we performed caspase-3/7 activity assays using the same tetra-peptide substrate as previously reported [32], [33], [41]. It was found that the expression of NAC089D-MYC considerably induced caspase 3/7-like activity in the XVE089D plants (Figure 4D). The caspase 3/7 activity was also checked in the wt control and NAC089 RNAi plants (line RNAi089-25). ER stress gradually induced caspase 3/7-like activity in both the wt and RNAi089-25 plants in response to chronic ER stress. However, the caspase 3/7-like activity in the RNAi089-25 plants was about half of that in the wt plants after 3 days of TM treatment (Figure 4E), suggesting that NAC089-regulated caspase 3/7-like activity is stress severity-dependent. It is possible that other pathways also regulate such caspase 3/7-like activity. Loss of cell viability, accumulation of H2O2 and rupture of plasma membrane are often associated with PCD [30]. To assess cell viability, roots of XVE089D plants were stained with fluorescein diacetate (FDA), which is a substrate for many endogenous esterases. NAC089D-MYC expression dramatically reduced the endogenous esterase activities (Figure 4F–G). Further 3, 3′-diaminobenzidine (DAB) staining demonstrated that H2O2 was accumulated in the roots when NAC089D-MYC was induced (Figure 4H–I). Propidium iodide (PI) binds to DNA, but it is often used to stain plant cell wall or plasma membrane (Figure 4J) because it is membrane impermeant. When NAC089D-MYC was induced, PI signals were observed in both the cytoplasm and nucleus (Figure 4K), indicating that expression of NAC089D-MYC reduced the rigidity of cell membrane. Another characteristic of PCD is the morphological changes in the nucleus which could be revealed by 4, 6′-diamidino-2-phenylindole (DAPI) staining. Intact and round nuclei were found in most of the XVE089D root cells without BE treatment (Figure 4L). In contrast, nuclei with stretches and speckles were observed in the XVE089D root cells when the plants were treated with BE (Figure 4M–N). Cleavage of genomic DNA at internucleosomal sites by endogenous nucleases is always associated with PCD and terminal deoxynucleotidyl transferase-mediated dUTP nick and labeling (TUNEL) assay is frequently used to label the fragmentation of nuclear DNA in situ [31]. Compared to the low level of background green fluorescence in normal-grown roots (Figure 4O), strong TUNEL-positive signals were observed in the XVE089D root cells when the plants were treated with BE (Figure 4P–Q). BE treatment had no obvious effect on the aforementioned histochemical staining in the wt control plants (Figure S8). These results suggest that NAC089 has the ability to promote PCD in plants.
PCD is a genetically controlled process that plays important roles in plant development and responses to abiotic stress or pathogens [51], [52]. Many PCD regulators that have been well characterized in humans, worms and flies are absent from the Arabidopsis genome, indicating that plants may use different regulators to execute PCD [52], [53]. To understand how NAC089 regulates plant PCD, we performed microarray (Agilent 4X44K) experiments with the BE inducible gene expression system [48]. BE treatment did not affect much of the gene expression in the wt plants as reported by other colleagues [54] (Figure S9), but up-regulated 1363 probes (fold change >2, P<0.01) in the NAC089D-MYC expressing plants (XVE089D-13) (Dataset S1). Gene ontology (GO) analysis revealed that the most significant GO term among the NAC089D-MYC-regulated genes is the transcription factor activity (Figure S10), indicating that NAC089 is an important transcriptional regulator. To validate the microarray expression data, 23 genes were selected from the microarray data and their expression was examined by qRT-PCR in two NAC089D-MYC expressing lines. It was found that the expression of these genes was highly induced in both transgenic lines, especially in line XVE089D-13 (Figure 5). Among the 23 selected genes, 13 of them were up-regulated more than two fold by the prolonged ER stress, especially after 12 hr TM treatment (Figure S11). The up-regulation of these 13 genes by ER stress was also checked in the wt and NAC089 knock-down plants (line RNAi089-25). Previously, the BAX inhibitor 1 (BI-1) was reported to be an important modulator of ER stress-mediated PCD in Arabidopsis [31]. The transcription factor WRKY33, which is required for resistance to necrotrophic pathogens, plays critical roles in autophagy [55]. These two PCD markers along with other cell survival UPR markers were also included in the expression study. It was found that the up-regulation of 6 NAC089D-MYC-regulated genes (i.e. AT1G69325, encoding remorin-like protein; AT1G79330, encoding metacaspase MC5; AT2G46240, encoding BCL-2-associated athanogene BAG6; AT3G52350, encoding unknown protein; AT4G30880, encoding lipid transfer protein; and AT5G39820, encoding transcription factor NAC094) and autophagy-related gene WRKY33 was impaired in the RNAi089-25 plants under ER stress condition comparing to the wt control (Figure 6A–B). These results indicate that NAC089 plays critical roles in regulating these ER-stress-induced genes including several PCD-related genes. ER stress also up-regulates several cell survival UPR marker genes and BI-1 in both the wt and RNAi089-25 plants (Figure 6B), suggesting that NAC089 plays minor role in regulating the expression of these genes. In order to know whether NAC089 directly regulates these downstream genes, chromatin immunoprecipitation (ChIP) experiments were carried out with NAC089D-MYC plants (line XVE089D-13) using anti-MYC antibody. It was found that NAC089D-MYC was enriched significantly with fold change greater than 2 at the promoter regions of 7 genes (i.e. AT1G65240, AT1G71390, AT1G79330, AT2G46240, AT4G30880, AT5G39820 and AT5G40010) (Figure 7), indicating that these genes might be the direct targets of NAC089. Among the 7 NAC089 possible targets, the up-regulation of 3 genes (i.e. AT1G65240, encoding aspartyl protease; AT1G71390, encoding receptor-like protein RLP11 and AT5G40010, encoding AAA ATPase 1) by ER stress was not suppressed in the NAC089 knock-down mutants (Figure 6A), suggesting that other factors may also up-regulate these genes under ER stress condition. We concluded that NAC089 has the ability to regulate some of the UPR downstream genes, including the PCD regulatory genes MC5, BAG6 and NAC094, and the autophagy regulatory gene WRKY33. The function of other NAC089 downstream genes in ER stress-induced PCD needs to be investigated in the future.
The unmitigated ER stress is believed to induce PCD in animals [3], as well as in plants [31], [56]. Given that PCD components are not highly conserved between animals and plants [30], [57], our knowledge on ER stress-induced PCD in plants is very limited [58]. Previously, BI-1 and IRE1 were reported to be the negative regulators of PCD in plants [31], [40]. In the current study, a membrane-associated transcription factor NAC089 was identified as an important transcriptional regulator of plant PCD under ER stress condition based on the following evidences: 1) NAC089 is up-regulated by UPR regulators bZIP28 and bZIP60 under ER stress condition; 2) NAC089 relocates from the ER membrane to the nucleus in response to ER stress; 3) Inducible expression of the truncated form of NAC089 induces PCD; 4) Partial loss-of-function of NAC089 confers resistance to chronic ER stress with reduced caspase 3/7-like activity; 5) NAC089 has transcriptional activity and binds to the promoter of many downstream targets; 6) Knock-down NAC089 suppresses the ER-stress-induced expression of several PCD regulators.
As in animals, plant development and adaptations to environmental stresses are intimately connected to PCD [59], [60]. In mammals, PCD is controlled predominately through functionally conserved proteins such as CED9/BCL-2 and BAX, but such genes have not been identified in plants [52]. Interestingly, the heterotrimeric G protein signaling was reported to be involved in ER stress-associated PCD. Null mutants of G beta subunit (AGB1) were more resistant to ER stress than either the wt plants or null mutants of G alpha subunit, but the underlying molecular mechanism was not known yet [61]. On the contrary, Chen and Brandizzi recently reported that the null AGB1 mutants were more sensitive to ER stress [62]. The function of AGB1 in ER stresses response needs to be further clarified. Caspases are cysteine-aspartic proteases that play essential roles in PCD in animals [4]. Plant caspase homologs are not found so far, and the metacaspases were demonstrated to have similar function in plant PCD [53], [63], [64]. Caspase-like activity has been detected in plant PCD associated with xylem formation and adaptations to heavy metal stress, pathogen infection, as well as exposure to ultraviolent-C [50], [65], [66], [67], [68]. In the current study, chronic ER stress induced caspase 3/7-like activity, and such induction was impaired in the NAC089 knock-down plants. Several NAC089 downstream targets including some known PCD regulators were also indentified in the current study. Among them, one metacaspase (MC5) and several other proteases were induced by ER stress, which was suppressed in the NAC089 knock-down plants. BAG (BCL2-associated athanogene) family proteins were originally identified as the anti-cell-death protein in mammals [69]. Among the seven animal BAG homologs found in Arabidopsis [70], overexpression of BAG6 induced PCD in yeast and plants [71], indicating that BAG6 is a pro-death protein in plants. In the current study, BAG6 was induced by ER stress in the Arabidopsis wt plants, which was impaired in the NAC089 knock-down plants. Previously, the soybean NAC transcription factor NAC6/NAC30 was shown to induce caspase 3-like activity and promote extensive DNA fragmentation when it was overexpressed in soybean protoplasts [33]. Here in the current study, we found that one of the direct targets of NAC089, NAC094, is the close-related homolog of soybean NAC6/NAC30 in Arabidopsis. ER stress-induced expression of NAC094 was greatly suppressed in the NAC089 knock-down plants. These results support that NAC089 controls the expression of several PCD-related downstream genes in Arabidopsis under ER stress condition. Interestingly, the autophagy-related gene WRKY33 [55]was also up-regulated by ER stress, which was dependent on NAC089. Other NAC089 downstream genes such as genes encoding protease and nuclease were also identified in the current study. The identification of NAC089 as a PCD regulator provides more opportunities for further understanding new molecular components involved in plant PCD, especially under ER stress condition.
NAC089 is regulated by ER stress at both transcriptional and post-translational levels. At the transcriptional level, NAC089 is up-regulated by ER stress, which is directly controlled by bZIP28 and bZIP60, two important regulators in plant UPR [9], [18], [44], [72]. At the protein level, NAC089 is an ER membrane-associated transcription factor (MTF) and it relocates from the ER to the nucleus under ER stress condition. Interestingly, bZIP28 and bZIP60 are also ER MTFs. bZIP28 is activated through regulated proteolysis. In response to ER stress, bZIP28 relocates from the ER to the Golgi where it is cleaved by two Golgi-resident proteases S1P and S2P, and the C-terminal lumen-facing domain is thought to be responsible for the sensing of ER stress [17], [18], [19], [72], [73]. The activation mechanism for bZIP60 is unconventional, and the activation of bZIP60 is dependent on the ER membrane-localized IRE1 proteins. Under ER stress conditions, bZIP60 mRNA is spliced by IRE1, which results in an open reading frame (ORF) shift and elimination of the transmembrane domain [9], [11], [12]. The N-terminal part of yeast IRE1 is inserted into the ER lumen and plays important role in direct sensing the unfolded proteins in the ER in yeast [74]. Recently, at least 13 MTFs in NAC family are found in Arabidopsis, of which some are activated during development and adaptations to environmental stresses [45], [75], [76], [77], [78]. However, the activation mechanisms of these NAC MTFs are still largely unknown. The NAC089 mRNA does not have the predicted double stem-loop structure that has been shown to be very important for IRE1 splicing [9]. Furthermore, there is no alternative spliced transcript of NAC089 observed in the ER stressed wt seedlings (Figure S12), suggesting that NAC089 might be activated in a manner different from bZIP60. The C-terminal ER lumen facing tail of NAC089 is very short and does not have the canonical S1P cutting site, which implicates that NAC089 might not be proteolytically processed in the same way as bZIP28. We did not include protease inhibitors in the NAC089 activation experiments because most of the protease inhibitors are not permeable to live plant cells. Further investigation of the activation mechanism of NAC089 will improve our understanding of MTFs in plants. Surprisingly, one rare nucleotide polymorphism caused by natural variation in the Arabidopsis Cvi ecotype results in premature stop and constitutive nuclear localization of NAC089, in which the C-terminus (114 AA) including the hydrophobic tail is not translated. Although Cvi ecotype is much more sensitive to fructose than Ler ecotype, expressing the Cvi NAC089 suppresses fructose sensitivity in Ler seedlings [79]. The truncated form (114 AA deletions) of NAC089 was found in the nucleus, however, the activation or nuclear relocation of NAC089 in response to fructose treatment is not reported, and how the truncated form of NAC089 represses fructose signaling is not clear. Since the deletion occurred in Cvi was found neither in over 100 Arabidopsis accessions nor in the Arabidopsis Genome 1001 sequence collections, the biological function of NAC089 in fructose signaling other than in the Cvi ecotype is elusive [79]. Recently, it was reported that fructose feeding induced ER stress in mice [80]. It would be interesting to determine whether the high concentration of fructose could also induce ER stress in Arabidopsis plant. Recently, NAC089 was reported to be involved in redox regulation [81]. Besides its effect on redox status, DTT also inhibits disulphide bond formation and therefore promotes protein misfolding. However, TM is a more specific ER stress inducer because of its specific effect on blocking protein N-glycosylation in the ER. In the current study, both TM and DTT treatments were employed to demonstrate the specific role of NAC089 in ER stress response. Our current study has also advanced the understanding of the function of NAC089 in ER-stress-induced plant PCD and the underlying molecular mechanisms.
How cells make the cell fate selection between life and death remains enigmatic. In human cells, ER stress activates all the three arms of UPR pathways; each branch has different effect on cell survival or cell death, but attenuation of each branch is different. Switch between cell survival and cell death outputs lies in part in the duration of individual branch activity, which guides the cell toward survival or demise [82]. In plants, except the PERK pathway, the bZIP28 and bZIP60 branches of signaling pathway have been previously discovered to regulate downstream genes involved in promoting cell survival [83]. Knock-outs of both branches in the zip28zip60 double mutant causes high sensitivity to ER stresses and accelerated PCD under prolonged ER stress condition [72]. IRE1A and IRE1B redundantly control the activation of bZIP60 and RIDD; knock-outs of both IRE1s in Arabidopsis also promotes PCD while knock-out of single bZIP60 gene has no PCD phenotype [40]. The ER stress-induced up-regulation of NAC089 is dependent on both bZIP28 and bZIP60. Different from the zip28zip60 mutant, knock-down NAC089 confers ER stress tolerance and over-expression of the truncated form of NAC089 promotes PCD. These results may not necessarily be controversial. Firstly, bZIP28 and bZIP60 regulate many survival genes whose expressions are almost completely abolished in the zip28zip60 mutant [72]. Lacking the expression of survival genes in the zip28zip60 mutant may lead to the accelerated PCD. Secondly, NAC089 has substantial constitutive expression and the NAC089 pathway may still operate for PCD in the zip28zip60 mutant even without further NAC089 up-regulation. Thirdly, it is possible to have other PCD pathways turned on to execute PCD in the zip28zip60 mutant. Recently, heterotrimeric G protein signaling [61], vacuolar processing enzyme (VPE)-triggered cell death [42], [84] and IRE1-mediated autophagy [85] pathways were reported to be involved in the ER stress-induced PCD in plants.
A hypothetical working model has been emerged from the current study (Figure 8). When Arabidopsis cells are confronted with ER stress, both bZIP28 and bZIP60 pathways are activated to mitigate the stress by up-regulation of genes involved in protein folding or ERAD to improve survival [1], [83]. The activated bZIP60 also induces its own transcription [44] and another transcription factor NAC103 [86] to amplify the cell survival signal. The ER-localized IRE1 protects cell through a process called RIDD to reduce the protein folding demand [40]. In the meantime, besides the constitutive high expression of NAC089 under normal condition, both bZIP28 and bZIP60 up-regulate the expression of NAC089 under ER stress condition. Up-regulation of NAC089 mRNA may increase the protein level of the membrane-associated NAC089 precursor. However, nuclear relocation of NAC089 is tightly controlled, in which bZIP28 and bZIP60 may play negative roles. When the ER stress is severe, NAC089 is activated and relocates from the ER to the nucleus, inducing the expression of PCD regulators to promote cell death. It is possible that the stress intensity and/or duration of ER stress might determine the signaling output and the final cell fate. Further understanding on how cells balance the cell survival and cell death effects in UPR is of great fascination.
All Arabidopsis (Arabidopsis thaliana) wild-type, T-DNA mutants and transgenic plants in the current study were in the Columbia (Col-0) ecotype background. The double mutant zip28zip60 was made as previously reported [72]. Methods for plant growth were described previously [43]. For short time treatment, different concentrations of TM (5 µg/ml), DTT (2 mM) or BE (10 µM) were added in the half-strength MS medium unless mentioned in the text. For long time treatment, 10 µM BE or various low concentrations of TM were supplied in the solid growth medium. Root length was measured and emergence rate of true leaves was calculated. Total chlorophylls were extracted from seedlings with 80% (v/v) acetone at 4°C overnight and measured from A663 and A646 readings for each sample [87]. All the data in the paper were subjected to Student's t-test or two-way ANOVA (analysis of variance) analysis.
The coding sequence of NAC089 was amplified with PCR and inserted into pSKM36 after digestion with AscI and SpeI restriction enzymes to produce the vector Pro35S:NAC089. Modified green fluorescence protein (mGFP) tag and 4X MYC tag were amplified and inserted into Pro35S:NAC089 at AscI site to generate the Pro35S:mGFP-NAC089 and Pro35S:MYC-NAC089 constructs, respectively. For RNAi construct preparation, part of the NAC089 gene sequence covering cDNA 651–1150 was inserted into pHANNIBAL in both sense and antisense orientations separated by an intron sequence. The entire RNAi cassette was cut with NotI and inserted into pART27 to make the RNAi expression vector. To express the dominant negative fusion protein ProNAC089:NAC089D-EAR, the sequence encoding EAR motif (QDLDLELRLGFA) was synthesized and firstly inserted into pCAMBIA1300; about 1 kb upstream sequence of NAC089 and sequence encoding the truncated form of NAC089 (NAC089D, aa 1–316) were amplified and subsequently inserted. To generate the conditionally overexpression construct [48], nucleotides encoding NAC089D was amplified and inserted into pER10M. For dual luciferase activity assay, fragment of the NAC089 promoter (−98 bp to −46 bp relative to the TSS site) was synthesized and inserted into pGreen0800-II after the 35S minimal promoter was introduced. Constructs expressing bZIP28D and bZIP60S were made as described [9], [18]. For protein expression in E. coli, the respective sequence of bZIP18D (aa 1–321) or bZIP60T (aa 87–217) was amplified and inserted into pET28 and pET32, respectively. All the primers were listed in Table S1 and error-free clones were introduced into plants by either transient expression or stable transformation.
Total protein was extracted from plants with extraction buffer described previously [17]. Membrane fraction and nuclear fraction were prepared with sucrose gradient centrifuge according to the standard protocol [88]. Proteins were resolved on 8–10% SDS-PAGE gels and visualized by western blotting using antibodies against c-MYC (Santa Cruz Biotechnology), nuclear protein marker histone H3 (Abcam) and ER protein marker BiP (Santa Cruz Biotechnology).
Caspase-like activity was measured with luminescent assays based on DEVD short peptides with Caspase-Glo 3/7 Assay Kit (Promega). Both caspase 3 and caspase 7 recognize the same DEVD substrate. Briefly, seedlings were harvested after various treatments and total proteins were extracted with liquid nitrogen in a buffer containing 100 mM sodium acetate, pH 5.5, 100 mM NaCl, 1 mM EDTA, and 5 mM DTT. To measure caspase-3/7 activity, 30 µl caspase-3/7 luminogenic substrate (Z-DEVD-aminoluciferin) was added to 50 µg protein extracts and incubated at 22°C for 1 hr protected from light. The luminescence of each sample was measured with the Synergy 2 Multi-Mode Microplate Reader (BioTek). For dual-luciferase activity assays [89], Arabidopsis leaf protoplasts were isolated from 4-week-old soil-grown seedlings and transfected according to a standard protocol [43] with various reporter constructs or cotranstransfected with different effectors. Firefly and renilla luciferase were quantified with Dual-Luciferase Reporter Assay Kit (Promega) according to the manufacturer's instructions in the Synergy 2 Multi-Mode Microplate Reader (BioTek).
For FDA and DAB staining, seedlings were stained with 2.5 µg/ml FDA (Sigma-Aldrich) in phosphate-buffered saline for 10 min or 1 mg/ml DAB (pH 5.5, Sigma-Aldrich) for 2 hr at room temperature, immersed into boiled ethanol for 10 min according to the standard protocol [31]. For PI and DAPI staining, samples were stained using DAPI (Sigma-Aldrich) at 1 µg/ml in 0.1% (v/v) Triton X-100 for 10 min or PI (Sigma-Aldrich) at 10 µg/ml for 1 min, and washed twice with water. For in situ TUNEL staining, roots were stained in microcentrifuge tubes (1.5 ml) using the in situ cell death detection kit (Takara) according to the manufacturer' instructions. Except DAB staining, which was observed under differential interference contrast (DIC) microscopy, other staining, BiFC and subcellular localization of mGFP-NAC089 were visualize with laser confocal fluorescence microscopy (Zeiss LSM A710).
ChIP was performed according to the standard protocols. Briefly about 3 g of 2-week-old XVE089D transgenic seedlings were treated with either 10 µM BE or DMSO (solvent control) for 16 hr and fixed with 1.0% formaldehyde for 10 min subsequently. Antibodies against c-MYC (Santa Cruz Biotechnology) and GST (IgG control, Abmart) were used for immunoprecipitation. Protein-A-agarose beads were blocked with salmon sperm DNA and used to pull down the protein-DNA complex. Equal amounts of starting plant material and the ChIP products were used for quantitative PCR. Primers were selected in the promoter regions of each selected gene. DNA levels were calculated relative to TA3 (AT1G37110) using a comparative threshold cycle method. The ChIP experiments were performed 3 times with biological replications and similar results were obtained. For microarray analysis or qRT-PCR, the wt control, XVE089D and NAC089 RNAi plants were grown vertically on agar plates for one week and then transferred to 1/2 MS liquid supplied with 10 µM BE or DMSO or TM for a period of time as noted. Total RNA was extracted and purified according to the manufacturer's instructions [43]. Agilent Arabidopsis gene chips (4X44K) were used to compare the gene expression profiles with three independent replications. P-values were calculated and used to select the genes that are up-regulated by NAC089D-MYC (cut-off: P<0.01, fold change >2). Microarray data from this article can be found in ArrayExpress under the accession number E-MTAB-1377. Quantitative PCR and RT-PCR were routinely conducted [43] and all the primers are listed in Table S1. GO analysis was performed with AgriGO (http://bioinfo.cau.edu.cn/agriGO/analysis.php).
EMSA was performed using a LightShift Chemiluminescent EMSA Kit (Pierce), according to the manufacturer's protocols [43]. Briefly, each 20 µL binding reaction contained 2 µl binding buffer, 0.3 µl Poly (dI-dC), 4 µg purified protein, 0.83 µmol biotin-labeled probe or certain amount of unlabeled probe as the competitor. The pNAC089 wt or pNAC089M1-M3 probes were created by annealing together complementing oligonucleotides and biotinylated with a labeling kit (Pierce). His-tagged bZIP28D or Trx-His-tagged bZIP60T proteins were expressed in E. coli strain BL21 and purified with Ni-NTA agarose beads (Qiagen). The binding reactions were allowed to incubate on ice for 1 hr and run on a 5% polyacrylamide mini-gel (37.5∶1 acrylamide-bisacrylamide in 0.5× Tris-Borate-EDTA (TBE) containing 3% glycerol). The complex was transferred to a membrane and developed according to a standard protocol.
|
10.1371/journal.pgen.1000749 | Stoichiometry of Base Excision Repair Proteins Correlates with Increased Somatic CAG Instability in Striatum over Cerebellum in Huntington's Disease Transgenic Mice | Huntington's disease (HD) is a progressive neurodegenerative disorder caused by expansion of an unstable CAG repeat in the coding sequence of the Huntingtin (HTT) gene. Instability affects both germline and somatic cells. Somatic instability increases with age and is tissue-specific. In particular, the CAG repeat sequence in the striatum, the brain region that preferentially degenerates in HD, is highly unstable, whereas it is rather stable in the disease-spared cerebellum. The mechanisms underlying the age-dependence and tissue-specificity of somatic CAG instability remain obscure. Recent studies have suggested that DNA oxidation and OGG1, a glycosylase involved in the repair of 8-oxoguanine lesions, contribute to this process. We show that in HD mice oxidative DNA damage abnormally accumulates at CAG repeats in a length-dependent, but age- and tissue-independent manner, indicating that oxidative DNA damage alone is not sufficient to trigger somatic instability. Protein levels and activities of major base excision repair (BER) enzymes were compared between striatum and cerebellum of HD mice. Strikingly, 5′-flap endonuclease activity was much lower in the striatum than in the cerebellum of HD mice. Accordingly, Flap Endonuclease-1 (FEN1), the main enzyme responsible for 5′-flap endonuclease activity, and the BER cofactor HMGB1, both of which participate in long-patch BER (LP–BER), were also significantly lower in the striatum compared to the cerebellum. Finally, chromatin immunoprecipitation experiments revealed that POLβ was specifically enriched at CAG expansions in the striatum, but not in the cerebellum of HD mice. These in vivo data fit a model in which POLβ strand displacement activity during LP–BER promotes the formation of stable 5′-flap structures at CAG repeats representing pre-expanded intermediate structures, which are not efficiently removed when FEN1 activity is constitutively low. We propose that the stoichiometry of BER enzymes is one critical factor underlying the tissue selectivity of somatic CAG expansion.
| Huntington's disease (HD) is a neurodegenerative disorder that belongs to a family of genetic diseases caused by abnormal expansion of CAG/CTG repetitive sequences. The instability of trinucleotide repeat expansions in germline and somatic cells has deleterious clinical consequences in HD. For instance, transmission of longer repeats to offspring results in an earlier onset of disease, where extensive somatic expansion in the striatum, the brain region primarily affected in HD, is proposed to accelerate disease pathology. Thus, understanding the mechanisms of trinucleotide repeat instability is a major interest. We have examined the role of oxidative DNA damage and base excision repair (BER) in somatic instability, which is tissue-selective and age-dependent. We show that oxidative DNA lesions abnormally accumulate at CAG expansions in a length-dependent, yet age- and tissue-independent manners, likely due to the secondary structures formed by CAG repeats that limit access of enzymes initiating BER. In addition, our data indicate that repair by BER enzymes of some of the accessible lesions results in somatic expansion when the ratio of FEN1 to POLβ is low, as found to occur in the striatum. Our results support BER enzyme stoichiometry as a contributor to the tissue selectivity of somatic CAG expansion in HD.
| Huntington's disease (HD) is a neurodegenerative disorder caused by aberrant expansion of a CAG repeat tract within the coding sequence of the Huntingtin (HTT) gene, resulting in the production of a mutant protein with a toxic elongated polyglutamine (polyQ) stretch. This dominantly inherited disease, which shares the same mutation mechanism with eight other neurodegenerative disorders, is characterized by preferential and progressive degeneration of the medium-spiny neurons in the striatum. The HD mutation is unstable in germline and somatic cells, and CAG repeat expansion in both cell types has deleterious clinical consequences. Transmission of the mutation to offspring, in particular by fathers, is characterized by an expansion bias, leading to the phenomenon of anticipation [1]–[3], whereby the disease tends to worsen over successive generations due to the production of mutant huntingtin protein with an increased glutamine tract length that triggers earlier disease onset and increased disease severity. In addition, somatic CAG repeat expansion occurs in several tissues, including the brain. Interestingly, different brain regions are unequally affected [2],[4]; somatic instability is extensive in striatal neurons but very limited in cerebellar neurons, which are largely spared by the disease [5],[6]. It is therefore proposed that somatic expansion in the striatum and other target tissues, leading to the production of increasingly toxic mutant huntingtin proteins, accelerates HD pathology and acts as a disease-modifier [5]–[7]. Consistent with this hypothesis, an early mutant huntingtin phenotype is delayed in HD mice in which somatic instability is prevented by a deficiency in either the Msh2 or Msh3 mismatch repair (MMR) protein [8],[9].
Several studies support the notion that DNA repair contributes to somatic instability of CAG repeats, though other DNA-associated processes, such as replication, may also play a role [10]. In particular, at least two repair pathways appear to regulate in vivo somatic instability in the brain of HD mouse models. In addition to MMR, whose role is well documented [8],[9],[11],[12], base excision repair (BER), which is specialized in DNA base damage removal, has recently been implicated. Indeed, somatic CAG repeat instability was reduced in HD mice lacking the DNA glycosylase OGG1, an enzyme that initiates BER of 8-oxoguanine (8-oxoG) lesions [13]. Furthermore, the global level of oxidative DNA damage increased with age in the brains of HD mice, while OGG1 activity did not change, suggesting that the level of DNA oxidation at CAG expansions might underlie the age dependence of somatic instability [13]. In vitro biochemical and yeast studies have further suggested that other BER enzymes might modulate somatic instability. DNA polymerases, including DNA polymerase beta (POLβ), a central participant in most BER responses, can promote CAG repeat extensions in vitro [13]. In contrast, Flap Endonuclease-1 (FEN1), involved in the long-patch BER (LP-BER) subpathway, prevents CAG repeat expansion in yeast [14].
BER is characterized by a sequence of highly coordinated steps [15],[16], starting with damaged base recognition and removal by a DNA glycosylase, leading to formation of an apurinic/apyrimidinic (AP) site. In the brain, OGG1 appears to be the major glycosylase involved in removing endogenous lesions such as 8-oxoG from the genome [17]. Cleavage at the AP site is performed by the major AP endonuclease, APE1, which incises 5′ to the damage leaving behind a 3′OH and a 5′deoxyribose phosphate (5′dRP), which is subsequently removed by the dRP lyase activity of POLβ. A DNA polymerase then fills the resulting gap. In single-nucleotide BER (SN-BER), POLβ incorporates the missing nucleotide. However, when the 5′-terminal residue is refractory to POLβ lyase activity, repair proceeds via LP-BER. In this case, multiple nucleotides are incorporated by one of several DNA polymerases, including POLβ, POLδ or POLε, through a strand displacement mechanism. Subsequently, FEN1 is required to remove the 5′-flap structure formed during LP-BER synthesis. Interestingly, a number of studies suggest that BER enzyme activities vary with age and between tissues, indicating that the repair kinetics of oxidative DNA damage may be age- and tissue-dependent [18].
We sought to clarify the role of DNA damage and BER in the tissue specificity and age dependence of somatic CAG expansion in HD. First, we asked whether DNA lesions accumulated at CAG repeats, particularly in the striatum. Second, we asked whether BER proteins and activities displayed a specific or unique pattern of expression in the striatum. We have systematically analyzed DNA damage at CAG repeats, BER protein levels, and BER activities in both the unstable striatum, and in the mostly stable cerebellum of young and aged HD mice. Our results show that the stoichiometry of BER enzymes, rather than DNA damage levels, correlates with the tissue selectivity of somatic CAG expansion.
Several studies have suggested that HD pathology contributes to neuronal aging. Specifically, oxidative stress and consequently oxidative DNA damage, which accumulates during normal aging, are elevated in HD neurons [19]. Thus, the “aging process” for HD neurons can potentially be envisioned as two separate components: (i) normal and (ii) disease-associated. We set out to determine the role of these two components in the process of somatic expansion. To this end, we compared various R6 transgenic mouse models of HD, which recapitulate several features of the human pathology, including somatic instability [6],[20]. R6/1 mice express the first exon of the human HTT gene with ∼125 CAG repeats and develop a progressive disease leading to death within approximately 8 months. The same transgene is expressed in R6/2 mice, yet is inserted in an alternate genomic location. R6/2–160 CAG and R6/2–100 CAG (spontaneously derived from R6/2–160 CAG) mice contain 160 and 100 CAG repeats, respectively. In both R6/2 lines, disease onset and progression is much more severe than in the R6/1 mice. Indeed, R6/2–160 CAG and R6/2–100 CAG mouse lines die at 3 and 4.5 months, respectively [21].
Somatic cell CAG instability was determined in R6/1 mice of various ages using standard PCR amplification and Genescan analysis from two brain regions that exhibit contrasting degrees of CAG instability, i.e. the striatum and the cerebellum (Figure 1A, top panel). To obtain a baseline reference for the number of CAG repeats, we analyzed tails from mice at 3 weeks of age, because the CAG repeat region from this tissue is mostly stable when taken from very young animals. As expected, quantification of instability showed that both the median and the amplitude of the Genescan profiles increased upon aging in the R6/1 striatum, but to a much lesser extent in the R6/1 cerebellum (Figure 1B, upper panels). As previously described, the profiles generated from aging striata, which display extensive somatic instability, were bimodal [6].
Comparison of 13 week-old R6/1 mice with R6/2–160 CAG and R6/2–100 CAG mice of similar ages, 12 and 16 weeks, respectively, showed that CAG repeat expansion was significantly higher in the striatum of R6/2 mice (Figure 1A and 1B, lower left panel). Thus, the process of somatic CAG instability was quicker in R6/2 mice than in R6/1 mice, regardless of the repeat length in the R6/2 mice, suggesting that disease-associated mechanisms and/or positional effects are involved in somatic expansion. However, when R6/1 and R6/2–160 CAG were compared at similar pathological stages (e.g. end-stages), we found that somatic instability was significantly higher in the striatum and the cerebellum of the R6/1 mice (Figure 1A and 1B, lower right panel). In agreement with previous results [6], these data indicate that normal aging contributes to somatic CAG instability.
It was recently proposed that DNA oxidation contributes to the age dependence of somatic CAG instability, due to damage accumulation in the brain during aging [13]. To clarify the role of DNA damage (and aging) in somatic expansion, we quantified AP sites within the whole genome in the striatum and cerebellum of R6/1 and R6/2 mice. AP sites are formed either spontaneously or as intermediates during repair of oxidized, deaminated or alkylated bases. AP sites increased with age in the striatum of R6/1 mice, with greater abundance at 8 months than at 3 months (Figure 2A). AP sites were also significantly higher at 8 months in the R6/1 striatum than in the striatum of littermate control animals, potentially reflecting increased oxidative stress or reduced repair in the HD mouse striatum. In contrast, the levels of AP sites detected in the cerebella of R6/1 were similar to those measured from control mice at 8 months (Figure 2A). To our surprise, we found that AP sites were significantly higher in the cerebellum vs striatum in age-matched animals for both the R6/1 and R6/2 lines (Figure 2A), indicating the level of DNA damage does not correlate with the tissue selectivity of somatic CAG instability in HD mice.
We then asked whether DNA damage specifically accumulates at CAG expansions in the striatum of aging mice, as proposed by Kovtun et al. [13]. We evaluated the level of DNA oxidation at CAG expansions based on digestion of genomic DNA by the bacterial DNA glycosylase Fpg followed by real-time quantitative PCR (QPCR) as previously described [22]. The efficacy of PCR amplification was predicted to be reduced if the amplified DNA region contained Fpg sensitive sites, which includes several oxidative base lesions (i.e. 8-oxoG, 8-oxoadenine, fapy-guanine, methy-fapy-guanine, fapy-adenine, aflatoxin B1-fapy-guanine, 5-hydroxy-cytosine, 5-hydroxy-uracil), as well as AP sites, all of which are converted to strand breaks by the enzyme. As a control, we amplified a portion of the orthologous HTT gene in mouse (Hdh), which encompasses seven interrupted CAG repeats. We chose primers to amplify PCR fragments from the CAG-expanded transgene and the control Hdh locus that were equivalent in size and GC content. The sizes of the CAG-expanded and Hdh PCR fragments are 440 bp (corresponding to 120 CAG repeats) and 420 bp, respectively, with the GC content reaching 67.2% and 72.2% (Figure S1).
Given that the relative PCR efficiency was >0.8, the number of Fpg sensitive sites at the murine Hdh control locus appeared to be negligible in the striata and cerebella of both wild-type and R6/1 mice at 6 and 38 weeks of age, suggesting that spurious oxidation did not significantly affect the assay (Figure 2B, upper panel). In contrast, Fpg sensitive sites accumulated significantly at CAG expansions in the R6/1 mice, as revealed by a reduced relative PCR efficiency of ∼0.6 (Figure 2B, lower panel). Surprisingly, damage accumulation at CAG expansions was neither age- nor tissue-dependent, as the levels were similar in the striatum and cerebellum of young (6 weeks) and old (38 weeks) mice (Figure 2B).
We then measured Fpg sensitive sites in the striatum of mice with different CAG repeat lengths. For this purpose, we analyzed HD knockin mice heterozygous for the HD mutation and bearing either 18, 48 or 115 CAG repeats [23]. Accumulation of Fpg sensitive sites was CAG repeat length-dependent, as only mice with 115 CAG repeats displayed significantly reduced PCR amplification (Figure 2C). Thus, DNA damage at CAG expansions is not sufficient to trigger somatic instability, as accumulation of Fpg sensitive sites did not correlate with the age dependence or tissue selectivity seen in HD. The repeat length-dependence of DNA damage accumulation suggests that a build-up of presumably oxidative DNA lesions at CAG repeats might be dictated by the propensity to form alternate structures in DNA.
DNA sequences containing CAG repeats spontaneously form secondary structures such as hairpins [24],[25]. To assess the role of DNA structure in oxidative DNA damage accumulation at CAG expansion, we synthesized substrates mimicking hairpin structures formed by CAG sequences and containing either an oxidized base (8-oxoG) or an AP site analog (tetrahydrofuran, THF) at the tip of the hairpin. Two series of hairpin oligonucleotides were generated containing an 8-oxoG or THF modification at (1) the 3rd position within a 3-nucleotide loop (referred to as CAG1oxoG and CAG1THF) or (2) the 2nd position within a 4-nucleotide loop (referred to as CAG2oxoG and CAG2THF) (Figure 3A). Design of hairpin oligonucleotides included sites for EcoRI and BamHI restriction enzymes (Figure 3A), which allowed for verification of hairpin structure formation. Digestion with either restriction enzyme revealed that the oligonucleotides formed the expected stable hairpin structure (Figure 3B). As additional controls, we designed linear oligonucleotides containing either an 8-oxoG lesion (referred to as 34-oxoG) or an AP site analog (referred to as 34THF). In contrast to OGG1, APE1 can incise single stranded oligonucleotides with varying efficiency, and thus, 34THF was used directly as a control [26],[27]. 34-oxoG, on the other hand, was annealed to a complementary oligonucleotide (34G) to form a control double stranded DNA substrate (34-oxoG/34G). To test whether OGG1 or APE1 were able to incise damage in hairpin substrates, increasing concentrations of human OGG1 or APE1 were incubated with the appropriate 5′-32P labeled oligonucleotide substrates. As shown in Figure 3C, no incision products were detected when using the aforementioned CAG hairpin substrates, even when OGG1 or APE1 concentrations were high (i.e. <2.5% product at 30 ng OGG1 or <3.5% product at 1 ng APE1). Conversely, both enzymes, even at low concentrations, efficiently incised their respective control substrates. Thus, OGG1 and APE1 are unable to efficiently incise hairpin substrates in vitro, supporting the idea that oxidative DNA lesions accumulate at CAG repeats because they can form secondary structures that are refractory to processing by BER enzymes.
Oxidative DNA damage is repaired by the BER pathway, which appears to be regulated in an age-dependent and tissue-specific manner [18]. We therefore asked whether the activity or stoichiometry of BER enzymes could contribute to the age dependence or tissue selectivity of somatic CAG instability in HD. We first determined mRNA expression levels of several major BER genes in the striatum and cerebellum of R6/1 HD and control mice at 40 weeks of age using quantitative RT-PCR. The BER genes that we tested were Ogg1, Ape1, Fen1 and Polβ. Expression of most of these genes was similar between the transgenic and control mice, both in the striatum and cerebellum, except for Ogg1, which was downregulated 2-fold in the cerebellum of R6/1 mice (Figure 4A).
We next analyzed protein levels of several major BER enzymes by Western blotting (Figure 4B). The results showed that FEN1 was more abundant in the cerebellum than in the striatum of 40 week-old R6/1 mice. FEN1 was also produced at higher levels in R6/1 mice compared to control animals. APE1, POLβ and OGG1 levels were not significantly changed when comparing striatum vs cerebellum or R6/1 vs control animals. However, OGG1 tended to be lower in the cerebellum of R6/1 mice relative to the striatum, whereas APE1 appeared to be higher in the cerebellum of R6/1 mice. As a consequence, the stoichiometry of BER proteins was different between the cerebellum and striatum of R6/1 mice (Figure S3A). Using recombinant FEN1 and POLβ proteins as standards and Western blotting, we more precisely determined the stoichiometric ratio between FEN1 and POLβ in the striatum and cerebellum of R6/1 mice (Figure S3B). The results revealed that the FEN1/POLβ molar ratio was 1.4+/−0.3 in the striatum and 6+/−1.2 in the cerebellum corresponding to a 4-fold difference, thus supporting that the stoichiometry of BER enzymes is different in the striatum vs cerebellum of R6/1 mice, largely due to a significant disparity in FEN1 protein levels. We also analyzed the high-mobility group box 1 (HMGB1) protein, which was recently identified as a cofactor of BER that stimulates the activities of APE1 and FEN1 [28]. Interestingly, we found that HMGB1 protein was 2–3 fold higher in the cerebellum than in the striatum of R6/1 animals, suggesting that APE1 and FEN1 activities may be further activated by HMGB1 in the cerebellum of R6/1 mice. Finally, the level of BER enzymes and HMGB1 were similar in the cortex, another brain region characterized by substantial somatic expansion, to those in the striatum of R6/1 mice (data not shown), indicating that the stoichiometry of BER proteins might modulate the propensity of a given tissue for somatic CAG expansion.
To evaluate further the role of BER in the tissue selectivity or age dependence of somatic CAG instability, we compared the enzymatic activities of the major steps of BER in extracts prepared from the striatum and cerebellum of R6/1 mice at both early- (6 weeks) and late- (36 weeks) stages. In addition, BER activities were compared between R6/1 and control littermate animals. Using standard oligonucleotide substrates (Figure S2), we performed in vitro biochemical assays allowing for evaluation of 8-oxoG lesion removal, gap filling, 5′-flap excision and AP endonuclease activities (Figure 4C). OGG1, POLβ, FEN1 and APE1 are most likely the main enzymes responsible for these activities, respectively, but other enzymes might contribute to the various activities as well. Strikingly, we did not observe significant differences between R6/1 mice and control animals regardless of the activity examined. Our results revealed that 8-oxoG incision tended to increase with age, both in the striatum and cerebellum of R6/1 and control mice, and to be higher in the striatum than in the cerebellum. In young animals (both R6/1 and controls), gap filling activity was significantly higher in the cerebellum compared to the striatum (by 2- to 3-fold). This difference disappeared at 36 weeks, due to an overall decrease in gap filling activity in the cerebellum (by 1.5- to 2-fold). AP endonuclease activity was slightly elevated in the cerebellum of both 36 week-old R6/1 and control mice (1.5 fold higher) when compared to the striatum. Most remarkably, 5′-flap incision activity was increased 5- to 10-fold in the cerebellum when compared to the striatum for both control and R6/1 mice. Furthermore, 5′-flap incision dramatically increased in the cerebellum upon aging, while it remained very low in the striatum. Finally, additional incision products were detected only in the cerebellum of 36 week old R6/1 mice, indicating that other exo- or endo-nuclease activities may be able to remove the 5′-flap structure in the cerebellum of aging transgenic animals.
Our results indicate that key steps implicated in BER, specifically downstream of 8-oxoG incision (OGG1) activity, are less efficient in the striatum than in the cerebellum of R6/1 mice. They also show that aging subtly modifies the stoichiometry of BER enzymes in a tissue-specific manner. Indeed, upon aging, 8-oxoG incision activity tends to increase both in the striatum and cerebellum, gap-filling activity decreases in the cerebellum and remains stable in the striatum, and flap endonuclease activity dramatically increases in the cerebellum, but remains stable, and low, in the striatum. These findings suggest that the LP-BER subpathway, which requires FEN1, is much less efficient in the striatum than in the cerebellum of R6/1 mice.
To further interrogate the role of BER in somatic CAG instability in vivo, we performed chromatin-immunoprecipitation (ChIP) experiments. To set up conditions, we first used an antibody against acetylated K9/14 histone H3 (AcH3K9/14) that immuoprecipitates transcribed regions. ChIPed DNA was amplified using two primer sets encompassing the CAG-expanded transgenic region and a comparable portion of the mouse Hdh gene as a control. The results show that AcH3K9/14 is specifically and highly enriched at CAG expansions and at the Hdh locus in both the striatum and the cerebellum of R6/1 and R6/2 HD mice (Figure 5A and 5B, lower panels). This study indicates that our experimental conditions allowed for significant and specific detection of proteins bound to both CAG expansions and the Hdh gene. We then employed several antibodies against BER proteins. α-OGG1 and α-APE1 antibodies did not reveal any specific enrichment at either CAG expansions or at the Hdh locus (data not shown). However, we did find that POLβ was specifically enriched at CAG expansions in the striatum, but not in the cerebellum of R6/1 mice. Interestingly, this enrichment tended to increase with age (Figure 5A). POLβ was also specifically enriched at CAG expansions in the striatum of R6/2 mice (Figure 5B). Thus, enrichment of POLβ at CAG expansions was tissue-specific. These in vivo data support the concept that POLβ promotes somatic expansion [13].
Having shown that the FEN1/POLβ molar ratio is 1.4+/−0.3 in the striatum and 6+/−1.2 in the cerebellum of R6/1 mice (Figure S3B), that FEN1-like nuclease activity is 5- to 10-fold more elevated in the cerebellum (Figure 4C), and that POLβ is specifically enriched at CAG expansions in the striatum of HD mice (Figure 5), we suspected that the FEN1/POLβ ratio might be critical in determining the propensity of a given tissue to experience somatic CAG instability. To explore this possibility, we examined the effects of varying FEN1∶POLβ stoichiometry on POLβ DNA synthesis using a 100-mer substrate consisting of a 23-mer primer followed by a single nucleotide gap containing a 5′-abasic site analog (THF) and nineteen CAG repeats (Figure 6A). The effect of FEN1, at different molar ratios, on POLβ DNA synthesis activity was evaluated as follows. We employed purified POLβ at a single concentration, determined to give predominantly single nucleotide incorporation, and varied FEN1 to give 1∶2, 1∶1, 2∶1, 4∶1 and 8∶1 FEN1∶POLβ stoichiometries. We found that enhancement of POLβ multiple nucleotide incorporation, presumably due to strand displacement synthesis, was achieved at FEN1∶POLβ ratios of 1∶1 to 2∶1 and, to a lesser extent, 4∶1. Interestingly, at the 8∶1 ratio, POLβ multiple nucleotide synthesis was inhibited (Figure 6B). Under identical conditions, we observed no significant change in FEN1 flap endonuclease activity, although we did observe increasing 5′-flap endonuclease activity with increasing FEN1 protein (data not shown). These results support a model whereby BER enzyme stoichiometry affects the outcome of DNA damage processing, and as a consequence, potentially influences trinucleotide repeat instability.
We have shown that oxidative DNA damage accumulates at CAG expansions in a CAG-repeat length-dependent, but age- and tissue-independent manner. Our data indicate that the DNA structure of CAG repeats likely contributes to this accumulation, as BER enzymes such as OGG1 and APE1 were unable to process lesions within hairpin configurations. In addition, comparing the striatum and cerebellum of HD mice, we have discovered a correlation between the propensity for somatic CAG expansion and BER protein stoichiometry and enzymatic activities. In particular, somatic instability correlates mainly with low flap endonuclease activity, as seen in the striatum. Our results also suggest that POLβ contributes to in vivo somatic expansion, as it is specifically enriched at CAG repeats in the striatum of aging animals. We thus propose a model whereby BER contributes to somatic CAG instability (Figure 7). In this model, accumulation of DNA damage at CAG repeats results from low accessibility of BER enzymes to lesions in pre-existing secondary DNA structures formed within CAG repeats. Repair of accessible lesions, which would arise due to the dynamic nature of trinucleotide repeat sequences, would be initiated by a protein such as OGG1 as proposed by Kovtun et al. [13] and potentially lead to somatic CAG expansion. This outcome would be particularly likely where there is poor cooperation between the strand displacement activity of POLβ and the 5′-flap excision activity of FEN1 during LP-BER.
The models proposed to explain disease-associated trinucleotide repeat instability involve specific stable secondary DNA structures. In vitro approaches based on substrates containing CAG/CTG repeats have shown that stable slipped strand DNA or hairpin structures form in a repeat length-dependent manner [24],[25]. Furthermore, three repair outcomes, including correct repair, escaped repair and error prone repair, have been observed for plasmid-based substrates that mimic slipped strand structures formed by CAG/CTG expansions, supporting the idea that processing of such structures by neuronal proteins can result in somatic expansion [29]. Although direct evidence is still lacking, several observations support that CAG/CTG expansions form secondary structures in vivo and the stability of these structures underlies the process of CAG/CTG repeat instability. First, longer repeats are more unstable in humans and mouse models of CAG/CTG repeat diseases [1],[23]. Second, CAG/CTG repeat sequence interruptions, which reduce the propensity to form slipped strand DNA structures, also prevent instability in mice and humans [30],[31]. Third, our data showing an extensive and repeat length-dependent accumulation of oxidative DNA damage at CAG expansions in HD mice imply that secondary structures form that are resistant to repair enzymes at the repeat locus. We in fact demonstrate that 8-oxoG and AP lesions, when located within the hairpin loop, cannot be efficiently incised in vitro by OGG1 and APE1 respectively. These results are in agreement with other studies showing that excision of 8-oxoG by OGG1 is paired-base-dependent and that the repair efficiency of some BER enzymes is affected by surrounding sequence contexts [32]. Thus, the probability that 8-oxoG lesions, an abundant endogenous DNA damage, are detected at CAG/CTG expansions by OGG1 should be significantly reduced if repeat expansions form secondary structures such as hairpins, as we indirectly demonstrate (in Figure 2B and 2C).
Our results showing that POLβ is enriched at CAG expansions, in combination with the results from Kovtun et al. [13] that provide evidence that OGG1 modulates the extent of somatic CAG instability, indicate that at least some of the oxidative lesions present at CAG/CTG repeats are (or become) accessible and processed by the BER pathway. This would imply that the secondary structures formed by CAG/CTG repeats are dynamic in nature. Several studies support that MMR, replication and transcription contribute to instability of CAG/CTG repeats, although the underlying mechanisms remain elusive [10]. These physiological processes induce chromatin remodeling and strand unpairing and, therefore, could provide a window for remodeling of secondary structures. While replication is unlikely to be involved in somatic instability in neurons, MMR and transcription could possibly interplay with BER at repeat expansions. During transcription, chromatin opens and the secondary structures in the transcribed strand have to be disrupted to allow RNA polymerase to proceed, only to re-form at a slightly different place. Thus, transcription could indirectly increase accessibility and the probability of repair of some oxidative DNA lesions by BER. The mechanism by which MMR contributes to somatic instability of CAG/CTG repeats is yet unknown. However, recent data indicate that functional MMR is required [33],[34], suggesting that MMR proteins not only bind, but also disrupt loops or secondary structures at CAG/CTG repeats. Whether BER cooperates with other DNA-associated mechanisms to promote somatic CAG instability is an intriguing possibility.
Using HD mice, we have found that the global level of AP sites was elevated in the mouse cerebellum, which displays little instability relative to the striatum (Figure 2A). In addition, abnormal accumulation of DNA damage at CAG expansions was similar in cerebellum and striatum and did not increase with age (Figure 2B). Thus, the propensity for somatic instability, which increases with age and varies between tissues, does not strictly correlate with levels of DNA lesions at CAG expansions, unlike the toxic oxidation cycle model proposed by Kovtun et al. [13]. We have found that the corresponding activity and relative stoichiometry of major BER enzymes, including POLβ, APE1 and FEN1 varied between the striatum and cerebellum and to a lesser extent with age (Figure 4 and Figure S3). In particular, FEN1 protein level and 5′-flap endonuclease activity were much lower in the striatum than in the cerebellum. We therefore propose that stoichiometry and the relevant activity levels of the corresponding BER enzymes, rather than the level of oxidative DNA damage at repeats, is crucial in determining the probability that repair of lesions at CAG expansions drives somatic expansion.
A 5′-flap endonuclease activity, such as that of FEN1, has for some time been implicated in models of instability for CAG/CTG repeats. In yeast, CAG/CTG instability is enhanced in strains defective for rad27 (FEN1) in a repeat-length dependent manner [14],[35]. It has been hypothesized that flap structures formed at CAG/CTG repeats during replication or repair inhibit FEN1, because they form complex secondary structures. In agreement, in vitro results have shown that secondary structures, such as hairpins, reduce processing by FEN1 at CAG/CTG repeats in a length-dependent manner [36]. Ligation of the unprocessed flap to produce an expansion mutation is supported by a study in yeast showing that overexpression of cdc9 (Ligase I) increases the rates of trinucleotide repeat expansion [37]. However, FEN1 haploinsufficiency did not change the extent of somatic instability in HD and DM1 mice, though intergenerational expansion tended to increase in the HD background [38],[39]. As complete inactivation of Fen1 is lethal, HD and DM1 mouse models could only be examined in the Fen1 heterozygous state. One cannot therefore exclude that a compensatory activity prevents exacerbation of repeat instability in a Fen1+/− background. By showing that FEN1 protein and 5′-flap endonuclease activity levels in the striatum and cerebellum are inversely correlated with the propensity for somatic expansion, we revive a role for FEN1 in preventing CAG repeat instability.
In BER, FEN1 is required to remove the 5′-flap structure generated during LP-BER strand displacement DNA synthesis [40],[41]. Interestingly, studies support that the strand displacement activity of POLβ is also crucial in LP-BER in neuronal cells [42]. Furthermore, in vitro experiments have shown that FEN1 strongly stimulates the strand displacement activity of POLβ and, reciprocally, POLβ stimulates FEN1 [43],[44]. It has been suggested that the functional interaction between POLβ and FEN1 controls incision product size in LP-BER: the tighter the cooperation, the shorter the product. Thus, cooperation between POLβ and FEN1 is believed to be essential to allow efficient repair of an oxidative lesion via LP-BER [43],[44]. Our results show that POLβ is enriched at CAG repeats in the striatum of R6/1 mice (Figure 5). Additionally, gap filling and flap endonuclease activities are much lower in the striatum than in the cerebellum (Figure 4), and the molar ratio of FEN1/POLβ proteins is decreased by about 4-fold in the striatum when compared to the cerebellum (Figure S3B). Additionally, we show that differential molar ratios of FEN1 influence DNA synthesis length within CAG repeat-containing substrates by POLβ in vitro, with protein stoichiometries consistent with those found in the striatum leading to increased synthesis and stoichiometries like those found in the cerebellum causing reduced nucleotide incorporation (Figure 6). Altogether, the findings suggest that poor cooperation between POLβ and FEN1, as apparently present in the striatum of R6/1 animals, would result in incorporation of long stretches of nucleotides by POLβ strand displacement activity and the formation of long 5′-flap structures that are generally resistant to FEN1 cleavage activity and precursors to somatic expansion (see Figure 7).
In vitro studies have shown that FEN1 is also stimulated by other BER partners, including APE1 [44]–[46], and by cofactors such as HMGB1 [28], which are both higher in the cerebellum when compared to the striatum (Figure 4). Thus, low APE1 and HMGB1 levels in the striatum might further contribute to the intrinsically low levels of flap endonuclease activity, thereby reducing the probability that structured 5′-flaps at CAG/CTG repeats are processed by FEN1. Alternatively, studies have shown that FEN1 possesses, in addition to its flap endonuclease activity, exonuclease and gap-dependent endonuclease activities, which can help process unusual 5′-flap structures, such as those generated by triplet repeat sequences during maturation of Okazaki fragments [47]. Interestingly, when assaying 5′-flap endonuclease activity (Figure 4C), we identified an alternative product of shorter size, present only in the cerebellum of 36 week-old R6/1 mice, suggesting that additional exonuclease or endonuclease activities exist in the cerebellum, but not in the striatum of old transgenic animals. These activities might help cerebellar neurons to remove the structured 5′-flaps formed at CAG repeats. Conversely, the low intrinsic endonuclease and exonuclease activities of FEN1 (or FEN1-related enzymes) in the striatum would lead to the persistence of structured flaps, thereby resulting in the generation of pre-expanded alleles, which could be processed through an error prone mode to facilitate somatic expansion as shown by Panigrahi et al. [29] (see Figure 7).
In conclusion, our results agree with a model that incorporates a role for oxidative DNA damage and BER in somatic CAG instability, at least in the context of HD. Our results suggest that, in the cerebellum, optimal cooperation between gap filling by POLβ and 5′-flap excision by FEN1 during the repair of oxidative lesions via LP-BER prevents (or at least limits) formation of slipped strand structures at CAG repeats. Conversely, in the striatum, poor cooperation between these two enzymes likely leads to the formation of complex intermediate structures, which, as shown by Panigrahi et al. [29], can then be processed through an error prone mechanism to foster somatic expansion.
Note. While this manuscript was under revision, two papers came out from Liu et al. [48] and Lopez Castel et al. [49] that support a contribution of poor coordination between specific enzymatic steps during DNA damage repair to somatic expansion of CAG/CTG repeats. The study of Liu et al. [48], based on in vitro experiments, indicates that dysfunctional coordination between POLβ and FEN1 during LP-BER triggers somatic expansion of substrates containing CAG repeats. Our study nicely complements these data by proposing that the lack of coordination between the two corresponding enzymatic steps of BER is correlated with the stoichiometric ratios of POLβ and FEN1 and is tissue-dependent. The study of Lopez Castel et al. [49], which is based on the use of mammalian cell lines impaired for Ligase I, extends this view by suggesting that coordination with the downstream ligation step is also crucial.
Hemizygous R6/1 and R6/2 (160 CAG) and R6/2 (100 CAG) mice from the Jackson Laboratory were maintained on a mixed CBAxC57BL/6 genetic background [21]. HdhQ111, HdhQ50 and HdhQ20 heterozygotes were maintained on a CD1 outbred genetic background [23]. The experiments were approved by the ethical committee C.R.E.M.E.A.S (Comite Regional d'Ethique en Matiere d'Experimentation Animale de Strasbourg).
The isolation of genomic DNA from mouse striatum and cerebellum for analysis of oxidative DNA damage was performed under conditions that minimize ex vivo oxidation artifacts according to the protocol developed by Lu et al. [22]. To this end, the silica-gel-membrane based DNeasy Tissue Kit (Qiagen) was used. Importantly, 50 µM of the free radical spin trap phenyl-tert-butyl nitrone (PBN, Sigma) was included in all buffers. High temperature and phenol use were avoided. DNA extracts were treated with RNase I. Striatum and cerebellum DNA extracted under these conditions was also used for CAG repeat sizing. Tail DNA for CAG repeats sizing was isolated using a standard protocol. Total RNA for CAG sizing and quantitative RT-PCR analysis was prepared using the RNeasy Mini Kit (Qiagen).
CAG repeat size was determined by PCR amplification using the HEX labeled primer 31329 and primer 33934 previously described [20]. PCR reactions were performed using the expand high fidelity DNA polymerase (Roche), according to manufacturer's instructions. PCR reaction products were subsequently purified using the Nucleospin Extraction II Kit (Macherey-Nagel). Products were then analyzed using the ABI Prism 3100 DNA analyzer instrument and GeneScan and Genotyper softwares. Size calibration was performed by including ROX 500 or ROX 1000 (Applied Biosystems) with the analyzed PCR products. Amplitude of Genescan profile was determined by calculating the number of peaks above 10% of the maximum fluorescent peak intensity. From the amplitude value, we deduced the median peak.
To quantitatively amplify CAG repeats from R6/1 and R6/2 DNA or RNA, we used the following protocol. DNA was amplified with primers 31329 and 33934, previously described [20] using the Herculase Hotstart DNA Polymerase (Stratagene). Concentrations of dNTP and primers were those recommended by the manufacturer. 8% DMSO was included in the reaction, as well as Sybr green (Molecular probe). The PCR reactions were performed and analyzed on a Light Cycler instrument (Roche). PCR cycling conditions were as follows: DNA was first inactivated for 3 min at 98°C, followed by 45 cycles consisting in 40 seconds at 98°C, 30 seconds at 60°C and 2 min at 72°C.
Reverse transcription was performed on 1 µg of total RNA using SuperScriptII (Invitrogen) and random hexamers according to the manufacturer instructions. We performed PCR amplification of cDNA on a Light-Cycler instrument (Roche). PCR primers for detection of Ogg1, Ape1, Polβ, Fen1, Hprt, Gapdh and 36B4 are available upon request.
Detection of AP sites was performed using the DNA damage quantification kit (Biovision), according to the manufacturer's instructions. The method is based on specific reaction of the Aldehyde Reactive (ARP) reagent with the open ring form of AP sites. Briefly, genomic DNA was treated with ARP tagged with biotin residues. AP sites in the DNA were then quantified using an avidin-biotin assay followed by a colorimetric reaction [50].
DNA damage at specific gene loci was assayed by cleavage of genomic DNA with bacterial formamidopyrimidine glycosylase (Fpg). Fpg is an AP-lyase that specifically excises 8-oxoG among other oxidized bases, and then creates a single strand break at the site of the abasic product. Quantitative real time PCR was used to determine the level of intact DNA at CAG repeats before and after DNA cleavage by Fpg. As a control, a portion of the murine Htt gene, which is similar in size and GC content to the CAG-expanded locus, was amplified. The ratio of PCR products after Fpg cleavage to those present in untreated DNA was used to determine the level of intact DNA. We followed the protocol described by Lu et al. [22], with the following modifications: Fpg from Sigma was used and incubated with genomic DNA for 1h at 37°C at a concentration of 0.2 µg Fpg/µg of DNA. These conditions, which were predetermined by an Fpg dose response curve and time course, allowed the reaction to reach a steady state. After inactivation at 60°C for 5 min, the Fpg reaction underwent an ethanol precipitation step. DNA was recovered in water and then submitted for quantitative PCR. Conditions for quantitative amplification of the CAG-expanded locus are described above. Quantitative amplification of the Hdh locus was performed using a commercial PCR master mix (Qiagen) with forward (5′-TCGAGTCGCTCAAGTCGTTT-3′) and reverse (5′-ACTTCGCAAACTGGGAACGG-3′) primers and PCR conditions were as follows: a first step of denaturation was performed at 95°C for 15 min, followed by 45 cycles consisting in a denaturation at 95°c for 30 seconds, hybridation for 30 seconds at 55°C and elongation for 1 min at 72°C.
Striata and cerebella were dissected and homogenized in lysis buffer containing 50 mM Tris-HCl pH 8.0, 10% glycerol, 5 mM EDTA, 150 mM KCl, a cocktail of protease inhibitors (Roche) and 1% NP-40. The extracts were incubated for 15 min on ice and centrifuged for 20 min at 13000 rpm and 4°C. Supernatants were collected and analyzed on SDS-PAGE gels. Rabbit α-OGG1 (Abcam), mouse α-POLβ (Biovision), rabbit α-APE1 (Abcam), rabbit α-FEN1 (Santa Cruz or Abcam), rabbit α-HMGB1 (Abcam) and mouse α-β-tubulin (Chemicon) were used at 1∶1000 dilutions and revealed with appropriate α-rabbit or α-mouse peroxidase-conjugated secondary antibodies (Jackson immunoResearch Laboratories) and the ECL chemiluminescence reaction (Pierce or Millipore).
Hairpin DNA substrates with specific modifications (Figure 3A) were purchased from Midland Certified Reagent Company, Inc. (Midland, TX). [γ32P]ATP 5′ radiolabeled oligonucleotides were generated as described [51]. After annealing, the labeled oligonucleotides were purified from unincorporated [γ32P]ATP by using a Bio-Rad Micro Bio Spin P30 column. Briefly, after spinning the column at 1000 g for 1 min to remove packing buffer, the columns used for purification of the 5′-32P THF substrates were washed twice with OPT buffer (25 mM Mops, pH 7.2, 100 mM KCl, 1 mM MgCl2) and the columns used for 5′-32P 8-oxoG substrates were washed twice with NEB2 buffer (New England Biolabs). Columns were centrifuged as above for each washing step. The labeled oligonucleotides were then eluted through the appropriate column by centrifugation at 1000 g for 4 min. The hairpin structure was verified by incubation of the 5′-32P labeled oligonucleotides (0.2 pmol) with BamHI or EcoRI (0 U to 4U; New England Biolabs) and subsequent electrophoresis on a 15% polyacrylamide urea denaturing gel. OGG1 incision assays were performed by incubating 5′-32P labeled 8-oxoG substrate (0.2 pmol) with hOGG1 protein (0 to 30 ng; New England Biolabs) in NEB2 buffer at 37°C for 15 min. APE1 incision assays were performed by incubating 5′-32P labeled THF substrate (0.2 pmol) with hAPE1 protein (0 to 1 ng) in OPT buffer at 37°C for 5 min. Reactions were inhibited by the addition of stop buffer (95% formamide, 20 mM ethylenediaminetetraacetic acid [EDTA], 0.5% bromophenol blue and 0.5% xylene cyanol), and then heated at 95°C for 5 min. Reaction products were resolved by 15% polyacrylamide urea denaturing gel electrophoresis and imaged using a Typhoon phosphoimager.
Cell extracts were prepared essentially as described [52]. Briefly, frozen striata or cerebella were homogenized in 10 mM HEPES–KOH, pH 7.7, 0.5 mM MgCl2, 10 mM KCl, 1 mM DTT buffer and then centrifuged at 2000 g at 4°C for 10 min. The pellet was resuspended in 20 mM HEPES–KOH, pH 7.7, 0.5 mM MgCl2, 420 mM NaCl, 0.2 mM EDTA, 25% glycerol, PIC (Complete Protease Inhibitor Cocktail EDTA-free, ROCHE), 1 mM DTT and gently stirred at 4°C for 20–30 min to allow for efficient nuclear lysis. The suspension was centrifuged at 14000 g at 4°C for 15 min. The supernatant was dialyzed against 40 mM HEPES–KOH, pH 7.7, 50 mM KCl and 2 mM DTT buffer overnight at 4°C. Cell extract concentrations were determined using a Bradford assay. OGG1, APE1, POLβ and FEN1 assays were performed using oligonucleotide substrates described by [26], see Figure S2. 0.1 pmol of [γ32P]ATP 5′-radiolabeled oligonucleotides were incubated with 10 µg of cell extract and assay buffer (75 mM KCl, 25 mM MgCl2, 7,5 mM dNTPs, 3,125 mM HEPES–KOH, pH 7.7, 1% glycerol, 0.25 µM EDTA) at 37°C. Incubation times were optimized for each repair activity: 3 h, 1 h, 2 h and 50 min were used for 8-oxoG incision (OGG1), AP-endonuclease (APE1), gap-filling (POLβ) and flap endonuclease (FEN1) assays, respectively. The samples were then treated with 0.5% SDS and 0.8 mg/mL of Proteinase K, heated at 55°C for 15 min, purified by phenol-chloroform extraction and resuspended in solution by adding equal volume of 98% formamide, 10 mM EDTA, bromophenol blue and xylene cyanol buffer. Reaction products were resolved by 20% polyacrylamide urea denaturing gel electrophoresis and imaged on radiographic films. Images were captured with GeneSnap and quantified with GeneTool softwares on Syngene Chemigenus XE machine.
2 nM of purified human DNA POLβ [53] was incubated with increasing concentrations of purified human FEN1 [54] (1 nM, 2 nM, 4 nM, 8 nM and 16 nM) and 200 fmol of substrate for 30 min at 37°C in 50 mM HEPES-KOH pH 7.5, 10 mM MgCl2, 0.5 mM EDTA, 2 mM DTT, 2 mM ATP, and 20 µM each dATP, dCTP, dGTP, and dTTP. Reactions were stopped by the addition of stop buffer consisting of 90% formamide and 20 mM EDTA, boiled for 5 min and then loaded onto a 15% polyacrylamide-urea gel and run at 300 V for 5 hrs.
ChIP experiments were performed essentially as described [55]. For each ChIP experiment, striata and cerebellar from 2 to 4 transgenic mice were pooled, cut into small fragments, fixed by adding 37% formaldehyde to a final concentration of 1% and incubated for 10 min at room temperature. Cross-linking was stopped by addition of glycine to 0.125 M. Tissue fragments were washed three times with cold phosphate-buffered saline and treated with sonication buffer (50 mM HEPES pH 7.9, 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% sodium dodecyl sulfate, 0.1% Na-deoxycholate) containing protease and phosphatase inhibitors. Tissue was then homogenized, and lysates were sonicated to obtain DNA fragments of 200 to 1000 bp, as revealed by ethidium bromide staining of aliquots separated on agarose gels (Figure S4). Samples were centrifuged to pellet debris and an aliquot was taken for gel analysis and input. The soluble chromatin fraction was pretreated for 1 h at 4°C with protein A Agarose/Salmon Sperm DNA -50% slurry- (Millipore). Samples were then incubated overnight at 4°C with α-DNA POLβ ab194 (ChIP grade, Abcam) or α-AcH3K9/14 (Upstate) antibodies. Protein A Agarose/Salmon Sperm DNA was then added, and the mixture was incubated for 2 h at 4°C. Agarose beads were washed twice for 10 min with sonication buffer, twice for 10 min with wash buffer A (sonication buffer with 500 mM NaCl), twice for 10 min with wash buffer B (20 mM Tris-HCl, pH 8.0, 1 mM EDTA, à.25 M LiCl, à.5% NP-40, 0.5% Na-deoxycholate), and finally with Tris-EDTA (TE, pH 8.0). Immune complexes were eluted from the beads with 1% SDS in TE (pH 8.0) and protein-DNA cross-links were reversed by adding 200 mM NaCl and heating overnight at 65°C. After treatment with proteinase K for 2 h at 42°C, the samples were purified by phenol-chloroform-isoamyl alcohol extraction and precipitated with ethanol. One-sixth (for amplification of CAG expansions) to one-fifteen (for amplification of a fragment of the Hdh gene) of the immunoprecipitated DNA and 1% of the input DNA were quantified by real-time quantitative PCR (see above). Results are expressed relative to the amount of input DNA per ChIP.
|
10.1371/journal.pntd.0004162 | Complete Protection against Pneumonic and Bubonic Plague after a Single Oral Vaccination | No efficient vaccine against plague is currently available. We previously showed that a genetically attenuated Yersinia pseudotuberculosis producing the Yersinia pestis F1 antigen was an efficient live oral vaccine against pneumonic plague. This candidate vaccine however failed to confer full protection against bubonic plague and did not produce F1 stably.
The caf operon encoding F1 was inserted into the chromosome of a genetically attenuated Y. pseudotuberculosis, yielding the VTnF1 strain, which stably produced the F1 capsule. Given orally to mice, VTnF1 persisted two weeks in the mouse gut and induced a high humoral response targeting both F1 and other Y. pestis antigens. The strong cellular response elicited was directed mostly against targets other than F1, but also against F1. It involved cells with a Th1—Th17 effector profile, producing IFNγ, IL-17, and IL-10. A single oral dose (108 CFU) of VTnF1 conferred 100% protection against pneumonic plague using a high-dose challenge (3,300 LD50) caused by the fully virulent Y. pestis CO92. Moreover, vaccination protected 100% of mice from bubonic plague caused by a challenge with 100 LD50 Y. pestis and 93% against a high-dose infection (10,000 LD50). Protection involved fast-acting mechanisms controlling Y. pestis spread out of the injection site, and the protection provided was long-lasting, with 93% and 50% of mice surviving bubonic and pneumonic plague respectively, six months after vaccination. Vaccinated mice also survived bubonic and pneumonic plague caused by a high-dose of non-encapsulated (F1-) Y. pestis.
VTnF1 is an easy-to-produce, genetically stable plague vaccine candidate, providing a highly efficient and long-lasting protection against both bubonic and pneumonic plague caused by wild type or un-encapsulated (F1-negative) Y. pestis. To our knowledge, VTnF1 is the only plague vaccine ever reported that could provide high and durable protection against the two forms of plague after a single oral administration.
| Yersinia pestis, the agent of plague, is among the deadliest infectious agents affecting humans. Injected in the skin by infected fleas, Y. pestis causes bubonic plague, which occasionally evolves into the very lethal and contagious pneumonic plague. Y. pestis is also a dangerous potential bioweapon but no plague vaccine is available. The current study describes the development of a vaccine highly efficient against plague in both its bubonic and pneumonic forms. The strategy consists of a live, avirulent, genetically modified Yersinia pseudotuberculosis that produces the capsule antigen of Y. pestis, named F1. The goal was to propose a vaccine that would be both easy to produce rapidly in large amounts with high quality, and easy to administer to individuals via a single oral dose. The VTnF1 strain described fulfills these demands. The immune response generated is long-lasting, involving both antibodies and memory cells directed against F1 and other antigens. We conclude that VTnF1 is a very promising candidate vaccine against plague.
| Plague has been one of the deadliest bacterial infections in human history, causing millions of deaths during three major historical pandemics and leaving an indelible mark engraved in human's collective memory. In addition to ancient foci of the disease in Asia and Africa, the last pandemic (plague of modern times), which started one century ago, allowed plague to develop new foci in previously unaffected territories such as Madagascar, Southern Africa, and the Americas. Despite considerable progress in its prevention and cure during the 20th century, plague has recently made a new comeback, causing close to 50,000 human cases during the last twenty years [1], including cases in countries where plague was thought to be extinct [2]. Therefore, plague is categorized by WHO (World Health Organization) as a re-emerging disease [1, 3].
The etiologic agent of plague, Yersinia pestis, is a highly pathogenic Gram-negative bacillus, very recently derived from the much less virulent enteropathogen Yersinia pseudotuberculosis [4]. Transmission of the plague bacillus to humans generally starts with the bite of an infected flea, causing bubonic plague, the most frequent clinical form of the disease. Y. pestis occasionally reaches the airways, and the resulting secondary pneumonic plague is highly contagious due to the emission of infected aerosols, causing inter-human transmission of pneumonic plague. This pneumopathy is systematically lethal in usually less than three days if no treatment is administered.
The possible use of the plague bacillus as a bioterrorist weapon is also a serious threat due to its pathogenicity and human-to-human transmission. Y. pestis has been classified by the Centers for Disease Control (CDC) of the USA among Tier 1 select biological agents. Different strains of Y. pestis showing resistance to antibiotics currently used to treat patients have been identified in Madagascar [5]. Antibiotherapy can therefore no longer be considered as sufficient against the natural and intentional danger of plague. Facing such a public health risk, vaccines may be one of the only remaining alternatives to limit the death toll in humans. A plague vaccine should confer protection against bubonic plague, the most frequent form of the disease in nature [1], at the origin of pneumonic plague outbreaks. The vaccine should also protect against pneumonic plague, the most contagious and fatal form of the disease.
No plague vaccine is currently licensed. The live attenuated Y. pestis strain EV76 and its derivatives have previously been used in humans [6, 7], and were found to confer protection. However, the genetic instability of Y. pestis represents a major obstacle in its use as live vaccine [4, 8]. Several molecular vaccine candidates have been recently developed, among which two molecular vaccines (RypVaxtm and rF1Vtm) are the most advanced in clinical trials [9, 10]. These vaccines rely on a combination of two peptides: the F1 antigen composing the Y. pestis capsule and the LcrV component of the Type Three Secretion System (TTSS) [9, 10], which are efficient targets of protective immunity against plague [6, 11]. Molecular vaccines are generally adjuvanted with alum, and thus are good inducers of antibody production but poor inducers of cellular immune response [12, 13]. Cellular immunity is, however, important for plague protection [14], and a weak cellular response could explain why F1-V vaccinated African Green Monkeys were poorly protected despite adequate antibody titers [15].
We recently proposed a vaccine strategy against plague based on an oral vaccination with a live, attenuated strain of Y. pseudotuberculosis [16, 17]. Because this species is a recent ancestor of Y. pestis, the two species are genetically almost identical, whereas Y. pseudotuberculosis has much lower pathogenicity and much higher genomic stability [4]. Due to their immunogenicity and antigenic complexity, live vaccines generate both humoral and cell-mediated immune responses without addition of adjuvant, and the response is directed against multiple target antigens, thus inducing an immunological response that could not be circumvented by genetic engineering of Y. pestis. In addition, a live vaccine, once developed and validated, could easily and rapidly enter mass production, and would be well suited in response to an emergency need. As a proof of concept, we had reported that oral inoculation of two doses of a live, naturally attenuated Y. pseudotuberculosis could provide 88% protection against bubonic plague [16].
The initial Y. pseudotuberculosis strain that we tested was not genetically defined [16]. To develop a vaccine strain both avirulent and genetically defined, the virulent Y. pseudotuberculosis IP32953 strain, whose genome is known [4], was irreversibly attenuated by deletion of genes encoding three essential virulence factors (the High pathogenicity island, YopK, and the pH6 antigen (PsaA [17]). To increase vaccine efficiency, an F1-encapsulated derivative was constructed. This was obtained by cloning the Y. pestis F1-encoding caf operon into a plasmid, and introducing this plasmid into the attenuated V674 Y. pseudotuberculosis, thus producing the V674pF1 strain. Oral vaccination using 108 CFU protected 100% against pneumonic plague caused by a challenge with 33 LD50 Y. pestis and 80% against a high-dose challenge (3,300 LD50) [17]. However, we subsequently found that vaccination with strain V674pF1 protected only 81% of the mice against bubonic plague caused by subcutaneous injection of a moderate dose (100 LD50) of Y. pestis, a level of protection that was judged insufficient. We also observed that production of F1 was not stable and homogenous in the V674pF1 vaccine strain, possibly accounting for the incomplete protection conferred.
The aim of this study was to generate a new vaccine strain that would allow a homogenous production of F1, and to evaluate the immune response elicited and its protective performances against both bubonic and pneumonic plague.
Animals were housed in the Institut Pasteur animal facility accredited by the French Ministry of Agriculture to perform experiments on live mice (accreditation B 75 15–01, issued on May 22, 2008), in compliance with French and European regulations on the care and protection of laboratory animals (EC Directive 86/609, French Law 2001–486 issued on June 6, 2001). The research protocol was approved by the French Ministry of Research (N° CETEA 2013–0038) and was performed in compliance with the NIH Animal Welfare Insurance (#A5476-01 issued on 02/07/2007).
The Y. pseudotuberculosis and Y. pestis isolates used in this study and their derivatives are described in S1 Table [17]. Bacteria were grown at 28°C in Luria-Bertani (LB) broth or on LB agar plates supplemented with 0.002% (w/v) hemin (LBH). Bacterial concentrations were evaluated by spectrometry at 600 nm and plating on LBH or LB plates. Chloramphenicol (Cm, 25 μg/ml), ampicillin (Amp, 100 μg/ml), kanamycin (Km, 30 μg/ml), spectinomycin (Spec, 50 μg/ml), or irgasan (0.1 μg/ml) were added to the media when necessary. All experiments involving Y. pestis strains were performed in a BSL3 laboratory.
To introduce the caf operon into the Y. pseudotuberculosis chromosome, the Tn7 transposition tool was used [18]. First, a mini-Tn7 containing a Cm resistance cassette was constructed by cloning the Cm-FRT KpnI-fragment from pFCM1 into pUCR6Kmini-Tn7 digested with KpnI. The resulting plasmid, named pUCR6K-miniTn7-Cm-FRT, was then digested by ApaI and EcoRI and ligated to the 5 kb-PCR fragment containing the entire caf operon from Y. pestis amplified with primers A (5’-ATAAGAATGAATTCGTGACTGATCAATATGTTGG-3’) and B (5’-CGTTAGGGCCCGTCAGTCTTGCTATCAATGC-3’), which added ApaI and EcoRI sites at the extremities of the fragment. To insert the caf locus into the Y. pseudotuberculosis V674 chromosome, plasmids pUC18R6KTn7-caf-CmR and pTNS2 (transposase provider) were introduced together into V674 by electroporation. Transposants were selected on LB agar plates containing chloramphenicol and verified for their sensitivity to ampicillin. Presence of the transposon Tn7-caf-CmR at the chromosomal att-Tn7 site was verified by PCR, using two pairs of primers: A (5’-CACAGCATAACTGGACTGATTTC-3’) and B (5’- GCTATACGTGTTTGCTGATCAAGATG-3’) for the left junction, and C (5’-ATTAGCTTACGACGCTACACCC-3’) and D (5’- ACGCCACCGGAAGAACCGATACCT-3’) for the right junction.
The recombinant Y. pseudotuberculosis strain containing the Tn7-caf-CmR region in its chromosome was initially named V674TnF1, and its use as a vaccine against plague is protected by patent application PCT/IB2012/001609, issued on August 7, 2012. For simplicity, we hereafter refer to this strain as "VTnF1".
To analyze F1 capsule production, bacteria were visualized by optical microscopy in contact with India ink, and an ELISA assay quantifying the F1 capsule on bacteria was performed as previously described [17].
A mini-Tn7 transposon containing a Km resistance cassette was constructed by cloning the Km-FRT SacI-fragment from pFKM1 into pUCR6Kmini-Tn7 digested with SacI. The recombinant pUCR6KTn7-Km plasmid was then digested with ApaI and XmaI, and ligated to the 5878 bp ApaI/XmaI fragment from pGEN-lux (LuxCDABE provider;[19]). The resulting pUCR6KTn7-luxCDABE plasmid was then digested with SpeI and XmaI and ligated to the 109 bp promoter region of ail (YPO2905), obtained from the Y. pestis DNA template by PCR with primers E (5'- CGCACTAGTTGGAATACTGTACGAATATCC-3') and F (5'- ataCCCGGGccagattgttataacaatacc). The resulting plasmid was named pUCR6KTn7-Pail-lux. For transposition of Tn7-Pail-lux into the Y. pestis chromosome, plasmids pUCR6KTn7-Pail-lux and pTNS2 were introduced into CO92 bacterial cells by electroporation. Transposants were selected for with LBH agar plates containing Km. Verification of the CO92::Tn7-Pail-lux recombinant was performed by PCR using the primer pairs A/B and C/D for chromosomal integration, and by measurement of photon emission using a Xenius plate reader (SAFAS Monaco) for bioluminescence activity. The virulence of the recombinant CO92::Tn7-Pail-lux derivative upon s.c. injection was checked and was found to be similar to that of the wild-type strain, with a median lethal dose (LD50) of 10 CFU. In vivo imaging was performed with an In Vivo Imaging System (IVIS 100, Caliper Life Sciences).
Mouse vaccinations were performed in a BSL3 animal facility as described previously [17]. Animals were seven-week-old OF1 female mice or (when specified) C57BL/6 mice from Charles River France. Bacterial suspensions (200 μl in saline) were given intragastrically (i.g.) to mice using a curved feeding needle. Animals were monitored for suffering (prostration, ruffled hair) every other day and were weighed to estimate the impact of vaccination. The virulence of the VTnF1 strain by the oral route was tested by infecting i.g. groups of mice (four per dose) with increasing doses of bacteria in order to determine the LD50. The LD50 value was calculated by the Spearman-Karber method [20]. In vivo dissemination of the VTnF1 strain was examined as described previously [17]. Feces (two fecal pellets) were collected from live mice and were homogenized in PBS using disposable homogenizers (Piston Pellet from Kimble Chase, Fisher Sci.). Peyer's patches, spleen, liver and mesenteric lymph nodes were collected aseptically from euthanized mice. They were homogenized in sterile PBS using three mm glass beads and an electric mill (TissueLyser, Qiagen). The bacterial load was determined by plating serial dilutions of the homogenates.
Mice were challenged either four weeks or six months after vaccination. Y. pestis strains were grown at 28°C on LBH plates, and suspensions in saline were prepared for infections. Mice were infected by s.c. injection (100 μl) in the ventral skin on the linea alba. The LD50 of the CO92Δcaf strain by this route was determined by infecting mice (six per dose) with serial dilutions of bacterial suspensions, and was found to be 100 CFU. To induce pneumonic plague, anesthetized mice were infected i.n. as previously described [17] by instillating 10 μl of bacterial suspensions in nostrils (5 μl each). Animal survival was followed for 21 days.
Mouse blood was collected three weeks after vaccination from live animals by puncture of the maxillary artery with a Goldenrod lancet (Medipoint, USA). Microtiter plates (NUNC) were coated either with F1 antigen, or a sonicate of the Y. pestis CO92Δcaf strain (10 μg/ml). The F1 antigen was obtained from Y. pestis as described previously [21]. The sonicate of the Y. pestis CO92Δcaf strain was obtained by sonication of bacteria grown at 37°C on LB agar as previously described [17]. The Yops antigens were purified as previously described [22], and assays were performed as described before [17]. Briefly, plates were blocked with 5% defatted dry milk and 0.1% Tween 20 in PBS. Sera serially diluted in PBS containing 0.1% BSA were incubated in wells and bound antibodies were detected using horseradish peroxydase (HRPO)–coupled rat antibodies specific for mouse IgG (Bio-Rad). HRPO activity was revealed using TMB substrate (OptiEIA, BD Pharmingen). Antibody (Ab) titers were calculated as the reciprocal of the lowest sample dilution giving a signal equal to two times the background. To analyze immunoglobulin isotypes, horseradish peroxidase (HRP)-coupled probes directed against mouse IgG1, IgG2a, IgG2b and IgM (Caltag) were used. IgG3 were detected using an uncoupled goat antibody (Abcam), and revealed with an HRP-coupled rabbit antibody against goat IgG (BioRad). For Western blotting, Y. pestis CO92, wild-type and Δcaf, S1 Table) were boiled in Laemmli sample buffer (Thermo) and were loaded on 12% acrylamide gel for SDS-PAGE separation using a MiniProtean device (BioRad). Migrated material was transferred from the gel onto a PVDF membrane (Amersham). All subsequent steps were performed following the western immunoblotting protocol recommended by Cell Signaling Technology (USA). Membrane strips were incubated overnight in pooled 1/100 diluted sera from naïve mice, or mice vaccinated one month earlier. Bound IgG were revealed using a secondary goat anti-mouse IgG coupled to horseradish peroxidase (BioRad). ECL Plus kit (Pierce) was used for peroxidase revelation, and membranes were photographed using a ChemiDoc apparatus (BioRad).
To evaluate the cellular memory in vaccinated animals, splenocytes were cultured as described previously [17]. Briefly, spleens from euthanized animals were dissociated and erythrocytes were lyzed using Gey’s hemolytic solution [23]. Cells extensively washed with cold PBS were resuspended in RPMI 1640 + Glutamax (Invitrogen) supplemented with 5% fetal bovine serum, penicillin + streptomycin, and 10 mM ß-mercaptoethanol. Cells (5x106/condition) were stimulated with either a sterile Y. pestis CO92Δcaf sonicate (5 μg/ml), sterile F1 antigen (5 μg/ml), or Concanavalin A (1 μg/ml; Sigma) as a positive control. After three days, the supernatant was collected and the cytokine content was determined using IFNγ, IL-1β, IL-10, and IL-17 assays (Duosets, R&D Systems).
The Log-rank (Mantel-Cox) test was used to compare survival curves. Unpaired Mann-Whitney or Student's t tests were used to compare bacteria numbers, animal weights, antibody titers, and cytokine production. The paired Mann-Whitney test was used for bioluminescence results. Analyses were performed with Prism 6.0 software (GraphPad Software).
Strain VTnF1 was constructed by inserting the caf operon encoding F1 [17] into the chromosome of the attenuated Y. pseudotuberculosis V674, using mini-Tn7 transposon technology [18]. F1 capsule production by recombinant VTnF1 grown at 37°C in LB broth was tested using an F1-specific rapid dipstick test [21], which was clearly F1 positive (S1 Fig). Microscopic visualization of VTnF1 in India ink revealed that all VTnF1 bacterial cells produced the F1 capsule, as visualized by the repulsion of ink particles with comparable thickness (Fig 1A). This contrasted with V674pF1 cultures, which displayed encapsulated and non-encapsulated bacteria. To quantify F1 capsule production, isolated colonies obtained after growth in LB broth at 37°C were tested using an F1-specific ELISA. All VTnF1 colonies were F1 positive and their F1 levels were homogenous (Fig 1B), whereas V674pF1 colonies exhibited various levels of F1 at their surface, indicating both heterogeneity and instability of F1 production. VTnF1 produced as much F1 as Y. pestis, and this production was temperature-dependent (Fig 1C).
Stability of F1 production was additionally tested after VTnF1 growth in vivo in mice. To this aim, VTnF1 was injected i.g. to animals, and five days later, bacteria were recovered from Peyer’s patches. All VTnF1 colonies were positive for F1 by ELISA (Fig 1B). F1 production was homogenous and comparable to F1 levels of colonies obtained after in vitro culture. This demonstrates that VTnF1 stably and homogenously produced F1 after growing in vivo in mice.
The V674 strain used to construct VTnF1 was strongly attenuated after intragastric inoculation into mice (LD50 >1010 CFU;[17]). VTnF1 exhibited an attenuation of virulence similar to V674 and the previous vaccine V674pF1, with an LD50 >1010 CFU. Mice vaccinated with VTnF1 at a dose of 108 CFU presented transient signs of infection (ruffled hair), and a slight delay in weight gain from day three to seven post-vaccination (average 1.7 g), but they recovered a normal weight from day 10 (S2 Fig). The F1 capsule is dispensable for Y. pestis virulence in many mouse strains [24–27], but not in others such as C57BL/6 [28]. When a high dose of VTnF1 (4x109 CFU) was inoculated orally to C57BL/6 mice (N = 7), no lethality was observed during the three weeks of follow-up. This confirmed that VTnF1 was strongly attenuated, even for C57BL/6 mice that are more susceptible to F1 action.
After oral inoculation of VTnF1 (108 CFU), the bacteria were detected on day five-six in the feces and Peyer's patches of all mice (Fig 2A and 2B), and in the spleen, liver and mesenteric lymph nodes (MLN) of most animals (Fig 2C and 2E). However, bacteria were almost completely cleared from the spleen, Peyer's patches and MLN after 15 days, and from the liver after 26 days. Feces tested monthly during the following 5 months remained negative, indicating a vaccine clearance from visceral organs and the gut lumen.
The humoral immune response elicited by vaccination with VTnF1 (108 CFU orally) was first evaluated by quantifying serum antibodies against purified F1 at different times post-vaccination over a period of six months. Anti-F1 IgG were detectable as early as four days after vaccination, and reached plateau values after 7 days (Fig 3A). They then maintained at high levels without significant evolution, as shown by the fact that titers observed after six months (d180) were not significantly different from those at d30 (p = 0.08, 14 mice per group). Analysis of the immunoglobulin isotypes revealed that both IgG1, IgG2a, IgG2b and IgG3 contributed to this humoral response, whereas IgM peaked rapidly after vaccination and then fell to low levels (Fig 3B).
IgG recognizing Y. pestis antigens other than F1 were quantified by ELISA using a sonicate of Y. pestis CO92Δcaf as target. All vaccinated mice had high amounts of IgG against Y. pestis CO92Δcaf antigens (Fig 3C). Western blotting analysis (Fig 3D) confirmed that in addition to F1, at least 12 target antigens were recognized by immune sera. Because conformational epitopes were lost due to the denaturing conditions used for electrophoresis, actual targets were probably more numerous. Among antigens strongly recognized are two components of the Caf operon (absent from the CO92Δcaf strain, Fig 3D): the Caf1 and Caf1M antigens (MW 15.6 and 28.7 kDa respectively). Finally, IgG against purified Yops were also measured because these molecules are essential Y. pestis virulence factors. Such IgG were observed in almost all mice, but with varying levels (Fig 3E). Altogether, our results indicate that the humoral immune response developed rapidly after a single-dose vaccination and involved all main IgG isotypes. The antibodies recognized several antigens other than F1 and were maintained at high levels for extended periods of time without recall vaccination.
To evaluate the antigen-specific T cell memory elicited by VTnF1, splenocytes taken from vaccinated animals were re-stimulated in vitro with either purified F1 antigen, or with a sonicate of the un-encapsulated Y. pestis CO92Δcaf. Cells from mice vaccinated one month earlier with VTnF1 produced IFNγ, IL-17, and IL-10 in response to F1 (Fig 4). Although these levels were low, they were significantly higher than those of control mice (Fig 4), indicating that vaccination mobilized F1-specific memory T cells producing both pro- (IFNγ, IL-17) and anti-inflammatory (IL-10) cytokines. IL-1β production was also measured, but levels were very low (<0.02 ng/ml) in all conditions.
In contrast, cells from VTnF1 vaccinated mice produced high amounts of IFNγ and IL-17 in response to non-F1 Y. pestis antigens, reaching levels at least 10 times higher than those observed for cells from naive mice (Fig 4A and 4B). Strikingly, levels of IL-17 were comparable to those induced by the mitogenic lectin ConA, used as a positive control, indicating that a large proportion of responding splenocytes of vaccinated animals recognized Yersinia antigens and displayed potent pro-inflammatory functions. This cellular response was also much higher than that stimulated by the F1 antigen alone, reflecting the mobilization of T cells directed against multiple Y. pestis antigenic targets.
Y. pestis antigens induced production of IL-10 by splenocytes from both naive and vaccinated mice, probably due to the response of innate immunity cells such as macrophages. However, splenocytes from vaccinated mice produced significantly higher levels of IL-10 than cells from naive mice stimulated with both F1 and other antigens, revealing the recall response of Y. pestis-specific memory cells with anti-inflammatory activity (Fig 4C).
To evaluate the durability of the cell-mediated response, splenocytes from mice vaccinated six months earlier were also tested. Levels of IFNγ, IL-17 and IL-10 in response to F1 and non-F1 antigens were lower on day 180 than on day 30 post-vaccination, but this difference was not statistically significant. Thus, although slightly reduced, the cell-mediated memory persisted after six months.
To determine the protective efficacy of VTnF1 against pneumonic plague, vaccinated mice were challenged intranasally with the fully virulent Y. pestis CO92 strain four weeks after a single oral vaccination (108 CFU of VTnF1). 100% of mice challenged with 105 CFU (33 LD50) of Y. pestis CO92 survived (Fig 5A). Vaccination also protected 100% of the animals exposed to an extremely severe challenge with 107 CFU CO92 (i.e. 3,300 LD50; Fig 5A), whereas the previous V674pF1 vaccine strain protected only 80% of the mice infected with this dose [17]. VTnF1 thus appears more protective.
To evaluate the protective efficacy of VTnF1 against bubonic plague, vaccinated mice were infected s.c. with Y. pestis CO92, four weeks after vaccination. A single oral dose (108 CFU) of the VTnF1 vaccine protected 100% of the mice against 103 CFU (100 LD50) of CO92 (Fig 5B). Compared to V674pF1 which only protected 81% (13/16) of animals against bubonic plague in these conditions, VTnF1 was again more protective. When animals vaccinated with VTnF1 received a very severe challenge by s.c. injection of 105 CFU CO92 (10000 LD50), 93% of mice (13/14) were still protected.
Because the immune memory is known to decline with time, the protection conferred by VTnF1 was evaluated six months (one third of an OF1 mouse's lifespan [29] after a single-dose oral vaccination with VTnF1. VTnF1 was completely undetectable in mice's feces at day 30 post vaccination and the following months. Upon s.c. challenge with 100 LD50 of Y. pestis CO92, 93% of the animals were still protected against bubonic plague. Furthermore, 50% of the mice survived an i.n. challenge with 33 LD50 of Y. pestis CO92 six months after vaccination (Fig 5D). This indicates that the immune memory elicited by the vaccine persisted and provided a long-lasting protection.
The capacity of VTnF1 vaccination to confer protection against bubonic plague caused by a non-encapsulated Y. pestis was also evaluated by challenging vaccinated mice with Y. pestis CO92Δcaf. A single oral dose of VTnF1 conferred 100% protection against a severe intranasal challenge with 107 CFU of Y. pestis (107 CFU, i.e. 3,300 LD50, Fig 5C). The same vaccination protected 93% of vaccinated mice against a severe s.c. challenge with 105 CFU (104 LD50; Fig 5D). Thus, VTnF1 conferred a strong protection against Y. pestis in the absence of the F1 pseudocapsule, indicating that, even if plague was caused by a natural or genetically modified F1-negative Y. pestis, vaccination with VTnF1 would provide a high level of protection.
To determine the stage of the infectious process at which immunity provided by VTnF1 controls Y. pestis proliferation, infection by a bioluminescent Y. pestis strain (CO92 Tn7ail-lux) was followed in vivo in live mice. The strain carries in its chromosome the lux operon under the control of the ail promoter, known to be very active during bubonic plague [30]. In unvaccinated animals, Y. pestis multiplied at the site of injection and spread to other organs, causing the death of two mice out of five at 92h post-injection. (Fig 6A). In VTnF1-vaccinated mice, the bioluminescence signal was much lower after 20 hours at the site of injection (Fig 6B) and after 44 hours it was no longer visible. Dissemination outside of the injection site was not observed in vaccinated mice. Therefore, vaccination induced fast-acting immune mechanisms that prevented the dissemination of Y. pestis from the site of injection.
To meet the demand for a vaccine able to confer protective immunity against both bubonic and pneumonic plague, we previously constructed the genetically modified Y. pseudotuberculosis V674pF1 candidate plague vaccine, which produces the Y. pestis F1 antigen [17]. Although this vaccine protected against inhalational exposure to Y. pestis, protection against bubonic plague was not complete. This lack of full protection was potentially explained by the observation that production of the F1 antigen was unstable. The objective of the present work was therefore to improve the vaccine by generating a new strain with improved robustness and efficiency.
Loss of F1 production by V674pF1 mainly resulted from plasmid instability. Others who used Salmonella as receiver of the caf operon had to apply a sustained antibiotic pressure in vitro to ensure plasmid persistence, but this pressure could not be maintained in vivo [31]. Here, the caf operon was transposed into the chromosome [18], and we show that production of F1 capsule by VTnF1 was comparable to that of Y. pestis, whereas F1 production by V674pF1 was much more heterogeneous and unstable. The stability of VTnF1 characteristics will allow the large-scale production of the live vaccine according to good manufacturing procedures.
VTnF1 provides high protection against both bubonic and pneumonic plague, and is more efficient than V674pF1 used at the same dose (108 CFU) and tested in the same conditions [17]. We had previously reported that only 80% of the mice vaccinated with V674pF1 survived a high-dose pneumonic plague challenge (107 CFU CO92), or a moderate bubonic plague challenge (103 CFU). In contrast, VTnF1 provided complete protection against commonly used bacterial challenges, and almost complete protection against very high challenges. Because V674pF1 and VTnF1 are both derivatives of the V674 attenuated strain, the only possible explanation for this difference of protection is the more homogenous and sustained production of the F1 pseudocapsule by VTnF1. The F1 antigen can activate macrophages [32], an adjuvant effect favorable to the adaptive immune response. Therefore, the efficiency of VTnF1 may result from a stronger stimulation of macrophages, and possibly also of dendritic cells which belong to the same lineage, thus fostering immunity more efficiently.
The high-level protection observed is especially remarkable as it was obtained with a single oral dose of vaccine. Such a very simple procedure is a key advantage as compared to the repeated injections required by most vaccines to confer a protection extended in time. Difficult to perform in the field, repeated injections are considered by public health authorities as a limitation for mass vaccination.
Mice vaccinated with VTnF1 very rapidly control Y. pestis at the skin entry site. This suggests that easily mobilizable effectors such as antibodies and phagocytes reach the infected tissue. Antibodies might play an essential role in protection conferred by VTnF1, as suggested by the high antibody titers of vaccinated mice. VTnF1 displays much more antigenic diversity than plague molecular vaccines currently under development, which are composed of only F1 and LcrV antigens. The immune response induced by VTnF1 targets various Y. pestis proteins in addition to Caf1 and Caf1M antigens. This target diversity is valuable to protect against bacteria, which can lose target antigens via gene deletions, as observed for the Caf1/F1 antigen [33]. The anti-F1 IgG are known to provide protection by opsonizing Y. pestis, facilitated by F1 abundance and surface localization [25, 26, 34]. Whereas the abundant anti-F1 IgG induced by VTnF1 probably play a central protective role against wild type (F1-encapsulated) Y. pestis [25, 27, 34, 35], the resistance of vaccinated mice to plague caused by an F1-negative Y. pestis strain demonstrates that the antibodies directed against other antigens also contribute to protection.
Y. pestis F1-specific IgG induced by VTnF1 are detectable as early as four days after vaccination, a fast kinetic comparable to that observed after vaccination with soluble, recombinant F1 [36]. This prompt onset of IgG production indicates that VTnF1 rapidly interacts with lymphoid cells, probably in Peyer's patches, mesenteric lymph nodes and spleen where VTnF1 was observed. The fast switch of the humoral response from IgM toward antibodies of isotypes IgG1, two and three, generally of high affinity, indicates a strong T-cell dependent response. The presence of abundant IgG3 also indicates that carbohydrate targets are recognized [37]. This diversity is favorable for opsonization via all Fcγ receptors and suggests the involvement of the various B-cells subtypes to the vaccine-induced response [38]. Because IgG levels escalate during the first week to reach the plateau levels observed during the following six months, their contribution to protection against plague at day 30 could already be available at day seven.
Cellular immunity plays an important role against plague [39, 40] and performs critical protective functions during humoral defense against pneumonic plague [41]. Molecular vaccines adjuvanted with alum are poor inducers of this part of the immune response [12, 13]. In contrast, VTnF1 triggered a strong cell-mediated response without adjuvant. Part of the recall cell response was directed against F1, but the most important part was directed toward non-F1 antigens. Its intensity, with IL-17 comparable to a mitogenic stimulation by Concanavalin A, indicates the engagement of a high percentage of splenocytes that are likely recognizing multiple antigenic targets. Memory cells produced IFNγ, IL-17, and IL-10, composing a mixed Th1-Th17 profile. An IFNγ-dependent type 1 immune response is essential for vaccine-induced protection against plague [39]. IFNγ derived from memory T cells instruct potent innate cell activation, resulting in a fast protective immunity against invading microorganisms [42]. IL-17 producing T lymphocytes (Th17 cells) are essential to cure pneumonic plague [43] due to the essential role played by IL-17 in the induction of antimicrobial peptides and attraction of polymorphonuclear leukocytes [44]. The memory response induced by VTnF1 also involves production of the anti-inflammatory cytokine IL-10, which balances potentially harmful effects of IL-17 [45]. In addition to these direct functions, T cells play an important role in antibody-dependent immunity [40] and thus potentiate the humoral response. Altogether, the immune response induced by VTnF1, by combining humoral and cellular mechanisms, has the characteristics required to efficiently clear Y. pestis.
The sustained humoral and cellular immunity six months after vaccination is unlikely to result from a prolonged stimulation of immunity by live bacteria because VTnF1 was undetectable in feces and organs of most mice one month after vaccination. IgG could be produced by long-lived plasma cells, which differ from so-called memory B cells, and produce abundant IgG without re-stimulation [46]. VTnF1 could have persisted in the gut after one month, for example by durably colonizing the cecum as recently reported [47], however this is unlikely because only virulent strains cause cecum foci, and they yield high levels of cultivable bacteria in feces.
The protective immunity provided by VTnF1 after a six-month period (93% against bubonic plague and 50% against pneumonic plague) is remarkably long-lasting since six months correspond to one third of the mouse life [29], which is comparable to #30 years in human lifespan. Considering that both antibody titers and cellular responsiveness remained high after six months, the almost full protection against bubonic plague may result from either compartment or more likely a synergy of the two [14]. The lower protection against pneumonic plague indicates that lung immunity is the most demanding and requires one or more components of the immune response, that are mandatory for protection but are the first to decline with time. The nature of this component of acquired immunity is yet to determine, but does not seem related to Ig isotype switching or the decline of antibody or IFNγ/IL-17 production. One possible explanation is that aging causes a modification of alveolar macrophages functions, with spontaneous activation and reduced responsiveness to external stimuli, thus contributing to lung susceptibility to infection [48, 49].
Live attenuated Y. pestis plague vaccines (EV76 and subclones) has been previously used with success in humans, but could generate strong side effects [6, 7], and, as all Y. pestis strains, are subject to easy genetic rearrangements due to high numbers of insertion sequences in the genome [4, 8]. In addition to genetic stability, VTnF1 combines the known advantages of replicating vaccines: elicitation of humoral and cell-mediated immune responses, robustness against mutant microorganisms, easiness of mass production and use, limited cost, etc., whilst providing guarantees in terms of attenuation, stability, and efficacy against both bubonic and pneumonic plague. Preparedness plans against bioterrorist attacks imply stockpiling millions of vaccine doses. However, stockpiles have a finite lifespan and thus demand regular production of new doses, a rapidly expensive strategy [50]. Live vaccines can be rapidly produced in mass amounts, and are now viewed as a valuable alternative.
In conclusion, we propose here a vaccine providing high-level protection against both bubonic and pneumonic plague after a single-dose immunization. VTnF1 is an easy-to-produce, genetically stable and irreversibly attenuated vaccine, providing a long-lasting and highly efficient protection against both wild type and un-encapsulated (F1-negative) Y. pestis. To our knowledge, VTnF1 is the only plague vaccine ever reported that could provide high and long lasting protection against both bubonic and pneumonic plague after a single oral administration.
|
10.1371/journal.pgen.1000658 | Broad-Scale Recombination Patterns Underlying Proper Disjunction in Humans | Although recombination is essential to the successful completion of human meiosis, it remains unclear how tightly the process is regulated and over what scale. To assess the nature and stringency of constraints on human recombination, we examined crossover patterns in transmissions to viable, non-trisomic offspring, using dense genotyping data collected in a large set of pedigrees. Our analysis supports a requirement for one chiasma per chromosome rather than per arm to ensure proper disjunction, with additional chiasmata occurring in proportion to physical length. The requirement is not absolute, however, as chromosome 21 seems to be frequently transmitted properly in the absence of a chiasma in females, a finding that raises the possibility of a back-up mechanism aiding in its correct segregation. We also found a set of double crossovers in surprisingly close proximity, as expected from a second pathway that is not subject to crossover interference. These findings point to multiple mechanisms that shape the distribution of crossovers, influencing proper disjunction in humans.
| In humans, as in most sexually reproducing organisms, recombination plays a fundamental role in meiosis, helping to align chromosomes and to ensure their proper segregation. Recombination events are tightly regulated both in terms of their minimum number (the rule of “crossover assurance”) and placement (due to “crossover interference”). Accumulating evidence, however, suggests that recombination patterns are highly variable among humans, raising numerous questions about the nature and stringency of crossover assurance and interference. We took a first step towards answering these questions by examining patterns of recombination in gametes inherited by viable, non-trisomic offspring. We found that the minimum number of crossovers is tightly regulated at the level of a chromosome (rather than chromosome arm), but with a notable exception: in females, chromosome 21 appears to frequently segregate properly in the absence of a crossover. We also found a set of double recombination events in surprisingly close proximity, consistent with a pathway not subject to crossover interference. These findings suggest that there are multiple mechanisms of recombination in human meiosis, which may buffer the effects of inter-individual variation in rates.
| Like most sexually reproducing organisms, humans undergo meiotic recombination. This process plays an important role in evolutionary dynamics, generating new combination of alleles on which natural selection can act ([1] and references therein), and in DNA repair. In humans, recombination is also fundamental to the successful completion of meiosis, helping to align homologous chromosomes and to ensure their proper disjunction [2]. Too little recombination or an abnormal placement of crossovers along the genome often results in aneuploidy, an outcome that leads to fetal loss or to severe developmental disabilities [3]. Thus, errors in the recombination process are clearly highly deleterious.
Nonetheless, there is substantial variation in recombination rates and patterns among humans. Individuals differ in their mean number of crossovers across the genome [4]–[6], in part due to genetic variation [7]. They also differ in the intensity of individual recombination hotspots, and in their use of hotspots genome-wide, variability that is again at least partly heritable [8],[9]. Moreover, this variation has detectable fitness consequences: mothers with a higher mean recombination rate have (slightly) more viable offspring [5],[9]. Together, these findings suggest that human recombination is naturally viewed as a quantitative phenotype subject to selection.
This perspective raises a number of questions about the selective pressures on recombination due to its role in meiosis, notably about their nature and stringency and the extent to which the system is buffered against variation. Answers to these questions have important implications for the relationship between recombination rate variation and the susceptibility to non-disjunction. Such answers are also essential to the study of the evolution of recombination rates and of genome dynamics [10],[11].
Chiasmata, the physical connections between chromatids later processed into crossovers, help to bind the homologs together, thereby aiding in their proper disjunction. In the absence of a chiasma, segregation is thought to be haphazard and to frequently result in aneuploidy, a role for recombination that likely imposes strong selective pressures for at least one chiasma per chromosome, even when the mean number is close to 1 [12].
In most species, however, the total number of chiasmata (in males and females) far exceeds the number of chromosomes (cf. [10]). Foci counts of MLH1, a mismatch repair protein that serves as markers for most (but perhaps not all) human crossovers, indicate that there is a chiasma on each chromosome arm examined [13]. Moreover, across mammals, the number of chromosome arms (i.e., two for metacentric chromosomes) appears to be a better predictor of the mean number of crossovers than is the number of chromosomes [10],[14]. These cytogenetic and evolutionary lines of evidence have led to the suggestion that, with rare exceptions, one chiasma per chromosome arm is required for proper disjunction ([15] and references therein; see also [10],[13],[14],[16],[17]).
The role of recombination in human meiosis shapes the placement of crossovers as well as their number. When more than one crossing-over event occurs on the same chromosome, the events are not spaced randomly, but instead tend to occur farther apart than expected by chance. The more even spacing of events on chromosomes due to this “crossover interference” may serve to further reduce the risk of non-disjunction [18]. While several models of interference have been proposed, the phenomenon remains poorly understood. In particular, evidence from model organisms, including mice, suggests that a small subset of events result from a second pathway, not subject to crossover interference [16]. A central prediction of the two-pathway hypothesis is the presence of rare, double recombinants in close proximity (see [19] for details).
Here, we used human pedigree data to assess whether the number of chiasmata is tightly regulated at the level of chromosomes or chromosome arms and to examine how often proper disjunction occurs in the absence of a chiasma. We also took advantage of the high spatial resolution of the data to investigate the spacing of crossover events.
Our point of departure was the number of crossovers observed in transmissions to viable, non-trisomic offspring (Supplementary Table 1 in Text S1). We constructed this distribution from 576 meioses in a large European-American pedigree (see Methods). The families were genotyped at approximately 400,000 single nucleotide polymorphisms (SNPs), providing excellent coverage of the genome (see Supplementary Table 2 in Text S1) and allowing us to localize crossover locations with high resolution [9]. For chromosome 21, we also considered the number of crossovers in an additional 152 female transmissions; these were inferred in European-American families that had been typed for 133 SNPs [20]. The distributions do not differ significantly between the two data sets (p = 0.86 by a Fisher's Exact Test).
The crossovers observed in gametes cannot be equated to the number of chiasmata in tetrads since, for every chiasma, only two of the four chromatids are recombinants. (The term chiasma is sometimes reserved for the physical manifestation of a crossover detected in cytogenetic studies; here, we employ it to denote a crossover in the tetrad, and use the term crossover to refer specifically to genetic exchanges visible in transmitted gametes.) We used the observed distribution of the number of crossovers in gametes to infer the distribution of chiasmata in the tetrads (also called bivalents) [21]. We were particularly interested in assessing how often there was no chiasma in the tetrad, i.e., how often nullichiasmatic chromosomes segregated properly. Following previous studies (e.g., [22],[23]), we assumed that one of the four chromatids is transmitted at random. We further assumed that there is no chromatid interference, i.e., that when there is more than one chiasma in the tetrad, the pair of chromatids involved in each genetic exchange is chosen independently and at random (cf. [24]). Under this model, given a chiasma, there is always a probability of one half of transmitting a recombinant chromatid. Thus, if there are one or more chiasmata per tetrad, at most half the gametes are expected to be non-recombinants. Assuming that no crossovers are missed due to insufficient marker coverage, a significant excess of non-recombinant gametes beyond one-half provides unequivocal evidence for the existence of tetrads that segregate properly without a chiasma. More generally, the shape of the observed distribution of crossover counts is informative about underlying patterns of recombination in tetrads.
To infer the distribution of the number of chiasmata in tetrads, we used two approaches: a likelihood method, variants of which have been applied in a number of studies [20],[22],[23],[25],[26], as well as a Bayesian approach that we developed and which we believe presents a number of advantages (see Methods). Results from the two methods were similar (see Supplementary Figure 1 in Text S1); to facilitate the comparison to earlier studies, we present those obtained from the likelihood approach.
Applying the approach to crossover data for each chromosomal arm separately, we found overwhelming evidence for frequent, proper disjunction in the absence of a chiasma (Figure 1). In both male and female transmissions, estimates of nullichiasmatic arms are as high as 20–40% for the smaller metacentric chromosomes, but are significantly above 0 even for some of the larger chromosome arms. Given the dense marker coverage of each arm (see Supplementary Table 2 in Text S1), we are missing at most a small fraction of crossover events, rendering our estimates robust. Thus, our findings establish that, in humans, proper segregation does not require the tight regulation of the number of chiasmata per chromosome arm.
We also assessed the stringency of the rule at the level of the entire chromosome. With these data, we can only test whether nullichiasmatic bivalents can segregate properly for the smaller chromosomes, because for the larger chromosomes, they rarely occur (as the recombination rate is simply too high). Indeed, simulations suggest that, even if nullichiasmatic chromosomes always segregated properly, we would have high power to detect their transmission only for the eight smallest chromosomes in males and for chromosomes 21 and 22 in females (see Supplementary Methods in Text S1). As expected, the larger chromosomes show no evidence of nullichiasmatic transmissions to viable offspring (Figure 2). But among the smaller chromosomes, there are apparent exceptions. In males, we found evidence for proper segregation in the absence of a chiasma for two chromosomes: in over 0.1% (the lower 5%-tile) of cases for chromosome 12 (p = 0.0176), and in over 4.2% of cases for chromosome 18 (p = 0.0120). In females, in turn, we inferred that at least 7.3% (p = 0.0002) of chromosome 21 transmissions occur properly in the absence of a chiasma (Figure 3).
Given that the estimated fractions of nullichiasmatic bivalents are relatively small, they may be sensitive to a modest number of missing crossover events. To evaluate this possibility, we performed a number of checks. For chromosomes 12 and 18, these analyses suggested that our results are tentative, as a subset of families are missing informative markers for some of the telomeric regions—enough to potentially inflate the apparent number of non-recombinant gametes and lead to an over-estimate of the fraction of nullichiasmatic tetrads (see Supplementary Methods in Text S1). For chromosome 21, however, the two data sets that we analyzed have good marker coverage and are missing at most a small number of events, indicating that our results are reliable (see Supplementary Methods in Text S1).
In interpreting the chromosome 21 results, a second consideration is the number of tests performed. As noted above, for the larger chromosomes, the probability of the data will be very similar under a model allowing for nullichiasmatic chromosomes and one that does not; only for the small chromosomes could the p-value derived from our likelihood ratio test be small. This reasoning suggests that we should be correcting for many fewer than 22 tests—possibly as few as two. But even if we were to conservatively correct for 22 tests, the results for chromosome 21 remain significant (e.g., if we use a Bonferroni correction, p = 0.0044). Thus, in females at least, the requirement for one chiasma per chromosome is not absolute.
To learn more about patterns of recombination underlying proper disjunction in humans, we used the high resolution of crossover locations in our data in order to better characterize crossover interference. Overall estimates of the strength of crossover interference are similar to those previously reported based on a smaller data set [27], although potential differences emerge between sexes and among chromosomes (see Supplementary Figure 2 in Text S1). Also consistent with earlier studies [27],[28], we find that the centromere is not a barrier to interference (see Supplementary Figure 3 in Text S1). This suggests that when a chiasma occurs close to the centromere of one arm, it will increase the odds of the other arm being nullichiasmatic.
Because of the high density of our markers, we were able to identify a set of double crossovers in close proximity (<5 cM), distributed across families and genomic locations (see Supplementary Table 4 in Text S1). This observation is highly unexpected under the standard statistical model of interference, the gamma renewal model (Figure 4). The excess of tight double crossovers is still apparent even when we focus on a set of stringently vetted double crossovers, which likely under-estimates the true number (Supplementary Figure 4 in Text S1). Thus, in humans, a non-negligible number of double crossovers occur surprisingly close together.
Our statistical analysis indicates that human crossovers are not regulated on the scale of an arm but on that of an entire chromosome (as found also in another mammal, shrews, using a cytogenetic approach [29]). In fact, the genetic length of a chromosome is extremely well predicted by a model that incorporates only two features: the need for one event and the length of the chromosome [19],[30] (see also Figure 5). The tight fit of the model further suggests that there are few chromosome-specific factors affecting the total recombination rate per chromosome and that beyond one event, additional crossovers occur in rough proportion to physical length.
As known for over half a century, the placement of these additional events is subject to positive crossover interference. Interestingly, however, this may not be true of all crossovers (see [19] for discussion). Indeed, we find an excess of rare, double crossovers in close proximity, which are better fit by the model of Housworth and Stahl (2003) [19], where there are two types of crossovers, only one of which is subject to interference, than by the standard interference model (see Supplementary Methods in Text S1). Thus, while other explanations remain, our results are consistent with the existence of a second crossover pathway in humans.
We also see exceptions to the rule of one chiasma per chromosome, most clearly for female (but not male; see Figure 2) transmissions of chromosome 21. Our analysis relies on assumptions of no chromatid interference and no meiotic drive of chromatids [31]. Violations of these assumptions could increase the variance in the transmission of non-recombinant chromatids, potentially serving as an alternative explanation for our findings. We extended our model to mimic this effect (see Supplementary Methods in Text S1) and found that a large variance in transmission probability could account for the observed distribution of crossovers on chromosome 21 even if there were an obligate chiasma to ensure proper disjunction (Supplementary Figure 5 in Text S1). However, studies in yeast, where this hypothesis can be tested directly, found no evidence of chromatid interference [24],[32]. In turn, if meiotic drive explains our findings, we might expect to see over-transmission of certain genotypes in females. Yet there is no evidence for transmission distortion of markers on chromosome 21 (see Supplementary Methods in Text S1). Additional explanations for our finding are strong, female-specific selection against recombinant gametes or the preferential transmission of non-recombinant chromatids. While both remain formal possibilities, of interest in their own right, there is no evidence for such selection in humans, nor is there (to our knowledge) a clear mechanism that would lead to the over transmission of non-recombinant chromatids.
Given that the model assumptions are well supported by studies in model organisms, and the frequent observation of trisomy 21 in humans [3], the most plausible interpretation of our findings is that nullichiasmatic chromosomes 21 occasionally experience proper disjunction. This conclusion is in qualitative agreement with the results of a subset of cytogenetic studies, which find that shorter bivalents sometimes lack an MLH1 focus (e.g., [33]–[35]); for example, Tease et al. [35] found that ∼3.5% of ooyctes from one female lacked a MLH1 focus on the chromosome 21 bivalent. However, the results of such studies vary markedly, likely due to the small number of cells and individuals that can be considered (e.g., [28], [34], [36]–[38]). Moreover, if there is a second crossover pathway in humans, MLH1 foci may not mark all crossovers [16]. Perhaps most importantly, in studies of pachytene cells, the fate of the daughter cells remains unknown [33]; in contrast, we are ascertaining viable offspring, where proper disjunction has clearly occurred.
Intriguingly, the pedigree data suggest that the proper segregation of nullichiasmatic chromosome 21 is fairly frequent. In the absence of a back-up mechanism to aid in the proper disjunction of nullichiasmatic chromosomes, chromatids are expected to segregate randomly to the poles, resulting in chromosomally unbalanced daughter cells in half the outcomes and in balanced cells in the other half. Thus, the rate of proper segregation of nullichiasmatic chromosomes should equal the rate of aneuploidy (in fact, it should be quite a bit lower, since aneuploidy has other sources). Yet fewer than 1% of all conceptions are thought to be aneuploid for chromosome 21—an estimate markedly lower than the fraction of nullichiasmatic transmissions that seem to segregate properly. This large discrepancy raises the possibility of a back-up mechanism in humans, similar to those that exist in Drosophila and yeast [33].
The absence of a chiasma on maternal chromosome 21 is known to be a risk factor for trisomy 21—the main cause of Down's syndrome [20],[22],[39]. If there is a back-up mechanism that aids in the proper disjunction of nullichiasmatic chromosomes, variation in its effectiveness (either across females or with age) could contribute to the risk of forming aneuploid gametes [31].
We previously estimated the location of crossover events in a large Hutterite (a founder population of European origin) pedigree that had been genotyped using Affymetrix 500 K genotyping arrays [9]. To this end, we required K = 5 or more consecutive informative markers to call a crossover event [9]. At the scale of a megabase or more, our sex-specific genetic maps were highly concordant with the those of Kong et al. [40], which are based on more meioses but fewer markers [9].
For these analyses, we focused on 52 overlapping, nuclear families of four or more genotyped offspring, as simulations suggested that our algorithm is highly reliable for families of more than three children (for K = 5) [9]. Based on the total 576 meioses, we constructed the distribution of the number of crossovers in (male or female) gametes, for each chromosome. For female transmissions of chromosome 21, we supplemented our data set with crossover numbers reported in Oliver et al. (2008) [20]. All transmissions were to viable, non-trisomic offspring.
To assign crossovers to chromosome arms, we relied on the centromere gap location in build May 2004 of the human genome (as provided by http://genome.ucsc.edu/). Events were assigned to the p arm if the start of the interval within which they were localized was left of the centromere; remaining events were assigned to the q arm. When the intervals spanned the centromere boundary, we conservatively added the events to both arms; few events fell in this category (see Supplementary Table 1 in Text S1).
Our starting point for inference was the model of Ott (1996) [25], in which the binomial distribution with parameter 0.5 describes the number of crossovers, Y, in a gamete given X chiasmata in the bivalent:(1)
This model assumes no chromatid interference in the distribution of chiasmata across chromatid pairs; in the Supplementary Methods in Text S1, we describe a simple extension that mimics some of the effects of chromatid interference. The object of inference was the probability distribution of the number of chiasmata in bivalents, described by the vector p, where(2)is the probability of x chiasmata in a bivalent. For computational convenience, we assumed that x lies in a finite range 0…xmax; we chose xmax = 20. To obtain maximum likelihood estimates of p for each chromosome and chromosome arm, we employed the EM algorithm of [25]. Confidence intervals were estimated using the parametric bootstrap procedure described in Yu and Feingold (2001) [23], based on 5000 permutations for chromosomes and 1000 for chromosome arms.
To test for the presence of nullichiasmatic bivalents among our sample of gametes (that necessarily underwent proper disjunction), we conducted a likelihood ratio test (LRT) of the unconstrained model (in which p0≥0) to the constrained model (p0 = 0). To fit the latter, we utilized the same EM algorithm subject to the additional constraint that p0 = 0. Because the asymptotic distribution for the LRT statistic is known to provide a conservative test [23], we calculated p-values using parametric bootstrap (again with 5000 permutations for chromosomes and 1000 for chromosome arms).
By taking a complementary, Bayesian approach to inference, we were able to incorporate extensive prior information about recombination patterns in order to improve inferential power, to quantify the deviation of p from a naïve model with no chromosome interference and to avoid possible statistical problems arising from a high-dimensional parameter and modest sample size.
Specifically, using the Kong et al. (2002) [40] estimate of the expected number of crossovers, λ, in a particular chromosome or chromosome arm, we employed a Dirichlet prior on p with parameter vector α, where(3)
The prior expectation of px is αx/J, which equals the Poisson probability of x chiasmata (given a mean of 2λ) and arises from a model of no chromosome interference. J controls the weight of prior information, and allows for deviation from this simple model. We explored values of J = 0.1, 1 and 5. Results were similar across values of J (not shown); for ease of interpretation, we present the results for J = 1. The Bayesian model was fit using Markov chain Monte Carlo (MCMC), which is described fully in the Supplementary Methods in Text S1.
The gamma model, a standard model for crossover interference, was previously found to be a good fit to inter-crossover distances in humans and in a number of other organisms (e.g., [27],[41]). In the gamma model, the locations of the chiasmata on tetrads occur according to a stationary gamma renewal process, where the genetic distances between chiasmata follow a gamma distribution with shape and rate parameters ν and 2ν, respectively. Under the assumption of no chromatid interference, the locations of the crossovers are obtained by thinning the chiasma locations independently, with a probability of 1/2.
Following Broman and Weber (2000) [27], we estimated the parameter ν from our data, for each sex and each chromosome separately (Supplementary Figure 3 in Text S1). The value of ν estimated from the Hutterite data is similar to that obtained by [27], but with tighter confidence intervals due to the larger number of meioses available in the Hutterites. We find some evidence for variation between chromosomes and sexes in the strength of interference (i.e., variation in ν). However, given the lack of fit of the gamma model (Figure 4), these findings should be interpreted with caution.
|
10.1371/journal.pntd.0002456 | Non-Invasive Sampling of Schistosomes from Humans Requires Correcting for Family Structure | For ethical and logistical reasons, population-genetic studies of parasites often rely on the non-invasive sampling of offspring shed from their definitive hosts. However, if the sampled offspring are naturally derived from a small number of parents, then the strong family structure can result in biased population-level estimates of genetic parameters, particularly if reproductive output is skewed. Here, we document and correct for the strong family structure present within schistosome offspring (miracidia) that were collected non-invasively from humans in western Kenya. By genotyping 2,424 miracidia from 12 patients at 12 microsatellite loci and using a sibship clustering program, we found that the samples contained large numbers of siblings. Furthermore, reproductive success of the breeding schistosomes was skewed, creating differential representation of each family in the offspring pool. After removing the family structure with an iterative jacknifing procedure, we demonstrated that the presence of relatives led to inflated estimates of genetic differentiation and linkage disequilibrium, and downwardly-biased estimates of inbreeding coefficients (FIS). For example, correcting for family structure yielded estimates of FST among patients that were 27 times lower than estimates from the uncorrected samples. These biased estimates would cause one to draw false conclusions regarding these parameters in the adult population. We also found from our analyses that estimates of the number of full sibling families and other genetic parameters of samples of miracidia were highly intercorrelated but are not correlated with estimates of worm burden obtained via egg counting (Kato-Katz). Whether genetic methods or the traditional Kato-Katz estimator provide a better estimate of actual number of adult worms remains to be seen. This study illustrates that family structure must be explicitly accounted for when using offspring samples to estimate the genetic parameters of adult parasite populations.
| Genetic epidemiology uses genetic data to uncover patterns of disease processes. To acquire data for these analyses, individual pathogens are collected and scored at genetic markers, and the resultant data are analyzed to infer biological patterns about the pathogen populations. In lieu of invasive sampling of adult pathogens in humans, researchers have relied on non-invasive sampling of parasite offspring (often shed in fecal samples). One potential problem with this approach is that analyses using the offspring data will be biased because many of the offspring are related and family sizes are likely to be unequal. We show that this sampling issue is relevant in a natural transmission zone in western Kenya and that it yields biases in three important parameters: genetic differentiation, inbreeding coefficients, and estimates of the amount of non-random association between loci (linkage disequilibrium). We also develop a method to remove these biases by removing the sibling structure present in the dataset. Finally, we suggest that our measure of family number, as well as other genetic measures, may be useful measures of the worm burdens in patients.
| Infectious disease research is rapidly adopting the tools of evolutionary biology and molecular ecology [1]–[5]. Molecular genetic data, evolutionary theory, and population genetic tools can provide methodology to uncover epidemiological processes that cannot be easily determined otherwise. Such processes include pathogen migration and gene-flow, strain divergence, and selection [6]–[9]. However, some pathogens can be difficult subjects for molecular studies because their adult stages cannot be ethically or pragmatically collected from their human hosts. Thus, researchers often rely on the collection of progeny to infer information about the adult population. Schistosome parasites are one such example. Schistosomes are dioecious blood flukes that become reproductively mature in the vasculature (mesenteric veins or the veins of the bladder plexus) of their hosts where they reside in primarily monogamous pairs [10], [11]. The adults are inaccessible, but their offspring can be collected as eggs that are shed in urine or feces. Consequently, schistosome offspring are often used as a proxy for the adult population, typically to infer worm burdens, and genetic structure among host individuals, host species, and geographic locations [12]–[20].
One challenge associated with using samples of offspring to assay genetic structure is that a sample of offspring may be misrepresentative of the adult population, and can thus give biased estimates of parameters of the adult population [9], [21], [22]. Two types of biased parameter estimates can result when offspring are produced by a small effective number of breeders, Nb (either because few adults were breeding and/or because those that did breed had highly skewed reproductive output). First, the sampling variance in population allele frequencies that arises from sampling the offspring of a small effective number of adult breeders will yield inflated estimates of genetic differentiation among hosts. Second, the strong sibling structure that will exist in a large sample of offspring having a small Nb will cause negative deviations from Hardy-Weinberg equilibrium (i.e. downwardly biased estimates of FIS) [e.g. 23], [24], [25] and inflated estimates of linkage disequilibrium (LD) among loci within hosts. The likelihood that these sampling artifacts will arise when sampling offspring depends on the number of breeding adults per host, the reproductive skew among those adults, and on the sample size of offspring collected. Furthermore, because it is relatively easy to collect large numbers of offspring, one can reach false conclusions with high statistical confidence.
The distribution of reproductive output among individuals in natural populations of organisms is usually highly skewed, causing the ratio of the variance of reproductive output to the mean reproductive output, (variance to mean ratio, VMR), of the adults in a population to be much greater than 1 [e.g. Table 10.2 in 26]. In laboratory infections of mice, schistosomes showed VMR ranging from 7.2 to 7.4 [10]. This reproductive skew produced a ratio of effective number of breeders (Nb) to census number of breeding adults (Nc) of 0.24. VMR has not previously been measured in natural populations of schistosomes. Therefore, in order to accurately measure important epidemiological parameters, it is essential to determine how large of an effect the above sampling issue will have on population genetic studies of this parasite.
Schistosomes are a substantial public health issue in tropical and developing countries. They are estimated to infect over 200 million people (approximately 1 out of 35) worldwide [27]. Schistosomiasis is a chronic and debilitating disease with a life cycle that is difficult to control. Long-lived and recurrent infections present an ongoing inflammatory challenge that can result in anemia, severe portal hypertension, malnutrition, poor growth, impaired cognitive development, increased suseptibility to coinfection, and increased pathology in coinfection [28]. The schistosome life cycle involves a snail intermediate host and a mammalian definitive host. Eggs are released with the urine or feces of the mammalian host, hatch in water, and release free-swimming miracidia. Miracidia infect snails and undergo asexual reproduction resulting in thousands of clonal cercariae that emerge from the snail daily. Cercariae penetrate the skin of their definitive host and establish long lived infections averaging 6–11 years [29]. Estimating population genetic parameters such as F-statistics of adult schistosome populations is important because they can reveal local transmission patterns and the distribution of genetic variation within and among hosts and geographic regions. Genetic data might also be useful in providing measurements of worm burdens and their effective population size, parameters that can be difficult to measure otherwise [30], [31]. Accurate estimates of these population genetic parameters are sorely needed for effectively targeting drug treatment efforts against schistosomes [32] and to ameliorate reduced drug susceptibility in schistosome populations, which has already been detected in a natural population [33].
The primary aim of this study was to investigate whether sampling artifacts are likely to influence population genetic studies of schistosomes from humans when offspring are sampled in lieu of adults. We collected data from 2,424 miracidia noninvasively sampled from 12 humans in western Kenya. To determine the amount of reproductive skew and family structure naturally present in samples collected from humans, we used microsatellite genotype data to cluster offspring into putative sibships. We investigated how family structure influenced LD, inbreeding coefficients (FIS) within hosts, and genetic differentiation (FST and GST) among hosts, and we developed a correction procedure to remove the bias introduced by family structure. We also investigated whether a single fecal sample taken from a human host would give an adequate representation of the genetic composition of worms in that host, or whether multiple fecal samples should be obtained from a host over several days.
Throughout the paper we use “infrapopulation” to refer to the adult worms in a host (patient), and “component population” to refer to all the adult worms in all the hosts of a host population [34]. The term “sample” refers to a sample of miracidia from a host (i.e. the offspring sampled from an infrapopulation). Initially, we measured population genetic parameters of adult and offspring schistosomes that were collected from mice as part of a prior study [10]. Because we detected the predicted biases in these samples from which both offspring and adult populations could be assayed (See Online Supporting Information, S1), we collected data from humans naturally infected with schistosomes so that we could determine if these sampling artifacts are relevant to samples from humans residing in a natural transmission zone.
We obtained miracidia from human fecal samples from twelve participants enrolled in a longitudinal study [35]–[38]. As these samples were considered discarded medical waste they were viewed as “exempt” by the University of New Mexico Internal Review Board. As part of the longitudinal study, patients were monitored periodically, and if infected, were treated with praziquantel. Patients were adult males who work in Lake Victoria near Kisumu, Kenya, and were either car washers or sand harvesters. Car washers stand knee to ankle deep in the lake as they wash vehicles near the edge of the lake. Sand harvesters stand up to chest-deep, shoveling sand from the bottom of the lake into boats to sell to concrete manufacturers. Both groups of men were exposed to schistosome cercariae as they worked. These patients presumably were exposed to the same pool of cercariae in the lake (spatially and temporally), such that we expect no spatial or temporal genetic subdivision among worm populations from different patients. In support of that expectation, no spatial genetic subdivision was found when cercariae were sampled from snails in this same region, nor was there evidence for LD or deviations from HWE in those cercarial samples [39]. Note that those cercarial samples were scored using the same microsatellite loci as used in this study [39].
To obtain miracidia, eggs were hatched using standard protocols [18]. The miracidia were lysed individually in the wells of a 96 well plate and genotyped at 12 microsatellite loci as described by Steinauer et al. [40]. GenBank accession numbers of the loci include the following: AF325695, AF202965, AF202966, AF202968, L46951, AF325698, AF325694 (Multiplex panel P17) and M85305, R95529, AI395184, BF936409, AI067617 (multiplex panel P22). Only those individuals having data for at least 10 of the 12 loci were included in the analysis.
We used sibship analyses to determine whether family structure was present in schistosomes collected from naturally infected human hosts. COLONY v.2.0 [41], [42] was used to partition individual miracidia into their probable sibling groups. COLONY implements a maximum likelihood approach that can incorporate genotyping errors to identify full-sibling and half-sibling families [42]. Using COLONY, we performed analyses with two different user-defined options, first using a monogamous mating system, and second using a polygamous mating system (to include half sibships). Because there was no empirical support for the presence of half sibling families (see below) and designating a polygamous mating system resulted in suboptimal full sibling family partitions, we used only the COLONY results where monogamous mating was specified (see online supplement S2). Nevertheless, the overall results were very similar when using the samples generated with a polygamous mating system (see Online Supporting Information, S2, for details on the performance of COLONY on these samples). Analyses with COLONY were run with full likelihood, with no priors or known allele frequencies, and one short run per dataset. Our sibling partitions from COLONY were similar to those derived from alternative software packages (see online supplement S3).
The results from COLONY were first used to calculate the number of families occurring in each offspring sample (FSF = “full sib families”). This number included families with a sibship size of 1. Because the total number of families identified in a sample should increase with the number of sampled miracidia, n, we also calculated FSF/n in order to compare among samples. This number gives a relative estimate of infrapopulation size. We employed a commonly used metric, the variance to mean ratio (VMR) for reproductive success, to quantify the amount of family structure in each sample of offspring. Ratios greater than one indicate a skewed distribution in reproductive output or a large variance in family sizes. It is important to point out that this metric will be downwardly biased when the sample size is much smaller than the true number of breeders, thus it is a minimum estimate. We also estimated the effective number of breeders [Nb; 43] within each patient as another relative measure of infrapopulation size using the Linkage Disequilibrium (LDNE) and Sibship Assignment (SA) [44] methods. The former was calculated using LDNE [45], and the latter with COLONY v. 2.
To correct both empirical and simulated samples for family structure, we randomly sampled one individual per family to create a reduced data set that no longer contained any full-sib individuals. We have named this the “one-per-family” approach. Because there is some inherent stochasticity associated with randomly selecting a single individual from each family, we used custom scripts to automate the process, which allowed us to create a large number of “one-per-family” samples for calculations of genetic parameters. By creating a large number of samples, we can effectively sample all single individuals from a given family and capture the associated mean and variance (See online supplement S4 for a discussion of the variance in these samples).
We predicted that small Nb and large family structure within samples of miracidia would increase LD among loci, cause negative deviations from HWE (more negative FIS), and inflate FST among hosts [9]. Therefore, these parameters were compared between the raw, uncorrected samples and the corrected samples from humans. To account for the reduction in sample size in the corrected samples after removing full siblings, we calculated Weir and Cockerham's estimation of FST (theta), which is unbiased with respect to sample size [46]. We also calculated standardized FST (FST of the sample relative to the maximum FST value possible given the dataset) [47] using RecodeData v. 0.1 [48]. To calculate pair-wise theta, we used the Geneclust package in R [49], [50]. We automated the process to iteratively (1) create a one-per-family dataset for each patient and (2) calculate all pair-wise FST values between patients. This process was repeated 1000 times, after which we calculated the mean and 95% CIs for the one-per-family FST values and compared them to the uncorrected data set. We repeated the above procedure to calculate a global (as opposed to pair-wise) value of Weir and Cockerham's FST using the R package Hierfstat [51]. We also calculated the corrected standardized GST (G″ST) [52] using GENODIVE [53]. GST is often used as an analog of FST because FST is dependent on within sample diversity. For assessment of our correction method, G″ST was also calculated in 10 randomly generated corrected datasets for comparison and tested for significance in each dataset using 10,000 permutations of the data. To calculate within-patient FIS for uncorrected (raw) and 1000 one-per-family samples, we used R scripts to combine observed and expected heterozygosities using the standard equation FIS = (He-Ho)/He [54].
We next exported 50 one-per-family data sets from R and imported them into GENEPOP to test for genotypic disequilibrium (a proxy for LD). For each patient we used 1000 batches and 10000 iterations per batch to calculate the percentage of loci pairs that were significantly associated with one another (p≤0.05, averaged over the 50 one-per-family data sets). Because statistical tests can be affected by sample size, we repeated the above procedure on “downsampled” empirical data sets, where we reduced the sample size of miracidia from each patient to the equivalent sample size in the one-per-family data sets from that patient. Unlike in the one-per-family data sets, in the downsampled datasets we removed individuals randomly without respect to the predetermined family structure. This process allowed for equitable comparisons of LD between the one-per-family data sets and the uncorrected (empirical) data sets with the same sample size.
To further validate the one-per-family approach to correcting for family structure, we also created simulated data sets where we could precisely control which individuals belonged to given families. We used a distribution of family sizes that matched the empirical distribution from the COLONY output to recreate the observed family structure. To create simulated adult schistosomes, we used the empirical allele frequencies to generate multilocus male and female genotypes in accordance with HWE. We next paired adults monogamously and randomly selected one allele from each parent to create offspring in accordance with Mendelian expectation. We generated 1000 offspring per adult pair and then randomly sampled a precise number of offspring per full-sib family in accordance with the COLONY output from the empirical data for family size. For example, if a patient had two families of sizes 5 and 30, then 5 full-sib offspring would be sampled from one pair, 30 full-sib offspring would be sampled from the next pair, and the remaining offspring would be discarded. Each offspring was assigned a unique individual and family ID to validate the downstream analyses. All simulated data sets were constructed with a script written in R 2.15.1 [55]. A fully annotated version of this script is freely available at dryad. Using the exact approach as described for the empirical data, we tested the effect of correcting samples on 1000 simulated datasets for measuring FST, FIS, and LD. Notice that with these simulated data sets we knew the family structure with 100% certainty, whereas with the empirical data we assumed that COLONY accurately captured all of the family structure. Thus, these simulated data sets allowed us to accurately verify the utility of our approach to correcting for family structure.
As even further proof of principle, we also conducted analyses with Kenyan schistosomes in a mouse model system [two mice infected with field-collected schistosomes; see 10 for laboratory methods]. Because we could sample both the adults and the offspring (which is not logistically feasible with human patients) the samples of offspring could be directly compared to the samples of adults for measurements of linkage disequilibrium, FST, HWE, and parentage analysis (See Online Supporting Information, S1, for details).
If reproductive rates are constant, then samples from a single patient should be genetically homogenous over multiple days. However, if reproduction occurs in “bursts”, then sampling a single fecal sample could miss diversity within a patient. For three patients 2–3 fecal samples were collected 1–2 days apart. To determine if these samples differed in genetic composition, we first calculated pairwise FST between temporal samples from the same patient using the raw data and tested their significance via 10,000 permutations of genotypes among samples using FSTAT 2.9.3 [56]. Standardized FST (F′ST) was calculated and standardized GST (G″ST) was calculated and tested for significance with 1000 permutations using GenAlEx v. 6.5 [57], [58].
Next, for the patients for which the raw data indicated significant temporal differences, we corrected them for family structure using our one-per-family approach. Families were resampled 1000 times and mean pairwise FST was calculated among sample days within each of the three patients. Because the reduced sample size should influence significance testing, we performed the same tests on downsampled samples to match the sample size of the corrected samples.
Another application of the calculation of the number of full sibling families in a sample of miracidia is to estimate the minimum worm burden within a patient (i.e. minimum number of breeding pairs). However, it is possible that with adequate sampling, this measurement, as well as other genetic parameters, may serve as relative or even absolute measurements of worm burden. The number of families detected, as well as other genetic parameters such as allelic diversity, should be strongly correlated with the number of worms that were reproductively active when the sample was taken. With this aim, we investigated the relationship of several parameters obtained from our genetic data and compared them to the World Health Organization gold-standard estimator of worm burden, the number of eggs per gram of feces as determined by the Kato Katz method [59]. The genetic parameters we measured included: the number of full sibling families (FSF), a standardized measure of full sibling families (FSF/n), allelic richness (number of alleles rarefied to the smallest sample size (AR), and the effective number of breeders [Nb; 43]. Nb was estimated using the sibling assignment method implemented in COLONY [44] and the linkage disequilibrium method implemented in LDNe [45]. Although the actual worm burdens in our patients cannot be determined, it is worth determining the extent to which the genetic estimators are intercorrelated and capture the same information as the Kato Katz method (eggs per gram of feces). Pearson's correlations between egg count (i.e. Kato Katz), and the genetic parameters: FSF/n, AR, and Nb were performed using Graphpad Prism v. 5.01 (GraphPad Software, Inc.). To further determine similarity among the estimators, a multivariate principal components analysis (PCA) using egg output (i.e., Kato Katz), FSF/n, AR, and both measures of Nb as variables was performed using Systat 11 (Systat Software, Inc.).
The samples of miracidia from humans varied substantially in the amount of family structure they contained (Table 1, Figure 1). The percentage of miracidia that belonged to a family of four or more (i.e., the number of miracidia which belong to a family with robust support) ranged from 18–91% among samples, and the variance to mean ratio (VMR) for the number of offspring per family ranged from 0.37 to 7.9. The lower values of VMR likely reflect our inability to detect a skew in samples with very large populations rather than equal reproductive output among families. When infrapopulations are large, one needs very large sample sizes to accurately describe the family-size distribution (else you wind up with mostly unrelated individuals). Indeed, several lines of evidence in our data indicate that samples with measured VMR less than one are likely those in which the sample size is much smaller than the actual number of families. First, the ratio of the estimated Nb to the sample size was high in samples with VMR<1 (Table 1, Figure 1). Furthermore, the percentage of individuals belonging to a family of 4 or more was highly correlated with VMR (Pearson's r = 0.807, P = 0.0008) indicating that the samples with a low VMR had a large number of families rather than having a small number of families of equal size. Thus, the measured reproductive skew of samples will be highly dependent on the infrapopulation size and the sample size.
In the analyses of samples of miracidia from humans in which the mating system was designated as “polygamous”, many half sibships were inferred in the COLONY partitions. However, most of these included only 2 or 3 members, or involved a large full sibling group with one or two half siblings. Such small groupings are likely to be spurious. Only three patients had half sibling families that consisted of greater than three individuals per FSF (two such families per patient). Furthermore, analysis of our simulated samples revealed that the majority of half-sib assignments were incorrect (see Online Supporting Information, S2). Thus, we do not see strong evidence for a large number of half sibling groups in our samples from humans, which indicates that these patients were not getting infected with large numbers of genetic clones derived from a single snail.
The global FST for the uncorrected data set was 27.8 times higher than the corrected, one-per-family data sets (Corrected dataset global FST = 0.00026; Uncorrected dataset FST = 0.0074; Uncorrected dataset standardized global F′ST = 0.027). G″ST indicated significant population subdivision of the uncorrected dataset (G″ST = 0.031, P = 0.001), and was greatly reduced in the 10 corrected datasets and permutation tests indicated no significant population subdivision (mean G″ST = −0.0013, range −0.004 to 0). Thus, correction of the dataset reduced standardized G″ST from 0.031 to a mean of −0.0013; and statistical significance was lost with this correction. Furthermore, pairwise FST between patients was greatly reduced in the one-per-family data sets (Figure 2). In fact, for the simulated datasets the mean pair-wise FST was reduced to 0. As expected, this analysis revealed that Weir and Cockerham's FST estimator is unbiased, though some comparisons differed from 0 due to decreased precision. The analysis using the empirical data set yielded a nearly identical result, with most of the genetic differentiation being removed from the corrected data sets. Interestingly, the mean pairwise FST for the empirical data set was slightly greater than 0 (approximately 0.001), suggesting a very low level of residual real or artifactual differentiation (see Discussion). The one-per-family correction worked well for both the empirical and simulated data and the confidence intervals surrounding the means from the 1000 iterations were narrow (smaller than the points on the plots). Also, it is important to note that simply downsampling the data randomly (without regard to family structure) to match the sample sizes of the corrected data yielded a mean pairwise FST value (from 5000 resampled datasets) that was very similar to the value for the uncorrected data (FST = 0.009). Thus, the change in FST value was due to the removal of family structure.
As predicted, samples with a large amount of family structure also yielded lower estimates of FIS as indicated by the negative correlation between FIS and VMR (r = −0.638, P = 0.0128) (Figure 3A). This relationship was not detected in the corrected samples (r = −0.3797, P = 0.112). The increase in FIS between raw and corrected samples was greatest in the samples having most family structure and these values were positively correlated (r = 0.733, P = 0.003) (Figure 3B).
All of the uncorrected samples of miracidia from humans showed significant genotypic disequilibrium, ranging from 38 to 91% of pairwise comparisons of loci (Figure 4). This percentage of loci was much lower in the corrected samples ranging from 0.015 to 0.065%, suggesting that most of the linkage disequilibrium was due to family structure. Correction also reduced the number of loci in LD when compared to the raw samples that were downsampled to match the same sample size as the corrected samples (Fig. 4). Correction also removed the relationship between VMR and the percentage of loci in disequilibrium.
The same sampling artifacts were detected in the samples from mice (See Online Supporting Information, S1, for details and methods). FST between the adult schistosome infrapopulations in the two mice was low and not statistically significantly different from zero (−0.026 [95% CI: −0.032–0.01]; P = 0.998). In contrast, FST was substantially higher (0.021 [95% CI: 0.014–0.029] and statistically significant between the samples of miracidia collected from each of the mice (P = 0.0001). Although, neither adult nor offspring samples deviated significantly from HWE, point estimates of FIS were more negative in the samples of miracidia, consistent with predictions. These samples of miracidia also showed many pairs of loci in significant linkage disequilibrium, while the adult samples showed no loci in significant disequilibrium.
When the temporal fecal samples from the same patient were analyzed in their raw form (no correction for sibships), allele frequencies of miracidia differed significantly among fecal samples collected on different days for two of the three patients (patient 2: F′ST = 0.021, P = 0.0009, G″ST = 0.021, P = 0.003; patient 3: F′ST = 0.001, P = 0.333, G″ST = 0, P = 0.513; patient 12: F′ST = 0.020, P = 0.0001, G″ST = 0.018, P = 0.002). We sampled patient 12 on three days. Pairwise tests indicated that the fecal sample collected from day 3 was significantly different from those on day 1 and day 2, but samples from day 1 and 2 were not significantly different (Day 1&2: G″ST = 0.006, P = 0.173; Day 1&3: G″ST = 0.028, P = 0.001; Day 2&3: G″ST = 0.019, P = 0.003).
We corrected the samples using the one-per-family method only for those two patients for which the raw data showed significant differentiation from the pairwise FST tests. Correcting the samples reduced global FST in all the patient samples and the amount of correction was the highest in the patient with the highest VMR (#12) (patient 2: raw/corrected FST = 0.006/0.003; patient 12: 0.006/0.001). After correction, there were no significant temporal differences between fecal samples. However, the differences between the uncorrected samples also became non-significant when they were downsampled to match the sample sizes of the corrected samples.
Egg output (i.e., Kato Katz method) was not correlated with either the number of full sibling families FSF/n (r = −0.097, P = 0.383), with AR (r = 0.153, P = 0.318), or either measure of Nb (SA: r = −0.049, P = 0.440; LDNE: r = −0.169, P = 0.300). However, PCA analysis indicated that all the genetic factors were strongly related and loaded heavily on one factor (loading values = 0.90–0.96), with egg counts loading heavily on a second factor (0.99). Factor one (genetic measures) explained 69.1% of the variation and factor two (egg counts) explained 22.2% of the variation. Further examination of univariate correlations indicated that all the genetic measures were strongly correlated (Pearson's r = 0.79–0.83 and P≤0.001) (Fig. 5). The strong intercorrelations among the genetic parameters suggests they are all capturing the same information about variation among patients in number of families contributing to each sample. However, that variation among patients is uncorrelated with variation in the traditional, Kato-Katz estimate of worm burden.
We demonstrated that samples of S. mansoni miracidia collected from human hosts can contain large numbers of full-sib families and that not explicitly accounting for this family structure could cause one to draw false conclusions concerning population structure. Importantly, this bias could occur in any system where offspring are sampled in lieu of the adults, and has previously been documented in a variety of systems [e.g. 21], [60], [61], [62]. As predicted, the presence of family structure in samples was correlated with reduced measures of FIS and inflated measures of linkage disequilibrium and FST values. Correcting for family structure in our samples of miracidia had a large effect on estimates of population genetic parameters. For example, the number of pairs of loci in linkage disequilibrium in the empirical samples was dramatically reduced after correction (from a maximum of 91% to a maximum of 0.065%). Similarly, correcting for family structure yielded global FST values that were 27 times lower than that from the uncorrected samples. These results were further corroborated by data simulations. FIS values were also downwardly biased by the presence of family structure, although this bias was not large enough to cause the offspring samples to significantly deviate from HWE (data not shown). Even though this bias appears to be small, it could mask important biological information, such as inbreeding or Wahlund effects, or even important laboratory artifacts (e.g., null alleles), that could otherwise be detected. A heterozygote excess over HWE expectations is predicted for samples that contain many siblings and the extent of the excess is increased in hosts having lower Nb and higher family structure [23]–[25], [63], [64].
The upward bias measured in global FST in this study may, at first glance, appear to be modest. However, the documented bias is non-trivial and could result in the misinterpretation of important biological processes. For example, even after Bonferroni corrections, pairwise FST tests among uncorrected patient infrapopulations (i.e. including siblings) show highly significant population differentiation among all patients. Consequently, both the global and pair-wise estimates of genetic differentiation may cause researchers to erroneously conclude that the samples were genetically differentiated. However, because all patients were exposed to the same pool of cercariae both spatially and temporally, and more importantly, because both the simulated and empirical data sets that were corrected for family structure yielded mean FST estimates close to zero (0, 0.001), we conclude that there is no real genetic differentiation between patients. This conclusion is further reflected in the data from the mouse system. Even though the adult populations were not differentiated, the uncorrected offspring datasets appeared to be. Thus, biases presented by including siblings in population genetic analyses could potentially affect the interpretation of all of the datasets presented here. More importantly, these biases could affect a wide variety of larger-scale studies that make genetic inferences based on sampling offspring.
Here, we also present a correction method that yields population genetic parameter estimates with this bias removed or greatly reduced. Although some loci remained in LD and the average FST of the pairwise comparisons of empirical data was marginally positive in the corrected samples, the correction yields much improved estimates that are likely to be more biologically meaningful. We hypothesize that the residual positive FST and loci in LD are due to family structure not recognized by the sibship analysis and thus not removed from the dataset after correction. This family structure could be due to failure of the sibship analyses to recognize small sibling families or the presence of small half sibling families in the samples (our sibship analyses did not uncover half-sibships with large membership in our samples). Half-sibships could theoretically be derived from clonal cercariae (products of asexual reproduction in the snail) that are transmitted together to the same definitive host and pair with genetically different partners. This mating pattern might occur if the cercariae are unable to move far from their infected snail (e.g. constrained to a small pool). However, our results suggest that most cercariae are well-mixed before infecting their hosts. Another way half-sibships could be produced is through mate switching. However, mate-switching is not likely to occur in our samples because each fecal sample represents a small window of reproduction and schistosomes are primarily monogamous [10], [11], [65].
We have shown that significant family structure is likely to be present in samples of miracidia collected from human patients. Therefore, using uncorrected genetic data collected from schistosome offspring (miracidia) to infer important epidemiological parameters of the adults is likely to generate false conclusions. Failure to correct for family structure can cause one to overestimate FST, linkage disequilibrium, and heterozygosity (negative bias in FIS). Previous schistosome epidemiological studies have used a hierarchical AMOVA approach in order to remove the bias caused by family structure in offspring samples [e.g. 12], [66]. Although this method may be successful by removing the among infrapopulation variance and thus bias caused by family structure so that the data may be interpreted at the top hierarchical levels (e.g. geography), this approach does not make corrections for comparisons at the infrapopulation level.
We observed a statistically significant FST>0 between the raw data from two sets of temporal samples that were each collected from a patient over multiple days. Tests using only the raw data would cause one to conclude that genetically different subsets of worms are producing eggs on different days. Family structure and differential representation of families in sequential fecal samples appear to be driving the statistical differences in our data. It should be noted that these biases can occur in samples with both large and small family structure (i.e. patient 12 and patient 2, see online supplement, S5). The differentiation between samples may be due to random sampling error (some families are missed by chance), or biological attributes that change a worm pair's contribution to a single fecal sample (i.e. location in host, age, competition among pairs). To answer this question, much larger samples are necessary. In any case, fecal samples from multiple days may be more representative than a single fecal sample. However, it still remains unclear whether sampling artifacts can be best overcome by increasing the sample size on one day, or collecting several small samples over multiple days.
The number of full sibling families detected in a sample of miracidia is a measure of the minimum number of worm pairs present in a patient. The strong covariation among genetic variables in our data suggests that genetic parameters could be used further: to depict relative worm burdens in patients. It is also possible that, given a large enough sample size, the true worm burden within a patient could be detected via the number of full sibling families. The challenge will be obtaining a large enough sample size to account for the true worm burden, a parameter that unfortunately may not be known without obtaining genetic data first. As shown in Table 1, the sample size necessary to detect most sibling families varies widely among patients. For example, for Patient 6, 75% of the miracidia collected were partitioned into robust full sibling families (>3) with a sample size of 81 miracidia. In contrast, with 412 sampled miracidia, only 49% were partitioned into robust families for Patient 3. However, it may be possible to obtain accurate worm burden estimates even without exhaustive sampling, by fitting the observed sibship sizes to an expected sibship size distribution to predict the number of unsampled sibships.
The lack of correlation among fecal egg counts (i.e., Kato Katz) and our genetic proxies for worm burden is interesting, particularly considering the strong correlations detected with genetic measures even with a low sample size of 12 patients. The Kato Katz methodology for egg enumeration is known to be highly variable between fecal samples collected from the same patient and has been deemed unreliable by some particularly when worm burdens are low [67]–[69]. However, others have found evidence of reliability of this method across a broad range of infection intensities [e.g. 70], [71]. It may be that the infection levels in our study were not broad enough for the Kato Katz method to accurately detect relative differences in the worm burdens of our patients. Although we have no independent data to determine which approach gives the most accurate estimates of the true number of adult breeding worms, we suggest that genetic methods potentially give more reliable estimates of the relative number of adult breeding worms per host. These methods should be explored further because they could be valuable tools for epidemiological studies that measure the success of control programs.
A previous study did not find a relationship between their calculated Ne from schistosome offspring populations and allelic richness [19]. However, it appears that their estimates of Ne may not have been very accurate because there was no correlation between Ne values from the same samples calculated by two techniques and they were estimated with large confidence intervals. This lack of accuracy could be due to the small number of markers used, relatively small sample sizes (10–30 miracidia per patient), and pooling of samples from several infrapopulations rather than using a subdivided breeders model [31] to calculate Ne. The lack of correlation may also be due to saturation of allelic richness since there is a limit to the number of alleles that will be found in a population and a correlation beyond this saturation point is not expected.
Genetic epidemiology is a powerful tool for infectious disease research. However, in cases where offspring must be collected in lieu of adults, data analysis and interpretation should be carefully considered. We have shown that samples of parasite larvae collected from humans can contain significant family structure, which can lead to inflated estimates of linkage disequilibrium and FST, and underestimates of FIS. The amount of bias in each of these parameters is positively correlated with the skew in reproductive output of individuals. It should also be noted that sibling structure and skewed reproductive output among individuals (small Nb) could skew additional population genetic parameters and analyses not evaluated here such as observed heterozygosity (Ho), gene diversity (He) [72], genetic distance, and clustering algorithms, thus, care should be taken in their interpretation. Correcting samples by performing a sibship analysis and then excluding all but one member of each full sibling family is effective at removing or reducing this bias. The number of full sibling families detected by our analyses gives an estimate of the minimum number of worm pairs within a patient and may be a reliable estimator of the relative worm burdens within patients, an important epidemiological parameter.
Microsatellite genotype data and annotated R scripts to perform the “leave-one-out” procedure will be made available at Dryad.
|
10.1371/journal.ppat.1005648 | An Oncogenic Virus Promotes Cell Survival and Cellular Transformation by Suppressing Glycolysis | Aerobic glycolysis is essential for supporting the fast growth of a variety of cancers. However, its role in the survival of cancer cells under stress conditions is unclear. We have previously reported an efficient model of gammaherpesvirus Kaposi’s sarcoma-associated herpesvirus (KSHV)-induced cellular transformation of rat primary mesenchymal stem cells. KSHV-transformed cells efficiently induce tumors in nude mice with pathological features reminiscent of Kaposi’s sarcoma tumors. Here, we report that KSHV promotes cell survival and cellular transformation by suppressing aerobic glycolysis and oxidative phosphorylation under nutrient stress. Specifically, KSHV microRNAs and vFLIP suppress glycolysis by activating the NF-κB pathway to downregulate glucose transporters GLUT1 and GLUT3. While overexpression of the transporters rescues the glycolytic activity, it induces apoptosis and reduces colony formation efficiency in softagar under glucose deprivation. Mechanistically, GLUT1 and GLUT3 inhibit constitutive activation of the AKT and NF-κB pro-survival pathways. Strikingly, GLUT1 and GLUT3 are significantly downregulated in KSHV-infected cells in human KS tumors. Furthermore, we have detected reduced levels of aerobic glycolysis in several KSHV-infected primary effusion lymphoma cell lines compared to a Burkitt’s lymphoma cell line BJAB, and KSHV infection of BJAB cells reduced aerobic glycolysis. These results reveal a novel mechanism by which an oncogenic virus regulates a key metabolic pathway to adapt to stress in tumor microenvironment, and illustrate the importance of fine-tuning the metabolic pathways for sustaining the proliferation and survival of cancer cells, particularly under stress conditions.
| KSHV is causally associated with the development of Kaposi’s sarcoma and primary effusion lymphoma; however, the mechanism underlying KSHV-induced malignant transformation remains unclear. The recent development of an efficient KSHV-induced cellular transformation model of primary rat mesenchymal stem cells should facilitate the delineation of KSHV-induced oncogenesis. In this report, we have used this model to investigate the metabolic pathways mediating the proliferation and survival of KSHV-transformed cells. In contrast to most other cancers that depend on aerobic glycolysis for their fast growth, we demonstrate that KSHV suppresses aerobic glycolysis and oxidative phosphorylation in the transformed cells. Significantly, suppression of aerobic glycolysis enhances the survival of the KSHV-transformed cells under nutrient deprivation. Mechanistically, KSHV-encoded microRNAs and vFLIP suppress aerobic glycolysis by activating the NF-κB pathway to downregulate glucose transporters GLUT1 and GLUT3. We have further shown that GLUT1 and GLUT3 inhibit constitutive activation of the AKT and NF-κB pro-survival pathways. Strikingly, GLUT1 and GLUT3 are significantly downregulated in KSHV-infected cells in human KS tumors. Furthermore, we have detected reduced levels of aerobic glycolysis in several KSHV-infected primary effusion lymphoma cell lines and a KSHV-infected Burkitt’s lymphoma cell line BJAB. Our results reveal a novel mechanism by which an oncogenic virus regulates a key metabolic pathway to adapt to stress in tumor microenvironment, and illustrate the importance of fine-tuning the metabolic pathways for sustaining the proliferation and survival of cancer cells, particularly under nutrient stress microenvironment.
| It has been recognized that metabolic reprogramming is a core hallmark of cancer[1]. The Warburg effect describes the dependence of cancer cells on aerobic glycolysis for their growth and proliferation[2]. Increased glucose uptake and aerobic glycolysis are widely observed in cancer and clinically exploited for diagnosis[3]. Aerobic glycolysis provides a fast supply of ATP to support the rapid growth and proliferation of cancer cells[3]. Recent works have shown that besides energy, cancer cells have special needs for macromolecular building blocks and maintenance of redox balance[4, 5]. Accordingly, metabolic adaptation in cancer cells has been extended beyond the Warburg effect[5]. Several types of cancers depend on glutamine or one carbon amino acids for growth and proliferation[4, 5].
Cancer cells often encounter a variety of stress conditions including low nutrients, low oxygen and excess byproducts in the microenvironment[4, 6]. To optimize the growth, proliferation and survival under diverse conditions, cancer cells must fine-tune the metabolic pathways. Hyperactivation of metabolic pathways can generate toxic products that are detrimental to the cancer cells[6]. For examples, overflow of oxidative phosphorylation produces reactive oxidative species while excess of aerobic glycolysis leads to the buildup of lactate and low pH in the microenvironment[6]. How cancer cells regulate metabolic pathways to adapt to different stress conditions is not entirely clear.
Kaposi’s sarcoma-associated herpesvirus (KSHV) is an oncogenic virus associated with several cancers including Kaposi’s sarcoma (KS) and primary effusion lymphoma (PEL)[7]. Infection by KSHV has become an excellent model for understanding the mechanism of oncogenesis. Experimentally, KSHV can efficiently infect and transform primary rat mesenchymal precursor cells (MM) and human mesenchymal stem cells[8, 9]. KSHV-transformed MM cells (KMM) efficiently induce tumors with features closely resembling KS[8]. In KS tumors, PEL and KMM tumors, most of tumor cells are latently infected by KSHV. These cells have restricted expression of viral genes including vFLIP (ORF71), vCyclin (ORF72), LANA (ORF73) and 12 precursor microRNAs (pre-miRNAs)[8, 10, 11]. Genetic analyses have revealed that viral miRNAs and vCyclin are critical for KSHV-induced oncogenesis by regulating cell cycle and apoptosis[10], and overriding cell contact inhibition[12], respectively.
KSHV infection induces Warburg effect in human endothelial cells (ECs) and lipogenesis in ECs and PEL cells, and these altered metabolic processes are required for maintaining KSHV latency[13–15]. Among the KSHV-encoded products, the miRNA cluster decreases mitochondria biogenesis and induces aerobic glycolysis in ECs[16]. KSHV also induces glutamate secretion in ECs[17]. Nevertheless, in these studies, KSHV infection did not lead to cellular transformation. Thus, whether metabolic reprogramming is essential for KSHV-induced cellular transformation remains unknown.
In this study, we have discovered that KSHV suppresses aerobic glycolysis and oxidative phosphorylation in KSHV-transformed cells and this reprogramed metabolic pathway is essential for adaptation to glucose deprivation. These findings indicate that fine-tuning of metabolic pathways is essential for the proliferation and survival of cancer cells, particularly under stress conditions.
KSHV-transformed cells (KMM) proliferated significantly faster than their uninfected/untransformed counterparts (MM), and KMM but not MM cells lost contact-inhibition and formed colonies in softagar (Fig 1A and 1B)[8]. To determine the metabolic state of KSHV-transformed cells, we examined the consumption of glucose, the main carbon source for most normal and cancer cells. In normal cells, glucose flows through glycolysis and tricarboxylic acid (TCA) cycle to generate ATP and NADH with the latter further serving as a substrate for oxidative phosphorylation to produce additional ATP, a process that consumes oxygen[4]. However, many cancer cells have fast ATP production through a higher rate of aerobic glycolysis, resulting in higher rates of glucose consumption and lactate production despite the presence of oxygen[3]. To our surprise, KMM cells consumed significantly less glucose than MM cells did (Fig 1C). This effect was even more dramatic after taking into account of cell proliferation rate (Fig 1D). KMM cells also produced less lactate, and had lower levels of intracellular ATP and oxygen consumption compared to MM cells (Fig 1E–1H). Thus, despite their rapid proliferation, KMM cells consume less glucose, and have lower activities of aerobic glycolysis and oxidative phosphorylation.
Because KMM cells consumed less glucose, we postulated that they might not require glucose to support their proliferation. Indeed, glucose deprivation affected neither the proliferation nor colony formation of KMM cells in softagar while it caused proliferation arrest of MM cells (Fig 2A and 2B). Glucose deprivation caused G1 arrest, reduced BrdU incorporation, increased apoptotic cells and decreased the intracellular ATP level in MM but not KMM cells (Fig 2C–2F). Collectively, these results indicate that KSHV has reprogramed the cellular metabolic pathways following cellular transformation.
KMM cells are latently infected by KSHV and mostly express only viral latent genes/products including vFLIP, vCyclin, LANA and miRNAs[8]. To identify KSHV genes/products that mediate metabolic reprogramming, we generated MM cells latently infected by KSHV mutants containing individual deletion of vFLIP, vCyclin or 10 of the 12 pre-miRNAs (miR-K1-9 and 11)[10, 12, 18]. We were unable to obtain cells stably infected by a mutant of LANA because of its essential role in persistent infection[19, 20]. Under normal culture condition, deletion of vFLIP or the miRNA cluster reduced cell proliferation rates to levels similar to those of MM cells (Fig 3A). Deletion of vCyclin had no effect on cell proliferation though a slower rate was observed at contact-inhibited high cell density[12]. We further examined the metabolic states of these cells. Deletion of vFLIP or the miRNA cluster but not vCyclin increased glucose consumption, lactate production, intracellular ATP and oxygen consumption to levels close to or even higher than those of MM cells (Fig 3B–3E). Furthermore, deletion of vFLIP or the miRNA cluster sensitized the cells to glucose deprivation, causing cell proliferation arrest similar to MM cells (Fig 3F). While vCyclin mutant cells continued to proliferate upon glucose deprivation, they did so at a rate slower than that of KMM cells (Fig 3F). Consistently, glucose deprivation caused G1 arrest, reduced BrdU incorporation and increased apoptotic cells in cells of vFLIP and miRNA cluster mutants (Fig 3G–3I). Interestingly, the basal level of dead cells in the vFLIP mutant cells were higher than those of MM and KMM cells (25% vs 8% and 3%, respectively), and were further increased upon glucose deprivation, reaching as high as 95% (Fig 3I), which could be attributed to the oncogenic stress in KSHV-transformed cells[10] and vFLIP activation of the NF-κB[21, 22]. In contrast, glucose deprivation had minimal effect on cell cycle progression and BrdU incorporation of vCyclin mutant cells (Fig 3G and 3H); however, it increased apoptotic cells to a level similar to that of MM cells (Fig 3I), which might explain the slower proliferation rate of vCyclin mutant cells than KMM cells (Fig 3F). We further correlated the metabolic states of these cells with cellular transformation. Cells of both vFLIP and miRNA cluster mutants failed to form any colonies in softagar, a phenotype resembling that of MM cells (Fig 3J). vCyclin mutant cells formed significantly less and smaller colonies than KMM cells did in normal culture condition as previously reported[12] but continued to form colonies upon glucose deprivation albeit at a reduced efficiency (Fig 3J). Together, these results indicate that both vFLIP and the miRNA cluster mediate KSHV reprograming of metabolic pathways, contributing to KSHV-induced glucose-independent cell proliferation, survival and cellular transformation. While vCyclin can override contact inhibition to promote cellular transformation[12], it does not contribute to KSHV reprogramming of metabolic pathways.
To identify the mechanism of KSHV inhibition of aerobic glycolysis and oxidative phosphorylation, we examined changes of gene expression of key enzymes in the glycolysis pathway following KSHV transformation. All glycolysis enzymes either had minimal change or were upregulated (S1 Fig); hence, they were unlikely the candidates that mediated KSHV suppression of glycolysis. GLUT1 and GLUT3 directly mediate glucose uptake, which is the first step in the glycolysis pathway[4]. Downregulation of GLUT1 and GLUT3 was observed at mRNA and protein levels (Fig 4A–4C). Importantly, deletion of vFLIP or the miRNA cluster was sufficient to restore the GLUT1 and GLUT3 expression levels (Fig 4D and 4E). Interestingly, the mRNA levels of both GLUT1 and GLUT3 detected by reverse transcription quantitative real time PCR (RT-qPCR) and the protein level of GLUT1 detected by Western-blot were even higher in vFLIP mutant cells than in MM cells. The results of flow cytometry were inconsistent, which were probably due to the fact that the antibodies detected surface expression while RT-qPCR and Western-blot detected the total levels of mRNA and protein in cells, respectively. Deletion of vCyclin did not affect GLUT1 and GLUT3 mRNA expression levels but marginally increased their protein levels (Fig 4D–4F). These results indicate that vFLIP and the miRNA cluster mediate KSHV downregulation of GLUT1 and GLUT3.
To investigate the mechanism of KSHV downregulation of GLUT1 and GLUT3, we searched for a common pathway regulated by vFLIP and the miRNA cluster. Both vFLIP and the miRNA cluster activate the NF-κB pathway[21–23], and both are required for the maximal activation of the NF-κB pathway in KSHV-transformed cells[10]. Because knock down of RelA, a key component of the NF-κB complexes, is sufficient to inhibit the NF-κB pathway in KMM cells[10], we examined the effect of knock down of RelA on the expression of GLUT1 and GLUT3 (Fig 5A and 5B). Knock down of RelA significantly increased the protein and mRNA expression levels of both GLUT1 and GLUT3 (Fig 5A–5D). As previously reported[10], knock down of RelA slightly decreased cell proliferation of KMM cells but had no effect on MM cells (Fig 5E). Importantly, knock down of RelA increased glucose consumption and lactate production in KMM cells (Fig 5F and 5G). These effects were even more obvious when adjusted for cell numbers. In MM cells, knock down of RelA slightly increased the glucose consumption but had no detectable effect on lactate production.
To confirm the above results, we carried out pharmacological inhibition of the NF-κB pathway with two specific inhibitors, JSH-23 and BAY11-7082. Both inhibitors significantly induced the expression of GLUT1 and GLUT3 at mRNA and protein levels (Fig 6A and 6B). Interestingly, neither knockdown of RelA nor the NF-κB inhibitors fully rescued the expression of GLUT3 in KMM, suggesting that another pathway, besides the NF-κB pathway, might be involved in the inhibition of GLUT3 expression in KMM cells. Inhibition of the NF-κB pathway increased glucose consumption and lactate production in both MM and KMM cells (Fig 6C and 6D). Importantly, the increased glucose consumption and lactate production rates were correlated with reduced cell proliferation rates in both MM and KMM cells and a reduced efficiency of colony formation of KMM cells in softagar (Fig 6E and 6F). Consistent with these results, inhibition of the NF-κB pathway sensitized KMM cells to apoptosis and inhibited BrdU incorporation (Fig 6G and 6H). Thus, the NF-κB pathway promotes cell proliferation and cellular transformation at least in part by inhibiting the expression of GLUT1 and GLUT3 to limit the glucose consumption in KMM cells.
To confirm if downregulation of GLUT1 and GLUT3 mediated KSHV inhibition of glucose consumption and lactate production, we overexpressed GLUT1 and GLUT3 in MM and KMM cells (Fig 7A and 7B). Overexpression of GLUT1 or GLUT3 was sufficient to increase glucose consumption and lactate production in KMM cells but the results were inconsistent with MM cells, which might reflect their cell surface expression levels (Fig 7C and 7D). While overexpression of GLUT1 or GLUT3 neither significantly affected cell proliferation of both MM and KMM cells under normal culture condition nor altered the sensitivity of MM cells to glucose deprivation, it reduced cell proliferation of KMM cells upon glucose deprivation (Fig 7E). Consistently, glucose deprivation increased the number of apoptotic cells in KMM cells with overexpression of GLUT1 or GLUT3 (Fig 7F). As expected, MM cells were sensitive to glucose deprivation. Overexpression of GLUT1 or GLUT3 increased the basal number of apoptotic cells in MM cells, which was further increased upon glucose deprivation (Fig 7F). Interestingly, overexpression of GLUT1 or GLUT3 had no effect on cell cycle progression in KMM cells in either normal culture condition or in medium deprived of glucose (Fig 7G). Thus, the glucose transporters regulate cell survival rather than cell cycle progression under nutritional stress conditions in KSHV-transformed cells. Finally, overexpression of GLUT1 or GLUT3 in KMM cells slightly increased the sizes of some colonies but significantly reduced the number of colonies in softagar in normal culture medium, which was further reduced upon glucose deprivation (Fig 7H). Taken together, these results indicate that suppression of GLUT1 and GLUT3 expression confers KMM cells lower levels of glucose consumption and lactate production, and endow them the potential for glucose-independent cell proliferation, survival and cellular transformation.
To determine the mechanism mediating GLUT1 and GLUT3 regulation of the survival of KMM cells, we examined two main cell survival pathways AKT and NF-κB. Overexpression of GLUT1 or GLUT3 in KMM cells reduced the phospho-AKT level (Fig 8A). The AKT downstream targets phospho-NF-κB p65 and phospho-4EBP1 were also reduced (Fig 8A). Accordingly, we observed increased levels of autophagy, which is regulated by the AKT pathway, in these cells. Specifically, there were increased LC3-II/LC3-I ratio, more cells with the typical LC3 punctate staining and increased number of punctates per cell in KMM cells with overexpression of GLUT1 or GLUT3 (Fig 8A–8D). These results indicate that GLUT1 and GLUT3 impair the AKT and NF-κB survival pathways in KMM cells. To determine if AKT pathway mediated the activation of NF-κB pathway in KMM cells, we treated cells with an inhibitor of the AKT upstream activator PI3K. Interestingly, the PI3K inhibitor and glucose deprivation reduced the total and phosphorylated p65 levels (Fig 8E). The PI3K inhibitor and glucose deprivation synergized with each other to further reduce the total and phosphorylated p65 levels. As shown in Fig 7F, KMM cells with overexpression of GLUT1 or GLUT3 were sensitive to glucose deprivation with increased numbers of apoptotic cells (Fig 8F). Treatment with the PI3K inhibitor alone was sufficient to increase the numbers of apoptotic cells in these cells, and further sensitized them to glucose deprivation. While KMM cells were resistant to glucose deprivation (Fig 2E), treatment with the PI3K inhibitor alone was sufficient to increase the number of apoptotic cells in KMM cells overexpressing the vector control, or GLUT1 and GLUT3, and further sensitized them to apoptosis upon glucose deprivation (Fig 8F). Collectively, these results indicate that the resistance of KMM cells to glucose deprivation is likely due to their reduced GLUT1 and GLUT3 levels, resulting in the enhanced persistent activation of the AKT-NF-κB pathway.
We have so far demonstrated that KSHV promotes cell survival under nutrient deprivation by downregulating GLUT1 and GLUT3 to suppress aerobic glycolysis. To demonstrate the pathological relevance of these observations, we examined the expression of GLUT1 and GLUT3 proteins in human KS tumors on a tissue array by dual-color immunofluorescence staining (Fig 9A and 9B). The expression of GLUT1 and GLUT3 was evaluated using a modified Histo-score (H-score) as described in the Materials and Methods. We observed significant downregulation of GLUT1 and GLUT3 in LANA-positive cells compared to LANA-negative cells in the KS tumors as well as adjacent uninvolved tissues (Fig 9C and 9D). A total of 27 specimens were retained following GLUT1 staining (S2 Fig). Of the 22 specimens that had robust LANA signal (detection of > 10 LANA-positive cells), 20 (90%) had significantly downregulation of GLUT1 in the LANA-positive cells compared to LANA-negative cells. Three specimens had weak LANA signal (detection of < 10 LANA-positive cells). Of the 2 specimens that had no detectable LANA signal (normal skin tissues), we detected strong GLUT1 signal. Among the specimens that had LANA-positive cells, the average GLUT1 signal was already negatively correlated with the numbers of LANA-positive cells (r = -0.5351, P = 0.0233 in Fig 9E). A total of 22 specimens were retained following GLUT3 staining (S3 Fig). Of the 17 specimens that had strong LANA signal (detection of >10 LANA-positive cells), 13 (76%) had significantly downregulation of GLUT3 in the LANA-positive cells compared to LANA-negative cells. Two specimens had weak LANA signal (detection of 10 or < 10 LANA-positive cells). Of the 3 specimens that had no detectable LANA signal (normal skin tissues), we detected strong GLUT3 signal. Among the specimens that had LANA-positive cells, there was already a tend of negative correlation between the average GLUT3 signal with the numbers of LANA-positive cells albeit it had not reached statistical significance (r = -0.3932, P = 0.0573 in Fig 9F). Together, these results suggest that KSHV suppression of aerobic glycolysis is present in the KS tumors.
PEL is another malignancy associated with KSHV infection. Since primary PEL specimens are rare, we examined the expression of GLUT1 and GLUT3 in three PEL lines that are only infected by KSHV including BCBL1, BC3 and BCP1 cells (Fig 10A). Compared to BJAB, a KSHV-negative and EBV-negative Burkitt's lymphoma cell line, the expression of GLUT1 was downregulated in all PEL lines. However, the expression of GLUT3 had no obvious difference among the cell lines examined. As there is no appropriate control for the PEL cell lines, we examined BJAB cells infected by KSHV (BJAB-KSHV). KSHV infection downregulated the expression of both GLUT1 and GLUT3 in BJAB cells. BCBL1 and BC3 cells had slightly slower proliferation rates compared to other cell lines. However, by day 1 post-seeding, we observed slower glucose consumption rates in all KSHV-infected lines compared to BJAB (Fig 10C). By day 2 post-seeding, BJAB and BCP1 cells no longer had detectable glucose in the culture medium. We detected a higher level of lactate production by BJAB cells than those of all the KSHV-infected cell lines at day 3 post-seeding (Fig 10D). These results indicate that aerobic glycolysis is likely suppressed in PEL cells though further investigations are required to understand the metabolic reprogramming in the PEL cells, as well as how it might affect cell proliferation and survival.
We have shown that KSHV downregulates the expression of GLUT1 and GLUT3 to inhibit glucose uptake resulting in the suppression of aerobic glycolysis and oxidative phosphorylation. Under glucose deprivation condition, downregulation of GLUT1 and GLUT3 is required for optimal cell survival and efficient colony formation of KSHV-transformed cells in softagar. Significantly, we have detected downregulation of GLUT1 and GLUT3 in KSHV-infected cells in KS tumors, suggesting that suppression of aerobic glycolysis is likely important in these tumors. Mechanistically, KSHV inhibits the expression of GLUT1 and GLUT3 through activation of the NF-κB pathway by vFLIP and the miRNA cluster. Downregulation of GLUT1 and GLUT3 further maximizes KSHV activation of the AKT-NF-κB survival pathway resulting in enhanced cell survival and cellular transformation. These results have also revealed a negative feedback loop of the AKT-NF-κB pathway imposed by the glucose transporters, which is disrupted by vFLIP and the miRNA cluster (Fig 11).
To adapt to diverse conditions for growth, proliferation and survival, cancer cells must undergo reprograming of the metabolic pathways[4]. For fast growing cancer cells, glucose is diverted to aerobic glycolysis from the TCA cycle and oxidative phosphorylation to provide rapid supply of the energy and substrates for synthesis of macromolecules[3]. Surprisingly, we have found that both aerobic glycolysis and oxidative phosphorylation are suppressed in KSHV-transformed cells. It has been recognized that metabolic pathways must be tightly regulated to ensure the homeostasis of cells, particularly under stress conditions as overflow of the metabolic pathways could generate cytotoxic products[4]. Mesenchymal stem cells have an intrinsic high level of aerobic glycolysis compared to differentiated cells[24]. Thus, suppression of the glycolytic and oxidative phosphorylation activities by KSHV might avoid the overflow the pathways. Interestingly, it has been reported that aerobic glycolysis is upregulated in untransformed KSHV-infected ECs compared to the uninfected control cells[13, 16]. AKT hyperactivation by KSHV is responsible for GLUT1 membrane exposure in KSHV latent infection of a human monocytic cell line[25]. Whether these contradictory observations are due to different cell types or the states of cellular transformation remains to be determined.
The origin of KS tumor cells remains unclear. Our previous studies indicate that KS tumor cells could be derived from mesenchymal stem cells[9, 10]. In the KMM model, KSHV-induced cellular transformation is immediate upon KSHV infection and is dependent on the KSHV genome[8]. If this scenario exists in human KS tumors, downregulation of GLUT1 and GLUT3, and suppression of aerobic glycolysis should be readily present regardless of the status of acute or persistent infection in the tumors. In contrast, if KS tumor cells are derived from endothelial cells, enhanced aerobic glycolysis without suppression of GLUT1 and GLUT3 should be expected. However, our results so far indicate that GLUT1 and GLUT3 are downregulated in KS tumors (Fig 9). It remains possible that KS tumor cells are derived from endothelial cells and are transformed by KSHV in KS tumors but the cellular transformation phenotype has not been genuinely recapitulated in any of cell culture systems, which could explain the discrepancies between the in vivo and in vitro phenotypes. Further studies are required to clarify these contradictions.
The Warburg effect in a tumor is often measured by the avidity of fluorodeoxyglucose (FDG), a glucose analog. Low FDG avidity was detected in pulmonary and lymph node KS but not in skin KS[26, 27]. However, occult KS lesions were detected by FDG-positron emission tomography and computed tomography (FDG-PET/CT) in advanced KS[28, 29]. It was also reported that 55% (5 in 9) of KSHV-associated MCD patients who had cutaneous KS showed mildly hypermetabolic cutaneous abnormalities in FDG-PET[30]. Therefore, whether KS tumors, particularly early stage KS tumors, have increased glucose uptake remains unclear. It should be noted that KS lesions are highly heterogeneous, consisting of LANA-positive spindle tumor cells, and various LANA-negative cell types including vascular and lymphatic endothelial cells, macrophages, lymphocytes, plasma cells and red blood cells[31]. Recent studies have shown that some cancer cells have low levels of aerobic glycolysis but they induce aerobic glycolysis in neighboring stromal cells, which in turn provide fuels for the cancer cells and contribute to the overall Warburg effect in the tumors[32–34]. This model, termed “reverse Warburg effect”, explains some challenges of the Warburg effect and reveals the complex metabolic interactions of tumor and tumor microenvironments. The detection of Warburg effect in a fraction of the KS tumors could also be a reflection of the aerobic glycolytic activities of the stromal cells rather than the LANA-positive tumor cells. In fact, our results have clearly shown that the LANA-positive tumor cells express lower levels of GLUT1 and GLUT3 than the LANA-negative cells. Furthermore, advanced KS tumors are often composed of diverse genetic alterations, some of which could result in metabolic reprograming in these cells. Further research is warranted to delineate the molecular basis underlying the metabolic heterogeneity in KS tumors.
The findings that KSHV inhibits aerobic glycolysis are analogous to results of several studies on PKM2. As a tumor-specific glycolytic enzyme, PKM2 promotes the proliferation of cancer cells by inhibiting ATP generation and antagonizing the Warburg effect in some cancers[35–38]. This observation was initially regarded as counterintuitive but it has become clear that the shift of glucose to the TCA cycle and oxidative phosphorylation can generate metabolic intermediates for the synthesis of lipids, nucleotides and amino acids in addition to ATP[4, 39]. On the other hand, compared to cancer cells that have upregulated levels of PKM2, KSHV-transformed cells are distinct in that they have lower levels of intracellular ATP and oxygen consumption, reflecting the general lower activities of the TCA cycle and oxidative phosphorylation. Since hyperactivation of oxidative phosphorylation can generate reactive oxygen species[6], minimizing ATP production and oxygen consumption might allow KSHV-transformed cells to maintain a balanced cellular redox status.
The fact that KSHV-transformed cells consume less glucose than the untransformed cells despite their faster proliferation rates implies that they might utilize other carbon sources to support their proliferation. Recent studies have shown that in addition to glucose, cancer cells also utilize glutamine, one carbon amino acids or fatty acids to support their growth[4, 5]. Indeed, PEL cells and KSHV-infected ECs utilize fatty acids to support their proliferation and survival[14, 15]. Whether KSHV-transformed cells also depend on these carbon sources for proliferation remains to be investigated. Regardless the alternative carbon sources, such metabolic reprogramming enables KSHV-transformed cells to adapt to glucose-deprived condition. We have shown that under this condition, KSHV-transformed cells maintain normal proliferation and cellular transformation while the untransformed cells undergo arrest and apoptosis. Cancer cells, particularly those in solid tumors, often encounter stress conditions including nutrient deprivation[40, 41]. Glucose concentrations are frequently 3- to 10-fold lower in tumors than in normal tissues[40, 41]. Thus, the observed metabolic reprograming provides the advantage for KSHV-transformed cells to survive in a stress tumor microenvironment. These findings are consistent with results of another study showing that deficiency in PKCξ promotes the plasticity necessary for cancer cells to survive and proliferate in the absence of glucose by reprograming their metabolism[42]. In fact, up to 30% of cancers are FDG-PET-negative, indicating the lack of excessive glucose consumption in these cancers[43]. A number of cancers can survive therapies aimed at curtailing the supply or utilization of glucose by reprogramming their metabolic needs[44, 45]. As a result, such treatment often leads to increasing cancer aggressiveness[44, 45]. It is important to note that, under normal culture condition, KSHV-transformed cells are capable of consuming glucose, and maintaining aerobic glycolytic and oxidative phosphorylation activities albeit at lower levels than the untransformed cells. Thus, KSHV-transformed cells likely have optimized their metabolic pathways to adapt to different proliferation conditions.
Our results show that both KSHV vFLIP and the miRNA cluster are required for suppressing GLUT1 and GLUT3 expression by activating the NF-κB pathway. While overexpression of vFLIP or the miRNA cluster is sufficient to activate the NF-κB pathway[21–23], both are required for the maximal activation of the pathway in KSHV-transformed cells[10]. The mechanism by which vFLIP and the miRNA cluster synergize with each other to maximize the activation of the NF-κB pathway remains unclear.
The NF-κB pathway transduces crucial survival signals and is frequently activated in cancer. GLUT3 is a NF-κB transcriptional target[46–48] and RelA inactivation can lead to upregulation of GLUT1 and GLUT3[48]. Silencing of RelA in murine tumors that heavily rely on NF-κB activation resulted in increased activity of aerobic glycolysis, rendering these tumors especially sensitive to metabolic challenges including glucose deprivation[48]. Indeed, silencing RelA or inhibition of the NF-κB pathway leads to upregulation of GLUT1 and GLUT3, and increase of aerobic glycolysis in KSHV-transformed cells (Figs 5 and 6). These results illustrate NF-κB as a central regulator of energy homeostasis and metabolic adaptation in addition to its pro-survival function. Importantly, NF-κB activation by KSHV miRNAs is essential for the survival, proliferation and cellular transformation[10]. Similarly, vFLIP is also required for KSHV-induced cellular transformation (Fig 3J). Thus, by activating the NF-κB pathway, both KSHV vFLIP and the miRNA cluster play critical roles in KSHV-induced cellular transformation by regulating energy homeostasis and metabolic adaptation in addition to providing survival signal. Interestingly, overexpression of GLUT1 and GLUT3 suppresses NF-κB activation (Fig 8A). Thus, higher levels of GLUT1 and GLUT3 might suppress the NF-κB pathway in primary cells. By activating the NF-κB pathway, KSHV vFLIP and the miRNA cluster inhibit the expression of GLUT1 and GLUT3 in KSHV-transformed cells, which further enhance the AKT and NF-κB signaling. These results have established a NF-κB signaling loop negatively regulated by the glucose transporters, which is disrupted by KSHV vFLIP and the miRNA cluster (Fig 11).
The PI3K/AKT pathway is often hyperactivated in malignant cells and is known to promote the survival and proliferation of cancer cells[49]. We have shown that KSHV-transformed cells have hyperactivated AKT. Both KSHV GPCR (ORF74) and ORF-K1 can activate the AKT pathway[50–52]; however, KSHV-transformed cells are predominantly latent with minimal expression of these two viral proteins[8]. Both cellular and viral IL-6 can also activate the AKT pathway through the gp130 receptor[53, 54]. We have shown that stable overexpression of GLUT1 or GLUT3 suppresses AKT activation. Thus, KSHV downregulation of GLUT1 and GLUT3 can maximize AKT activation. While increased glucose uptake is known to enhance AKT activation[55, 56], the roles of glucose transporters in AKT activation are unclear. Our results indicate that glucose metabolism and glucose transporters might regulate AKT signaling by distinct mechanisms.
AKT is an important driver of the tumor glycolytic phenotype and stimulates ATP generation through multiple mechanisms[57, 58]. In particular, AKT1 simulates aerobic glycolysis by promoting the transcription and incorporation of GLUT1 into the plasma membrane[59–61]. Thus, suppression of aerobic glycolysis in KSHV-transformed cells where there is hyperactivation of AKT appears to contradict with these observations; however, this might be due to the intrinsic high level of aerobic glycolysis in the mesenchymal stem cells[24]. Similarly, hypoxia and HIF1α also enhance aerobic glycolysis[4, 5], and HIF1α is upregulated in KSHV-infected cells[62]. It is possible that NF-κB is the dominant pathway that regulates the expression of GLUT1 and GLUT3, and aerobic glycolysis in KSHV-transformed cells though further investigations are required to clarify these issues.
AKT is a known upstream regulator of NF-κB[49]. Indeed, chemical inhibition of the AKT pathway reduces the level of activated NF-κB, and sensitizes KSHV-transformed cells to glucose deprivation (Fig 8E and 8F). However, the activated AKT only partially accounts for the NF-κB activities as activation of the NF-κB pathway by both vFLIP and miRNAs is independent of the AKT pathway[21–23]. Nevertheless, AKT hyperactivation is essential for the survival and proliferation of KSHV-transformed cells, particularly under stress conditions such as glucose deprivation, and KSHV downregulation of GLUT1 and GLUT3 can maximize the AKT activation.
In summary, KSHV suppression of aerobic glycolysis and oxidative phosphorylation through inhibition of glucose uptake enables the adaption of KSHV-transformed cells to different proliferation and survival conditions. Our findings illustrate the importance of fine-tuning of the metabolic pathways in cancer cells, which could be explored for therapeutic application.
Rat primary embryonic metanephric mesenchymal precursor cells (MM), KSHV-transformed MM cells (KMM)[8], MM cells infected by KSHV mutants with a cluster of 10 precursor miRNAs deleted (ΔmiRs)[10], vFLIP deleted (ΔvFLIP)[18], vCyclin deleted (ΔvCyclin)[12] and 293T cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM; with 25 mM glucose, 4 mM L-glutamine and 2 mM sodium pyruvate) supplemented with 10% fetal bovine serum (FBS; Sigma-Aldrich, St. Louis, Mo) and antibiotics (100 μg/mL penicillin and 100 μg/mL streptomycin). Only MM and KMM cells at early passage (<15) were used for the experiments. For glucose starvation, cells were cultured in DMEM without glucose (with 4 mM L-glutamine and 2 mM sodium pyruvate), supplemented with 10% FBS (Sigma-Aldrich). PEL cell lines BCBL1, BC3 and BCP1, and EBV-negative Burkitt’s lymphoma cell line BJAB and KSHV-infected BJAB (BJAB-KSHV) were cultured in RPMI-1640 medium with 10% FBS. JSH-23 (inhibitor of NF-κB nuclear translocation), BAY 11–7082 (an inhibitor of IκBα phosphorylation) and LY294002 (PI3K inhibitor) were purchased from Sigma-Aldrich.
Softagar assay was performed as previously described[12]. Briefly, a total of 2x104 cells suspended in 1 ml of 0.3% top agar (Cat. A5431, Sigma-Aldrich) were plated onto one well of 0.5% base agar in 6 well-plates and maintained for 2–3 weeks. Colonies with a diameter of >50 μm were counted and photographed at 40× magnification using a microscope.
Cell cycle and BrdU incorporation were performed at the indicated time points as previously described[12]. Cell cycle was analyzed by propidium iodide (PI) staining. BrdU incorporation was performed by pulsing cells with 10 μM BrdU for 1 h and then stained with a Pacific Blue monoclonal antibody to BrdU (Cat. B35129, Life Technologies, Grand Island, NY). Apoptotic cells were detected by Fixable Viability Dye eFluor 660 staining (Cat. 650864, eBioscience, San Diego, CA) and with a PE-Cy7 Annexin V Apoptosis Detection Set (Cat. 88810374, eBioscience) following the instructions of the manufacturer. Flow cytometry was performed in a FACSCanto System (BD Biosciences, San Jose, CA) and analysis was performed with FlowJo (FlowJo, LLC, Ashland, OR).
Cells were seeded in 24-well plates, media were changed 24 h later and assays were carried out at the indicated time points in normal medium or in glucose-free medium (glucose starvation). Glucose and lactate concentrations were measured in the culture media using the Glucose Colorimetric/Fluorometric Assay Kit (Cat. K606-100, BioVision, Milpitas, CA) and the Lactate Assay Kit (Cat. MAK064, Sigma-Aldrich), respectively, according to the manufacturer’s instructions. Intracellular ATP levels were determined in cell lysates using the ATP Bioluminescent Somatic Cell Assay Kit (Cat. FLASC-1KT, Sigma-Aldrich) according to the manufacturer’s instructions. Oxygen uptake was measured in 24-well plates using a Seahorse XF24 Extracellular Flux Analyzer (Seahorse Bioscience, Billerica, MA). Cells were seeded at 2x104 cells per well and incubated overnight in normal growth medium. Next day, medium was changed to 8.3 g/L DMEM base medium at pH 7.4 supplemented with 200 mM GlutaMax-1, 100 mM sodium pyruvate and 32 mM NaCl in the presence of 25 mM glucose, and oxygen consumption was continuously measured.
The coding sequences of rat GLUT1 (GenBank accession no. NM_138827.1) and GLUT3 (NM_017102.2) were cloned into the FseI/PacI sites of pSMPUW-IRES-Bsd (Cat.VPK-219, Cell Biolabs, San Diego, CA) by PCR amplification to generate expression vectors named pSMPUW-IRES-Bsd-GLUT1 and pSMPUW-IRES-Bsd-GLUT3. The primers were 5’-AGTGGCCGGCCATGGAGCCCAGCAGCAAGA-3’ (forward) and 5’-AGTTTAATTAATCACACTTGGGAGTCAGCC-3’ (reverse) for GLUT1, and 5’-AGTGGCCGGCCATGGGGACAGCGAAGGTGA-3’ (forward) and 5’-AGTTTAATTAATCAGGCATTGCCAGGGGTCT-3’ (reverse) for GLUT3 with all the restriction enzyme sites underlined. All constructs were confirmed by direct DNA sequencing. The mCherry coding sequence from the pmCherry-C1 vector (Addgene, Cambridge, MA) and the rat LC3 coding sequence from the pEGFP-LC3 vector (Addgene) were cloned into the XbaI/BamHI and BamHI/NotI sites of the pCDH-hygro vector to generate a fusion mCherry-LC3 expression vector named pCDH-mCherry-(rat)LC3. To obtain the recombinant lentivirus, pSMPUW-IRES-Bsd overexpression plasmids were cotransfected with pMDLg/pRRE, pRSV-Rev and pMD2.G packaging plasmids into actively growing HEK293T cells by using Lipofectamine 2000 transfection reagent. Virus-containing supernatants were collected 72 hr after transfection and filtered to remove cells, and target cells were infected in the presence of 8 μg/mL polybrene. MM-Vector, KMM-Vector, MM-GLUT1, KMM-GLUT1, MM-GLUT3 and KMM-GLUT3 cells were selected with 10 μg/mL Blasticidin after transduction.
siRNA knock down of RelA was performed as previously described[10]. Briefly, the small interfering (si)RNA targeting Rat RelA (GenBank Access. No. NM_199267.2) transcript was designated siRelA (sense: GUGACAAAGUGCAGAAAGAUU; antisense: UCUUUCUGCACUUUGUCACUU). A scrambled oligonucleotide containing a random sequence was obtained from the manufacturer (Ambion, Thermo Fisher Scientific, Waltham, MA) and used as a control. Reverse transfection of siRNA duplex was performed using Lipofectamine-RNAiMAX (Invitrogen, Carlsbad, CA). Transfection was performed at a final concentration of 50 nM.
Total RNA was isolated with TRI Reagent (Cat. T9424, Sigma) according to the instructions of the manufacturer. Reverse transcription was performed with total RNA using Maxima H Minus First Strand cDNA Synthesis Kit (Cat. K1652, Thermo Fisher Scientific). qPCR analysis was performed on Eppendorf Real Plex using KAPA SYBR FAST qPCR Kits (Cat. KK4602, Kapa Biosystems, Wilmington, MA). The relative expression levels of target genes were normalized to the expression of internal control genes, which yielded a 2-ΔΔCt value. All reactions were run in triplicates. The cycle threshold (Ct) values should not differ more than 0.5 among triplicates. Rat β-actin was used as an internal control. The primers were 5’-GCGAGCTCTTTGAATGTGTG-3’ (forward) and 5’-GGCTCAGGTCCTTCACGTAG-3’ (reverse) for GLUT1, 5’-ATGTTGGCCAGTCAAGTTCC-3’ (forward) and 5’-CTGTCACCTCTGGGAGCAG-3’ (reverse) for GLUT3, and 5’-GCAGGAGTACGATGAGTCCG-3’ (forward) and 5’-ACGCAGCTCAGTAACAGTCC-3’ (reverse) for β-actin.
Total cell lysates were separated in SDS-polyacrylamide gels, electrophoretically transferred to nitrocellulose membranes (GE Healthcare, Piscataway, NJ). The membranes were incubated sequentially with primary and secondary antibodies. The signal was developed using Luminiata Crescendo Western HRP substrate (cat. WBLUR0500, EMD Millipore, Billerica, MA). The antibodies used for Western blot include rabbit monoclonal antibodies (mAbs) for GLUT1 (cat. ab115730, Abcam, Cambridge, MA), phospho-AKT (Thr308) (cat. 2965, Cell Signaling Technology, Danvers, MA) and NF-κB p65 (cat. 8242, Cell Signaling Technology); rabbit polyclonal antibodies against GLUT3 (cat. ab41525, Abcam), phospho-4E-BP1 (Ser65) (cat. 9451, Cell Signaling Technology) and AKT (cat. 9272, Cell Signaling Technology); and mouse mAbs for LC3 (cat. CTB-LC3-1-50, COSMO BIO CO., Tokyo, Japan), phospho-NF-κB p65 (Ser 536) (cat. 3036, Cell Signaling Technology) and β-tubulin (7B9, Sigma).
Cells fixed with 80% methanol (5 min) were permeabilized with 0.1% PBS-Tween for 20 min. The cells were then incubated in PBS containing 10% normal goat serum and 0.3 M glycine to block non-specific protein-protein interactions followed by GLUT1 or GLUT3 antibody (ab115730 or ab41525, respectively) at 1/500 dilution for 30 min at room temperature. The secondary antibody used was Alexa conjugated at 1/2000 dilution for 30 min at room temperature. Flow cytometry was performed with a FACS Canto II flow cytometer and analyzed with FlowJo. All runs included a control without the primary antibody.
KMM, KMM-GLUT1 and KMM-GLUT3 cells were seeded on 24-well culture plate that contained 12 mm diameter round glass coverslips (2x104 cells per well). After infection 48 h, cells were fixed with 4% paraformaldehyde in PBS for 15 min at room temperature and washed with PBS. Samples were incubated with 0.5 μg/ mL 4-, 6-diamidino-2-phenylindole (DAPI) in PBS for 1min, then were mounted in FluorSave Reagent (Calbiochem, San Diego, CA). Samples were imaged with laser-scanning confocal microscopy (Nikon Eclipse C1).
Formalin-fixed, paraffin-embedded tissue microarray consisting of tissue specimens from healthy subjects and patients with KS were obtained from the AIDS and Cancer Specimen Resource (ACSR). Sections were de-paraffinized in xylene, rehydrated through graded ethanol, quenched for endogenous peroxidase activity in 3% hydrogen peroxide in methanol for 10 min and processed for antigen retrieval by microwave heating in 1 mM EDTA at pH 8.0. Immunostaining was performed using an anti-LANA antibody LN35 (cat. ab4103, Abcam) and an anti-GLUT1 antibody (ab115730, Abcam) or an anti-GLUT3 antibody (cat. sc-30107, Santa Cruz Biotechnology, Santa Cruz, CA) antibodies, followed by Alexa-488 and Alexa-568 conjugated secondary antibodies (Thermo Fisher Scientific). Nuclei were stained with 4’,6-diamidino-2-phenylindole (DAPI). The stained cells were viewed under a confocal fluorescence microscope with a 60x objective. Tissue sections without incubating with primary antibody were used as negative controls. For each specimen, three images of representative areas were acquired and a total of 200 to 500 cells were counted unless stated otherwise. The scoring of the expression of GLUT1 and GLUT3 was performed using a modified Histo-score (H-score), which included a semi-quantitative assessment of both fraction of positive cells and intensity of staining. The intensity score was defined as no staining (0), and weak (1), moderate (2), or strong (3) staining. The fraction score was based on the proportion of positively stained cells (0%-100%). The intensity and fraction scores were then multiplied to obtain H-score, which ranged from 0 to 3 and represented the levels of GLUT1 and GLUT3 expression.
Data were expressed as the mean ± standard error of the mean (s.e.m.) from at least three independent experiments, each with three repeats unless stated otherwise. The differences between groups were analyzed using Student’s t-test when two groups were compared and using one-way ANOVA when more than two groups were compared unless otherwise noted. Correlation was determined using Spearman’s correlation coefficient. Statistical tests were two-sided. A P < 0.05 was considered statistically significant. Statistical symbols “*”, “**” and “***” represent P-values < 0.05, < 0.01 and < 0.001, respectively, while “NS” indicates “not significant”. All analyses were performed using the GraphPad Prism program (GraphPad Software Inc., San Diego, CA).
|
10.1371/journal.pntd.0005966 | Programmatic factors associated with the limited impact of Community-Directed Treatment with Ivermectin to control Onchocerciasis in three drainage basins of South West Cameroon | The CDTI model is known to have enhanced community participation in planning and resource mobilization toward the control of onchocerciasis. These effects were expected to translate into better individual acceptance of the intervention and hence high Treatment Coverage, leading to a sustainable community-led strategy and reduction in the disease burden. A survey revealed that after 10–12 rounds of treatment, prevalence of onchocerciasis was still high in three drainage basins of South West Cameroon and transmission was going on.
We designed a three (3)-year retrospective (2012, 2013 and 2014), descriptive cross-sectional study to explore the roles of operational challenges in the failure of CDTI to control the disease as expected. We administered 83 semi-structured questionnaires and conducted 12 in-depth interviews with Chiefs of Bureau Health, Chiefs of Centers, CDDs and Community Heads. Descriptive statistics was used to explore indicators of performance which were supported with views from in-depth interviews.
We found that community participation was weak; communities were not deciding time and mode of distributions. Only 6 (15.0%) of 40 Community Drug Distributors reported they were selected at general community meetings as required. The health service was not able to meet and discuss Community-Directed Treatment with Ivermectin activities with individual communities partly due to transportation challenges; this was mostly done through letters. Funding was reported to be inadequate and not timely. Funds were not available to conduct Community-Self Monitoring after the 2014 Mass Drug Administration. There was inadequate health staff at the frontline health facility levels, and some Chiefs of Center reported that Community-Directed Treatment with Ivermectin work was too much for them. The mean operational Community Drug Distributor-population ratio was 1 Community Drug Distributor per 317 populations (range: 194–464, expected is 1:250). Community Drug Distributor attrition rate was 14% (2012), 11% (2013) and 12% (2014) of total Community Drug Distributors trained in the region. Lack of incentive for Community Drug Distributor was primary reason for Community Drug Distributor attrition. Number of Community Drug Distributors trained together by health area ranged from 14 to 127 (mean ± SD = 51 ±32) with duration of training ranging from 4–7 hours (mean ± SD = 5.05 ± 1.09). The trainings were conducted at the health centers. Community Drug Distributors always conducted census during the past three distributions (Mean ± SD = 2.85 ± 0.58). Community-Self Monitoring was facing challenge. Several of the community heads, Chiefs of Bureau Health and Chiefs of Center agreed that Community-Self Monitoring was not being carried out effectively due to lack of incentives for monitors in the communities.
Inadequate human resource, funding issues and transportation challenges during distribution periods reduced the ability of the health service to thoroughly sensitize communities and supervise CDTI activities. This resulted in weak community understanding, acceptance and participation in the process. CDTI in our study area did not achieve sustainable community-led campaign and this may have led to the reduced impact on Onchocerciasis.
| River blindness is caused by a very tiny, thread-like worm. The disease is better controlled when affected communities are included in the planning and carrying out of distribution of Ivermectin used to treat the disease. For a community to be able to prevent people from getting this disease, members must take Ivermectin once or twice a year, continuously for about 20 years. Hence, the organization in charge of controlling river blindness (African Programme for Onchocerciasis Control–APOC) decided that when a control programme is started in a community, the community must be involved and assisted to take full charge of the programme so that within 12 years the community can sustain the distribution of Ivermectin for as long as necessary. This community directed strategy prevented river blindness in many communities. However, after 10–12 years of implementation, studies found that river blindness largely persists in communities in three drainage basins in South West Region of Cameroon. This paper discussed the operational challenges that the programme may have faced in these areas.
| The Community-Directed Treatment with Ivermectin (CTDI) approach adopted by the African Programme for Onchocerciasis Control [1] accelerated progress towards control of Onchocerciasis through increased community participation and higher treatment coverage of affected populations [2,3]. The CDTI model is known to be cost-effective, fosters high level of community acceptance and empowerment [4]. The success of the CDTI process has been attributed to the nature of collaboration among the stakeholders. The stakeholders involved in the CDTI process are some 146 000 affected communities, health services (ministries) of participating countries, Non-Governmental Developmental Organizations (NDGOs) and external donor countries and organizations [5–7].
The partners have specified roles relating to planning, resourcing, drug distribution, monitoring and evaluation, social mobilization and training of implementers: The NDGOs and external donors are responsible for providing technical/operational support, funds for logistics, training and monitoring as well as ensuring that Ivermectin gets to national offices of the programme [6]. The governments, through the ministries of health are responsible for approaching and sensitizing communities, training of implementation staff, supervision of the process and ensuring an adequately motivated staff. APOC prescribed the required number of training, supervision, monitoring and review meetings to be carried out at each level of the health service annually [7]. The communities are responsible for selecting community drug distributors (CDDs) and deciding how to motivate them, plan when and how to distribute Ivermectin, distribute the drug and ensure members adhere to treatment, conduct Community Self-Monitoring (CSM) and report to the health service [8].
Central to the CDTI process is the maximized structural community participation [1,9,10]. To win the trust of communities and build strong partnership however, the health service must carry out series of advocacy meetings (Fig 1) [8]. This requires adequate and well-resourced health staff at the front line health facility levels.
Ivermectin mass drug administration process involves annual training of staff, census to determine drug needs of communities, distribution of the drug, conduction of CSM and reporting to the health services, the health service providing feedback to communities in addition to supervision and review meetings. These activities mandate adequate and timely availability of human, material and financial resources to ensure successful implementation. This manuscript reviewed the implementation process during the 2012, 2013 and 2014 distributions.
Cameroon health system is coordinated by a Ministry of Public Health headed by a Minister, who works through ten (10) Regional Delegations of Public Health headed by medical doctors. A region is divided into Health Districts, headed by district medical officers who are also medical doctors; there are eighteen (18) health districts in the South West Region. A health district is further divided into health Areas. A health area comprises about 5–10 communities that are served by a health center which may have smaller health posts in addition. The health center is usually headed by a senior nurse called Chief of Center. A nurse in charge of a health post is called Chief of Post. Health decisions at this level are taken by a Health Area Committee. This is the level where the health system interacts with communities, also referred to as frontline health facilities. In CDTI, CDDs collect Ivermectin from here and report directly to health personnel at the frontline health facilities.
A community or village in Cameroon is headed by a chief or community head. The community is demarcated into what is known as Quarters, comprising of about twenty (20) houses depending on the size of the community. Each quarter has a head who takes decisions with household heads of all households in the quarter. The quarter heads and community head forms the village council. They oversee the work of CDDs, other developmental activities and resources in the community. Decisions are disseminated by quarter heads through household heads or through an announcer called ‘town crier’. However, the influence of these traditional rulers varies among the tribes and is generally eroding away very fast.
The implementation of CDTI in Cameroon started in 1998. There are currently 15 CDTI projects in Cameroon (the third largest after Nigeria and DR Congo). The two CDTI projects, South West I & II situated in the South West Region are responsible for reaching some 1,367 endemic communities with Ivermectin [8]. They were approved in 1998 and 1999, and became operational in 1999 and 2000 respectively. The process started in 1999 in communities of the Meme and Mungo river drainages, and in the year 2000 in communities of the Manyu river drainage (South West Regional Delegation of Health Database).
South West CDTI Project I covers the Meme and Mungo drainage basins. It achieved 100% geographic coverage from 2001–2014, except in 2008 (97%) and 2009 (98.33%). Therapeutic coverage rose from 32.56% in 2001 to 82.83% in 2010, and has since achieved at least 81% as at 2014. The South West CDTI Project II, which covers the Manyu drainage basin, achieved a geographic coverage between 95% - 100% from 2001–2014. Therapeutic coverage rose from 37.2% in 2001 to 83.7% in 2009, the project has since achieved closely same coverage as at 2014.
According to ONCHOSIM predictions, the outcome of elimination of onchocerciasis depends on precontrol endemicity level, frequency of MDA and treatment coverage (TC) achieved [11]. Given the precontrol hyperendemic (60–98% prevalence) levels in communities in our study area, CDTI projects here must achieve and sustain annual TCs above 80% in order to bring the disease under control in about 18 years [11]. Kim et al. predicted that the elimination scenario for onchocerciasis is feasible by 2028 in some areas but could go beyond 2045 in countries with operational challenges [12]. Certain challenges have been identified with the CDTI process. These include: maintaining timely drug-collection mechanisms; integrating CDTI with existing primary-healthcare services; strengthening local health infrastructure; achieving and maintaining an optimal treatment coverage; establishing and scaling up community self-monitoring; designing and implementing operational research locally; ensuring the adequacy of community-directed distributors; increasing the involvement of local non-governmental developmental or community-based organizations in the programme; achieving financial sustainability; implementing equitable cost-recovery systems; and engaging in effective advocacy among the stakeholders, especially with the affected communities [13].
After twelve (12) annual MDA with Ivermectin in communities in three drainage basins in the South West Region of Cameroon, a situation analysis of onchocerciasis through entomological and parasitological (epidemiological) surveys revealed that prevalence and transmission did not reduce as predicted (Prevalence of microfilaria: Meme drainage basin– 52.7%, Mungo drainage basin– 41.0%, Manyu drainage basin– 33.0%) [14,11]. This paper explored the possible roles operational challenges could have played in the failure of CDTI to control the disease in these areas, by examining the implementation practices over the past three distributions (2012–2014).
We carried out this study to document the programmatic factors associated with the limited impact of CDTI to control onchocerciasis in three drainage basins. We also wanted to assess the community perceptions as well as their level of participation in the CDTI process.
We hypothesized that stakeholders were not able to fully implement the CDTI protocols and hence communities did not adhere adequately to treatment.
The study was carried out in five (5) health districts located in three forest drainage basins of South West Region of Cameroon. These drainage basins receive about eight (8) months of rainfall and hence create long periods of favorable ecological environment for the insect vector (blackflies). About 90% percentage of the road network in the drainage basins were not tarred. The poor road network, coupled with long rainy seasons pose transportation challenges to health supervision teams.
The respondents were the South West Regional Coordinator for Neglected Tropical Diseases (NTDs), the Chiefs of Bureau Health (CBH) of five (5) Health Districts (1 female and 4 males), 12 Chiefs of Centers (COC) (12 Health Areas; 8 females and 4 males), 40 CDDs (7 females and 33 males; 30 farmers; 34 married; all had at least primary school leavers’ certificate) and 24 Community Heads from communities across the five health districts. Once a community was included in the study, its head and CDDs were sampled. The COCs and CBHs of the health areas and health districts of the community also become part of the respondents. Since communities were randomly selected, bias was reduced.
A 3-three year retrospective, cross-sectional and descriptive approaches were used to explore possible weak links in the collaboration of the stakeholders and context specific factors that may be acting as implementation barriers to the CDTI process in the these drainage basins.
Data was collected with standard, semi-structured questionnaires and guides for in-depth interviews developed with reference to prescribed functions of stakeholders in the CDTI process [4,8,15]. Twelve (12) in-depth interviews were held with a community head, CDD, COC, and CBH from each drainage basin and also with the regional coordinator for NTDs. The interviews were held in English, recorded with a 4GB capacity Xgenx digital voice recorder (GDVR-901) and were transcribed into Microsoft word text format.
Performance of CDTI implementers were measured as number of training, census, supervision, CSM, reporting and review meetings conducted during the past three distributions (2012–2014). Adequate and timely arrival of funds and drugs, adequate health workers, amount of sensitization done, community participation as well as physical challenges to carrying out of the process were also assessed. Data collection was done from May to July, cleaning and analysis was done during August and September, 2015.
Quantitative information was entered into a template created in Epi info version 3.5.4. The data was imported into excel and cleaned. It was then exported to SPSS version 20 and analyzed. Descriptive measures were used to explore indicators of good performance of the CDTI process. All statistical differences were considered significant at p < 0.05.
The in-depth interviews were transcribed verbatim and translated where necessary into English by a trained transcriber into separate Microsoft word documents, through the familiarization process. The transcripts were imported into Atlas.Ti, transformed into rich text format (rtf) and given file names recognizable by Atlas.Ti. The framework approach was used to screen the data following the development of an initial thematic framework. The transcripts were then coded following the emerging themes previously identified to retrieve meaningful views expressed by our informants. Frequently expressed views were used to explain and to compare with observed trends in the quantitative data.
Ethical approval was obtained from the Institutional Review Board, Faculty of Health Science, University of Buea. Administrative authorization was obtained from the South West Regional Delegation of Public Health. Administrative authorizations were also obtained from health districts after thorough review of the study protocol. The objectives, importance and ethical provisions of the study were explained to the respondents and informed verbal consent was obtained before the questionnaires and in-depth interviews were administered.
The communities were participating in CDTI mainly by selecting CDDs and individual members were giving cash incentive to CDDs. All the health personnel interviewed stated that the communities were yet to appropriate the CDTI concept. One CBH stated: “APOC and ministry of public health and the rest I can grade them to be good, not very good. Is the community that is bringing weakness in the partnership”.
Distribution activities were planned top-down. All 24 community heads reported they did not participate in deciding the period (time of year) and the mode of distributing Ivermectin; whether by door-to-door or at a central location. This was confirmed by the statement of another CBH: “I can say we have not really […] involved them in the planning and when you go and plan something like that and come to tell people without involving them, they become reluctant to participate”. The communities were however given the opportunity to choose their CDDs and decide how to motivate them. Out of 40 CDDs sampled, 25 (62.5%) reported having been selected at a meeting of community leaders, 6 (15.0%) at a meeting of all community members and 3 (7.5%) by health workers (Fig 2).
The primary media used in creating awareness of CDTI activities among the communities included TV/Radio announcements at the regional level; banners, posters, badges (T-shirts and caps for CDTI personnel and key community members) at district and health area levels; at community levels, it was done through CDDs, town criers and religious groups. Review of the sensitization records showed that over the past three distributions (years), 2 TV announcements (alongside news scrow during the distribution weeks), 131 radio announcements, 9 872 posters and 11,300 fliers (smaller posters) were used to sensitize communities in our study area which had a population of 546 136 (Table 1). The posters and fliers were normally pasted at key public places such as health centers or given to CDDs and other key community members.
Only three (3) out of twenty-four (24) community heads reported having been completely involved in the sensitization of their members (Fig 3). Instead of the personal visit to the head of communities and their leaders, health areas were informing them about distribution through letters. One community head reported: “I only hear that […] is where they deposit the drugs and other health units go for it. That’s all the information that I know, but for the […] other issue that they will call or they bring the drugs and the community now […] decides on how to distribute it, to me we have never have such information”. The CBHs and COCs also admitted that the communities have not been fully sensitized on their roles. The view of one COC is presented as follows: “I just know, I think that we have not really sensitized them well, we have not really involved them well. They do not know their duties […]. And so I think that the weakness is our own”. One CBH also pointed out: “The second thing I will say is the failure of the health personnel to sensitize the community members to actually understand what their role is as partners”. The scores by the community heads based on their knowledge of communities’ roles as stakeholders in CDTI are shown in Fig 4.
The goal of APOC was to establish community-led CDTI projects that are sustainable within twelve (12) years after start of the project in a particular jurisdiction [4]. This requires full participation of communities in the design of implementation strategies, sensitization of its members and mobilization of human and financial resources, CDTI works best by the principle of active social participation [1,9,10]. Exploring the operational challenges faced by CDTI projects in our study communities revealed that a top-down approach to planning of CDTI activities may have resulted in low community participation. As observed by Gyapong et al., active community participation fosters better adherence to treatment [17]. Active community participation could have guided the program to organize MDAs at periods when the community members are most available to optimally participate. Practices such as communities not selecting CDDs through general community elections has also been reported in other CDTI projects [9,14]. In another study, community-member participation in CDD selection was observed to be a positive predictor of adherence to treatment. [18].
Studies have demonstrated that when communities are informed and empowered, they are able to deliver health interventions that are more complex than Ivermectin MDA [19,20]. The inability of the health service to engage and thoroughly educate communities on their roles in the CDTI process therefore, constituted a huge implementation barrier. Hence, communities lacking knowledge of their roles regarding planning, resourcing and monitoring of the process, and perceiving CDTI as exclusive responsibility of the government after 16 years of MDA is blamable on lack of sensitization. Lack of sensitization may also have resulted in poor knowledge of the causes of onchocerciasis and the dynamics of treating using Ivermectin among community members and hence they may not have adhered optimally to treatment. It was important that the health service closely followed APOC’s guidelines for mobilizing communities for CDTI through series of meetings with the people in their respective communities instead of using letters [8]. In an APOC midterm evaluation conducted in Cameroon, implementers asserted that more health sensitization was needed to achieve the revised target of 80% therapeutic coverage; the report added that sensitization was more critical in the face of 25 severe adverse effects plus one death and 12 severe adverse effects plus two deaths in 2008 and 2009 respectively and also to remove strong suspicion among CDDs that health staffs were siphoning their cash incentives [21]. In the face of the current goal of eliminating onchocerciasis as a disease of public health concern by 2025, sensitization must also be strategized to actively identify systemic non adherers to be educated and properly treated since they may hold a reservoir of the infection in the population [22]. Again, sensitization through television and radio broadcastings, posters and fliers must be scaled up and augmented with home visits by CDDs to announce and educate household members on Ivermectin distribution.
Financial issues may have also affected the implementation of CDTI in the area. The government and other donor agencies must endeavor to meet their budgetary allocations to the programme to ensure smooth implementation. The subcontracting arrangements between funding partners possibly introduced complexities into funding procedures. The programme could have benefited from a more streamlined funding process that minimizes delays and ensures adequate funding for all aspects of implementation. The cost recovery approach adopted by Cameroon from the beginning of CDTI may have hampered the ability of health staff to persuade the communities to accept their ‘new responsibility’ of having to provide cash incentive to CDDs [22]. Clearly, inappropriate community support to CDTI derailed APOC’s objective of establishing a community-led distribution in our study area, possibly because the principles that foster community ownership, empowerment and sustainability were not applied [21,23–26]. It therefore would be helpful for the health services to engage the communities to explore all possible alternatives of providing incentives to implementers at the community level. Another issue that may have challenged CDTI activities here is inadequate health workers. Integration of NTD programmes has been generally reckoned as cost effective. However, this could mean increased work load for implementers at frontline health facility levels and unbearable costs to communities. Given the human resource limitations at the frontline health facilities in our study area, a health staff having to implement several health interventions including primary health care services, coupled with frequent travels for meetings and workshops posed serious challenges to the CDTI process [21]. Consequently, Inadequate health staff at the frontline health facility level negatively influenced the way communities were approached and sensitized, the quality of training given to CDDs and CSM facilitators, supervision and monitoring of distribution. Lack of close supervision of distribution by health staff was also attributed to human resource gap in another study carried out in some communities of the Mungo river drainage [27]. Inadequate health staff partly explain why instead of training CDDs and CSM facilitators in their own communities, they were trained together (both new and old) in large numbers at health centers. An interesting dimension to the human resource limitation was high CDD attrition. The midterm evaluation conducted in Cameroon before this study reported a better CDD: population ratio of 1:123 (1:86–1:300) [21], indicative that the situation may have gone worse. High CDD attrition and hence turnover necessitated annual selection and training of new CDDs for replacements. The high attrition may have been caused by CDDs not receiving cash incentive for CDTI work but receiving such on other interventions using the CDTI approach. Again, organizing MDAs during the peak of farming seasons may have been a key contributory factor to CDD attrition. This partly explains why most of the CDDs in our study communities were missed out during the data collection. The problem may have been further compounded by inability of health staff to clearly explain modes of CDD remuneration under the CDTI concept. However, some CDDs expressed willingness to continue their work in spite of lack of incentives. They expressed feeling important as they became involved in community-level decision making and are sometimes referred to as ‘doctors’ by community members. In the midterm evaluation report, health staff believed more training of implementers would be needed in order to achieve the revised coverage of 80% [21]. The inclusion of hypoendemic areas in treatment to meet current goal of WHO to eliminate ochocerciasis by 2025 implied training of more implementers to meet the increased workload.
Another possible setback faced by CDTI in these areas was the inability of the programme managers to carry out distribution in the dry months of the year. Initial meetings with communities, sensitization, training of CDDs and CSM facilitators, supervision and monitoring of distribution are activities that heavily depend upon efficient transportation. Most communities in these areas become hard to reach during the rainy season due to poor nature of roads. Fares usually double and health staff spend hours in transit or simply avoided these areas. Transportation challenges has been noted to have adversely affected CDTI operations in other jurisdictions [21,27]. These physical and health system effects created barriers to effective implementation of annual distributions and may have actually led to low therapeutic coverage of the population [28,29].
The key internal mechanism put in place to identify operational challenges with the CDTI is Community Self-Monitoring (CSM). However, CSM was not a common practice among the study communities. This was likely due to inability of the health service to mobilize well-motivated CSM facilitators and to develop strong partnership with the communities in this regard. Monitoring of CDTI activities is very key since programme failure would result in onchocerciasis recrudescence and hence rolling back of the socioeconomic and health gains made against the disease over the years [21,30]. Understanding the weaknesses of CDTI would be necessary in maintaining its relevance in the current NTD landscape.
The CDTI approach gave huge impetus to onchocerciasis control on the continent through increased geographic and therapeutic coverage, and hence the possibility of using it to deliver other health interventions has since been explored to some extent [2]. The approach thus looks promising to current NTD control efforts on the continent, more importantly, given the fact that most onchocerciasis endemic countries in Africa are also faced with serious health professional crisis. Especially, among the most affected, rural and poor communities. The use of CDDs and other community health workers therefore, would be crucial in bridging the health professional gaps of these countries. Again, the CDTI approach may become a key strategy in ESPEN’s integrated approach to controlling the five NTDs amenable to preventive chemotherapy (namely: onchocerciasis, lymphatic filariasis, schistosomiasis, soil transmitted helminthes and trachoma), given the fact that the approach proved efficient in delivering similar interventions, including distribution of Long-Lasting Insecticide-Treated Bed Nets, Vitamin A and home management of malaria [1,31]. Also, the CDTI approach holds the potential to become more relevant in delivering more other interventions in the near future. Following the discovery of thermostable dog rabies vaccine, studies are now ongoing to ascertain the prospects of a scaled-up community-led administration [32]. Further, the CDTI approach is in perfect harmony with international calls for inclusion of populations in the health decision making process. The objective to create a lean NTD control entity, as a result of which ESPEN was created, is most likely possible only at regional and country levels. ESPEN would need the contribution of the huge CDD work force within communities to be able to sustain the momentum and health gains achieved by APOC [31]. Finally, the approach would remain relevant into the foreseeable future because it is probably the most cost efficient. Prior to inception of CDTI, the World Bank, UNICEF, UNDP, WHO special programme for Research and Training in Tropical Diseases (TDR) and OCP jointly conducted a trial study in five countries across West, Central and Eastern Africa to compare the effectiveness of vertical campaigns to a community-directed approach. The study demonstrated that coverage, cost-effectiveness, community acceptance and empowerment were better for CDTI as compared to vertical campaigns [4]. Also, structural community participation prepares communities better for subsequence sustainability and continuation of intervention as opposed to vertical approaches.
Since parasitological and entomological surveys were not done in this study, it is difficult to conclude that implementation challenges were responsible for the inability of CDTI to control onchocerciasis in the study area as expected. The recent parasitological and entomological studies conducted in our study area cited vector competence, favorable breeding conditions for the vector as possible factors responsible for the persisting transmission and high prevalence, though this paper did not compare coverage data from other parts of the country, its findings would likely be same for other endemic areas with similar transmission potentials and geographic characteristics. In such terrains, there would probably be a need for an additional intervention like vaccination when it becomes available, or moving from annual to biannual treatment. The current approach of Test and Treat being used in Cameroon may increase the health gains if it is followed comprehensively.
Certain critical weaknesses existed in the implementation process of CDTI in our study area. This included weak community participation towards planning of CDTI activities, sensitization of community members, resource mobilization and monitoring of the process. This may have actually led to low adherence to Ivermectin treatment among community members.
Also, inadequate staff at the frontline health facilities, funding issues and transportation challenges derailed efforts of the health service towards implementing adequate training, supervision and monitoring of the process.
Future studies should combine entomological, epidemiological, as well as data on the programme performance in order to identify and better explain the factors responsible for certain control outcomes. Also, it would be interesting to examine how closely figures for indictors such as therapeutic coverage, reported by CDTI projects reflect what is actually achieved in the populations.
The guidelines on engaging and mobilizing communities for CDTI activities must be comprehensively followed. This should reflect a change in perception of health staff of communities as mere beneficiaries of the process to an attitude that regards the importance of the roles of communities. The roles of communities should be clearly communicated to them, and they should be trained to totally assume those roles. Planning of CDTI activities must begin from community level, with increased community participation in the planning, supervision and monitoring of the process. Again, integration of CDTI with other primary health care deliveries must be quickly broadened to cover many interventions as possible. However, this must be done with caution in order not to dilute the CDTI concept.
|
10.1371/journal.ppat.1001319 | Sheep and Goat BSE Propagate More Efficiently than Cattle BSE in Human PrP Transgenic Mice | A new variant of Creutzfeldt Jacob Disease (vCJD) was identified in humans and linked to the consumption of Bovine Spongiform Encephalopathy (BSE)-infected meat products. Recycling of ruminant tissue in meat and bone meal (MBM) has been proposed as origin of the BSE epidemic. During this epidemic, sheep and goats have been exposed to BSE-contaminated MBM. It is well known that sheep can be experimentally infected with BSE and two field BSE-like cases have been reported in goats. In this work we evaluated the human susceptibility to small ruminants-passaged BSE prions by inoculating two different transgenic mouse lines expressing the methionine (Met) allele of human PrP at codon 129 (tg650 and tg340) with several sheep and goat BSE isolates and compared their transmission characteristics with those of cattle BSE. While the molecular and neuropathological transmission features were undistinguishable and similar to those obtained after transmission of vCJD in both transgenic mouse lines, sheep and goat BSE isolates showed higher transmission efficiency on serial passaging compared to cattle BSE. We found that this higher transmission efficiency was strongly influenced by the ovine PrP sequence, rather than by other host species-specific factors. Although extrapolation of results from prion transmission studies by using transgenic mice has to be done very carefully, especially when human susceptibility to prions is analyzed, our results clearly indicate that Met129 homozygous individuals might be susceptible to a sheep or goat BSE agent at a higher degree than to cattle BSE, and that these agents might transmit with molecular and neuropathological properties indistinguishable from those of vCJD. Our results suggest that the possibility of a small ruminant BSE prion as vCJD causal agent could not be ruled out, and that the risk for humans of a potential goat and/or sheep BSE agent should not be underestimated.
| Prion diseases, also referred as transmissible spongiform encephalopathies, are fatal neurodegenerative diseases caused by proteinaceous infectious particles denominated “prions.” Prion diseases acquired their first real public relevance with the outbreak of bovine spongiform encephalopathy (BSE) (“mad cow disease”) in the United Kingdom in the 80s and its link with the appearance of a new, variant form of Creutzfeldt-Jakob disease in humans. Recycling of ruminant tissues in meat and bone meal has been proposed as origin of the BSE epidemic. During this episode, sheep and goats have also been exposed to BSE-contaminated meal, so transmission to this species may have occurred. We analyzed the human susceptibility to sheep and goat passaged-BSE prions by using transgenic mice expressing human prion protein (PrP). When different sheep and goat BSE isolates were inoculated in these transgenic mice, higher susceptibility than that observed for cattle BSE was detected and the disease manifestation was similar to that observed in mice inoculated with the new variant of Creutzfeldt-Jakob disease. Our findings suggest that humans are at least equally, and might be even more, susceptible to a sheep or goat BSE agent compared to a cattle BSE one.
| Transmissible Spongiform Encephalopathies (TSEs) are fatal neurodegenerative diseases which include Scrapie in sheep and goats, Bovine Spongiform Encephalopathy (BSE) and Creutzfeldt-Jakob disease (CJD) in humans. Prions, the causal agents of these diseases are thought to be infectious protein particles essentially composed of a misfolded isoform (PrPSc) of the cellular prion protein (PrPC) [1], [2]. Scrapie has been detected more than two centuries ago, without epidemiological evidence of human transmission. BSE was diagnosed in cattle in the 80s [3] and subsequently acquired epidemic characteristics in several European countries. Ten years later, a variant form of CJD (vCJD) was identified in humans and linked to the consumption of BSE-infected products [4], [5]. During the BSE epidemic, sheep and goats have also been exposed to BSE-contaminated Meat and Bone Meal, so BSE transmission to these species may have occurred [6]. Sheep and goats are experimentally susceptible to BSE [7], [8], [9], [10], [11] and one confirmed [12] and one suspected [13] BSE-like case have been reported in goats in France and the United Kingdom (UK), respectively. While BSE infection is mostly restricted to the nervous system in cattle [14], [15], [16], [17], PrPSc is widely distributed in lymphoid tissues of experimentally BSE-infected sheep [18], [19], suggesting that infected sheep could provide a secondary and more dangerous source of BSE infection for humans.
Considering the protein-only hypothesis, one of the most difficult aspects to explain within prion diseases is the existence of prion strain diversity. Prion strains can be defined as isolates or sources of prion infectivity, that when transmitted to the same host, present distinct disease phenotypes, characterized by their incubation times, clinical signs, PrPSc biochemical properties, histopathological lesion profiles and PrPSc deposition patterns in the brain [20]. Intra-species prion transmission is characterized to be very efficient, maintaining these phenotypic traits on serial subpassaging. Although PrP primary sequence is highly conserved among mammals, inter-species prion propagation is limited by the so called transmission barrier, showing often at first passage lower attack rates and extended incubation times [21]. During years this transmission barrier has been called species barrier, suggesting to reside essentially in the degree of PrP amino acid sequence homology between donor and receiver [22], [23]. Nowadays there is strong evidence that also PrPSc conformation plays a critical role, not only in the cross-species transmission events, but also in the existence of different prion strains [24], [25], [26], [27], [28], [29].
In previous reports we demonstrated that BSE experimentally passaged in sheep homozygous for the A136R154Q171 allele of ovine PrP showed an enhanced virulence in transgenic mice expressing bovine and porcine PrP, compared to the original cattle BSE [29], [30]. The susceptibility of humans to a sheep or goat passaged BSE agent is still unknown. PrP polymorphism at codon 129, where either methionine or valine is encoded, has been described as a key factor influencing human prion susceptibility [31], [32], [33], [34] and seems to be particularly important in vCJD manifestation, as all but one clinical vCJD cases diagnosed so far are homozygous for methionine. In an attempt to evaluate the potential risk for humans of a sheep or goat BSE agent in this study we analyzed the susceptibility of transgenic mice expressing methionine 129 human PrP to sheep and goat BSE isolates.
In two parallel studies, performed in two different laboratories, the susceptibility of 2 lines of human-PrP transgenic mice to BSE agent was assessed without or after an intermediate passage in either sheep or goat. Both human-PrP transgenic mouse lines overexpress human M129 PrP under the mouse Prnp promoter on a mouse PrP null background but they differ in the genetic construction, genetic background and human PrP protein expression levels: a PAC insert, a Sv129 background and a 6-fold expression levels for tg650 mice [35] and a MoPrP.Xho vector, a 129Ola/B6CBA background and a 4-fold expression levels for tg340 mice (generated as described in Material and Methods). Tg650 (at INRA, France) and tg340 (at CISA-INIA, Spain) were inoculated with several cattle, sheep and goat BSE isolates (see Table 1 for a description of the isolates used). The transmission efficiency was evaluated by the appearance of TSE clinical symptoms and by the presence of PrPres in the brain. The transmission data available to date, together with those obtained comparatively with vCJD or sCJD agents, are shown in Tables 2 and 3.
Both tg650 and tg340 lines were fairly susceptible to vCJD isolates with 100% clinical attack rates and mean survival times around 500 and 600 days post-infection (d.p.i.), respectively. These features were stable upon subpassaging, suggesting an absence of transmission barrier for this agent (Tables 2 and 3). Inoculation of cattle BSE isolates to tg340 mice produced markedly different results, as previously reported with tg650 mice [36]. At first passage, only one out of fifteen tg340 mice inoculated with BSE2 and BSE0 isolates was scored positive for brain PrPres and at a very late stage (739 d.p.i.) without clear clinical signs. The remaining inoculated mice failed to develop a clinical disease or to accumulate detectable levels of PrPres in the brain up to ∼700 days after inoculation. On second passage performed with brain homogenate from a PrPres-negative mouse (succumbed at 576 dpi) from the first passage (BSE2 isolate), 3 out of 4 inoculated tg340 mice tested positive for brain PrPres by western blot with a survival time of 572±37 d.p.i. It is important to note that all the cattle BSE isolates tested in this study were transmitted as efficiently as vCJD isolates or other BSE-related sources to bovine PrP transgenic mice (Table 2 and 3), thus suggesting that they may have a comparable infectivity in the absence of an apparent transmission barrier [36]. Overall, these results indicate that both human-PrP transgenic mouse lines exhibit a strong transmission barrier to cattle BSE, suggesting that human PrPC Met129 is a “bad substrate” for cattle BSE prions.
We next examined the transmission efficiency of sheep- and goat-passaged BSE prions. Several experimental or “natural” isolates of distinct origin were selected (Table 1). Upon primary transmission to both tg340 and tg650 mouse lines, sheep and goat BSE isolates showed significant higher transmission ability than cattle BSE isolates. Thus the attack rates approached 100% (clinical signs, PrPres detection in the brain) with almost all the sheep and goat BSE isolates used (Tables 2 and 3). Depending on the isolate used, the survival times varied between 571±67 and 749±50 d.p.i. in tg650 and 615±84 - 695±22 in tg340, thus much closer to the survival times observed for vCJD isolates inoculated in these mouse models (Tables 2 and 3). On second and third passage, 100% attack rates were obtained with all the sheep and goat BSE isolates tested. The incubation times were stable or for some isolates, slightly decreased and approached that of variant CJD. Overall, these data suggest a lower or an absence of apparent transmission barrier to sheep and goat BSE in human PrP transgenic mice.
Importantly, the comparatively higher attack rates seen with sheep and goat BSE isolates are not related to their initial PrPres content as dilution experiment results indicate that cattle BSE isolates (Ca-BSE2 and Ca-BSE3) contained higher PrPres levels in their brains than the sheep and goat BSE isolates (Figure 1).
Because distinct PrPSc conformations appear to encipher/encode distinct strains, brain PrPres electrophoretic mobility and glycoprofile characterization constitutes standard criteria to distinguish between strains. Brain PrPres of human PrP tg650 and tg340 transgenic mice inoculated with cattle, sheep and goat BSE isolates were analysed by western blot and the signature obtained was compared to that of variant CJD (Figures 2 and 3). A typical PrPvCJD banding pattern, characterized by low size fragments (∼19 kDa fragment for the aglycosyl band) and prominent diglycosylated species was consistently observed in the challenged, PrPres-positive mice. This signature clearly differed from that observed after inoculation of mice with sporadic CJD (Figure 3 and [35]). Similar results were obtained when the immunoblots were performed with the 12B2 anti-PrP antibody, whose epitope (89WGQGG93 according to the human PrP sequence) is known to be poorly protected from proteinase K digestion [29], [37] in vCJD and BSE-related isolates (Figure 4). The only exception was observed after second passage of one cattle BSE isolate (Ca-BSE2) in tg340 mice, one of the three positive mice presented a brain PrPres profile clearly distinct from PrPvCJD, that was comparable to that of typeI sCJD-inoculated tg340 mice with predominantly monoglycosylated and higher size fragments (∼21 kDa for the aglycosyl band) and preserved detection by 12B2 antibody (Figure 4).
The regional distribution of PrPres and vacuolation in the brains are standard criteria to differentiate between strains/TSE agents [38]. We thus compared the neuropathological phenotypes of cattle, sheep and goat BSE agents by PrPres histoblotting and histopathological examination. The PrPres deposition pattern of cattle, sheep and goat BSE were clearly similar in both tg650 and tg340 mice on 2nd passage (not shown) and on 3rd passage in tg650. As illustrated in Figure 5, large plaque-like PrP deposits were detected throughout the brain, predominantly in the cerebral cortex, corpus callosum, thalamic nuclei, optic tract, brain stem and cerebellum, a distribution which is similar to that seen in the brains of vCJD-infected tg650 mice (Figure 5 and [35]).
At microscopic level, abundant amyloid-like plaques were present (Figure 6), as suggested by histoblotting. These plaques were associated with severe vacuolisation of the surrounding tissue (‘florid like’ aspect see Figure 6), precluding the establishment of a reliable standard lesion profile. However similar distribution of the vacuolar changes was observed in the brain of mice inoculated with the different BSE and the vCJD isolates. It mainly involved thalamic, hippocampal and cerebral cortex areas, while brainstem and cerebellar cortex remained poorly affected (Figure 6).
Once known that the phenotypes of cattle, sheep and goat BSE appear indistinguishable in human PrP mice, we proceed to analyze in more detail the potential elements involved in the change on BSE transmission characteristics after passage into sheep or goat. One of the cattle BSE isolate studied (Ca-BSE2) was passaged into bovine (tg110) and ovine PrP transgenic mice (ARQ allele) [39] to propagate BSE agents with different PrP primary sequence (these isolates were termed Ca-BSE2/TgBov and Ca-BSE2/TgOv, Table 1). The Ca-BSE2/TgBov isolate did not induce a clinical disease nor PrPres accumulation in tg340 mice while an intermediate passage on the ovine PrPARQ sequence (Ca-BSE2/TgOv) restored the disease susceptibility, with survival times, biochemical and neuropathological features similar to those obtained with experimental sheep BSE isolates (Figure 7 and data not shown).
The opposite experiment was also performed. One sheep and one goat BSE isolate were passaged (twice for goat BSE) in bovine PrP transgenic mice (generating isolates Sh-BSE0/TgBov and Go-BSE1/TgBov, Table 1) before re-inoculation to human PrP transgenic mice (Figure 7). None of the mice inoculated with these isolates developed the disease nor accumulated PrPres, although one of them (Sh-BSE0/tgBov) produced the shortest disease in bovine PrP transgenic mice (Figure 7). This last result suggests that when sheep- or goat-BSE agents recovered their original bovine PrP sequence, the human transmission barrier was re-established. Moreover, the Sh-BSE0/TgOv isolate (which maintains its PrP ovine sequence) showed a full transmission rate in human transgenic mice with similar survival times as those of the original Sh-BSE0 isolate (Figure 7). Overall these data suggest that PrPSc primary sequence plays a critical role in the capacity of BSE prions to propagate on the human Met 129 PrP sequence.
In this study, we compared the transmission features of cattle and sheep/goat BSE prions in two different models of transgenic mice expressing Met129 human PrP (tg650 and tg340 lines) in two different laboratories. In general, the transmission results obtained in both human-PrP transgenic mouse lines were very comparable. Some shortening in survival times was observed in tg650 mice (compared to the tg340 mice line), which was probably due to higher PrP expression levels in this line. Worryingly, our results support the view that an intermediate passage of BSE agent in small ruminants accelerates the appearance of a vCJD-like disease in human PrP mice or markedly increases its transmission efficiency. Because the apparent phenotype of cattle and sheep/goat BSE prions is conserved, these data also unravel an important role of PrPSc primary sequence in the cross-species transmission capacities of prion strains.
The transmission efficiency of cattle BSE isolates in both human-PrP transgenic mouse models was apparently low. With all BSE isolates, whose high infectivity has been demonstrated in bovine-PrP transgenic mice (Tables 2 and 3), very low attack rates were obtained on primary transmission to both tg650 and tg340 mice. Three passages were necessary to achieve a degree of fitness comparable to vCJD in the same mouse line. This low BSE transmission efficiency to human PrP transgenic mice -occasionally accompanied by a strain shift- has also been described by others [40], [41], [42], and suggests a strong although not absolute transmission barrier. Although the exact characteristics and further evolution of the vCJD epidemic still entail uncertainties owing to prolonged incubation times, this apparent high transmission barrier of humans to cattle BSE might be an explanation for the currently low vCJD incidence, considering the high exposure to BSE during the “mad cow” crisis.
Remarkably, a different picture emerged when the sheep and goat BSE isolates were inoculated to human PrP transgenic mouse models. Attack rates approaching 100% were observed from the primary passage onwards and mean incubation times were more consistent with those measured after transmission of vCJD. On further passaging, the neuropathological phenotype and PrPSc type of cattle and sheep/goat BSE agents appeared indistinguishable from the vCJD agent propagated in these mice, as previously demonstrated in bovine transgenic mice [29], thus strongly supporting the view that the same BSE prion strain has been propagated whatever the infecting species. Hence, these observations reproduced in two distinct human transgenic lines with different genetic background and PrP expression levels support the view that transmission efficiency of BSE prions is increased by an intermediate passage in sheep or goat. Although the electrophoretic pattern of sheep/goat and cattle BSE PrPres appeared similar in human-PrP transgenic mice, other assays are currently performed to further compare the biochemical or biophysical properties of the respective proteins are ongoing.
Importantly, the higher attack rates obtained after sheep and goat BSE transmissions compared to cattle BSE are not in accordance with the initial PrPres content of these isolates. In addition, the data from inoculation to BoPrP-Tg reporter mice suggest that cattle BSE and sheep and goat-BSE isolates could have similar transmission efficiency (Table 1 and 2) in the absence of apparent transmission barrier [36]. Furthermore, when the human PrP transgenic lines were inoculated with the BSE agent passaged into bovine and ovine transgenic mice, the transmission results were comparable to those of the cattle and sheep BSE isolates (Figure 7), further supporting the crucial role of the PrPSc primary sequence in the increase of transmission efficiency. Taken together all these considerations suggest that the higher transmission efficiency of sheep and goat BSE isolates in comparison to cattle BSE isolates cannot be linked to a higher infectious titer of the inoculum but must be the outcome of a modification in the pathogenicity of the agent.
Commonly, transmission barriers are determined considering attack rates and quantified by measuring the fall in the mean survival times between the first and second passage. Hence, if we consider PrPres detection as an indicator of successful transmission, our results imply that humans could be significantly more susceptible to a sheep or goat BSE agent than to a cattle BSE agent. On the other hand our results suggest that cattle BSE infection could produce very long latency in humans, with conversion efficiency far below the threshold of detectable PrPres, which is also very worrying since it suggests the possibility of silent carriers.
Our observations, made in two different mouse genetic backgrounds, suggest that the different transmission properties acquired by BSE after passage into either sheep or transgenic mice expressing ovine PrP are strongly related to the ovine PrP primary sequence, rather than to other host species-specific factors. Thus the transmission barrier observed with cattle BSE was fully restored when sheep/goat BSE experienced intermediate passaging into bovine transgenic mice before reinoculation to human PrP mice. In contrast, when the ovine sequence of sheep BSE was maintained, through passage into ARQ ovine PrP transgenic mice, the efficient transmission to human PrP mice was maintained. Apparently, an ovine/caprine PrPSc sequence appears to facilitate human PrP conversion by the BSE agent, compared to a bovine one.
The PrP primary sequence influence seems to depend strongly on the strain involved, since no PrPres was found in either first or second passages of sheep scrapie in tg340 mice (unpublished observations), suggesting no infection, in accordance with the lack of epidemiological evidence linking scrapie with human TSE. Moreover, the low transmission efficiency observed for the cattle BSE agent is not exclusively linked to the bovine PrP sequence since other uncommon BSE strains (BSE-L) are efficiently transmitted to human-PrP mice [41], [43]. Considering the conformational selection model [20], our results would suggest that M129 human PrPC prefers a BSE PrPSc with conformational characteristics templated by the ovine sequence, to a bovine BSE PrPSc. Because a similar increased transmission efficiency of sheep/goat BSE has been reported in wild type mice [44] and transgenic mice expressing elk [45], bovine [29] and porcine [30] PrP, the better structural compatibility conferred by sheep/goat primary PrPSc sequence may not be limited to human PrPC. One explanation might be an alteration in the quaternary structure (after passage into sheep/goat) generating PrPSc polymers less degraded or more rapidly/easily amplified favouring or enhancing the initial conversion. This question is currently being addressed by sedimentation velocity [46] and PMCA experiments. Another possibility, within the quasispecies concept [20], [47], might be that BSE prions confrontation with the sheep and goat primary PrP sequence increases the variety of BSE substrain components, with the following emergence of a markedly adapted component in response to the selection pressure imposed by the interspecies transmission events. On the other hand, this component would not be distinguishable from bovine-passaged BSE prions due to the current limits of the standard biological methods and/or the molecular tools employed here to characterize prion strains. Whatever the mechanism, the notion that a passage through an intermediate species can profoundly alter prion virulence for the human species has important public-health issues, regarding emerging and/or expanding TSEs, like atypical scrapie or CWD.
Although extrapolation of results from prion transmission studies by using transgenic mice has to be done very carefully, especially when human susceptibility to prions is analyzed, our results clearly indicate that Met129 homozygous individuals might be susceptible to a sheep or goat BSE agent at a higher degree than to cattle BSE, and that these agents might transmit with molecular and neuropathological properties indistinguishable from those of vCJD. Although no vCJD cases have been described in Val129 homozygous individuals so far it is relevant to analyze if similar results will be observed in this genotype. This issue is currently being addressed in transmission experiments using transgenic mice expressing Val129 human PrP.
Taken all together, our results suggest that the possibility of a small ruminant BSE prion as vCJD causal agent could not be ruled out, which has important implications on public and animal health policies. On one hand, although the exact magnitude and characteristic of the vCJD epidemic is still unclear, its link with cattle BSE is supported by strong epidemiological ground and several experimental data. On the other hand, the molecular typing performed in our studies, indicates that the biochemical characteristics of the PrPres detected in brains of our sheep and goat BSE-inoculated mice seem to be indistinguishable from that observed in vCJD. Considering the similarity in clinical manifestation of BSE- and scrapie-affected sheep [48], a masker effect of scrapie over BSE, as well as a potential adaptation of the BSE agent through subsequent passages, could not be ruled out. As BSE infected sheep PrPSc have been detected in many peripheral organs, small ruminant-passaged BSE prions might be a more widespread source of BSE infectivity compared to cattle [19], [49], [50]. This fact is even more worrying since our transmission studies suggest that apparently Met129 human PrP favours a BSE agent with ovine rather than a bovine sequence. Finally, it is evident that, although few natural cases have been described and so far we cannot draw any definitive conclusion about the origin of vCJD, we can not underestimate the risk of a potential goat and/or sheep BSE agent.
Animal experiments were carried out in strict accordance with the recommendations in the guidelines of the Code for Methods and Welfare Considerations in Behavioural Research with Animals (Directive 86/609EC) and all efforts were made to minimize suffering. Experiments were approved by the Committee on the Ethics of Animal Experiments of the author's institutions (INRA and INIA); Permit Number: RTA06-091 and CT05-036353.
The isolates used in this study are described in Table 1. For mouse inoculation, all isolated were prepared from brain tissues as 10% (w/v) homogenates in 5% glucose.
The tg650 transgenic mouse line over expresses human PrP M129 at a 6-fold level on a mouse PrP null background [35]. The tg340 mouse line expressing about 4-fold level of human PrP M129 on a mouse PrP null background has been generated following the same procedure previously described for the generation of other transgenic mouse line expressing different species PrP [51]. The details of this procedure are described below. Tg110 and tg540 mouse lines expresses bovine PrP at levels approximately 8-fold that in cattle brain [51], [52].
All inocula were prepared from brain tissues as 10% (w/v) homogenates. Individually identified 6–10 week-old mice were anesthetized and inoculated with 2 mg of brain homogenate in the right parietal lobe using a 25-gauge disposable hypodermic needle. Mice were observed daily and the neurological status was assessed weekly. When progression of a TSE disease was evident or at the end of lifespan, animals were euthanized because of ethical reasons. Once euthanized, necropsy was performed and brain was taken. A part of the brain was fixed by immersion in 10% formol to quantify spongiform degeneration by histopathology and PK resistant PrP accumulation (PrPres) by immunohistochemistry (IHQ) or histoblotting and the other was frozen at −20°C to determine presence of PrPres by Western blot (WB). In all cases, survival time and attack rate were calculated for each isolate. Survival time was expressed as the mean of the survival days post inoculation (d.p.i.) of all the mice scored positive for PrPres, with its correspondent standard error. Attack rate was determined as the proportion of mice scored positive for PrPres from all the mice inoculated. When all mice were scored negative for PrPres, the survival time range was shown. Brain homogenates from PrPres positive mice, when available, were used for further passaging. When all mice were scored negative for PrPres on primary passage, PrPres-negative brain homogenates were used for second passage.
175±20 mg of frozen brain tissue were homogenized in 5% glucose in distilled water in grinding tubes (Bio-Rad) adjusted to 10% (w/v) using a TeSeE™ Precess 48™ homogenizer (Bio-Rad) following manufacturer instructions. Presence of PrPres in transgenic mice brains was determined by Western blot, following the procedure described below and using the reagents of the ELISA commercial test (TeSeE, Bio-Rad). 10–50 µl of a 10% (w/v) brain homogenate were diluted in a 10% (w/v) negative sheep brain homogenate, to obtain a 200 µl final volume. Homogenates were incubated for 10 min at 37°C with 200 µl of a 2% proteinase K solution (in buffer A). PrPres was recovered as a pellet after addition of 200 µl of buffer B and a centrifugation at 15,000× g for 7 min at 20°C. Supernatants were discarded and pellets were dried inverted over absorbent paper for 5 min. Pellets were solubilised in Laemmli buffer and samples were incubated for 5 min at room temperature, solubilised, and heated at 100°C for 5 min. Samples were centrifuged at 20,000× g for 15 min at 20°C and supernatants were recovered and loaded on a 12% Bis-Tris Gel (Criterion XT, BioRad or NuPage, Invitrogen). Proteins were electrophoretically transferred onto PVDF or nitrocellulose membranes (Millipore). Membranes were blocked O/N with 2% BSA blocking buffer. For immunoblotting, membranes were incubated with either Sha 31 [53] or 12B2 [37] monoclonal antibody (Mab). Immunocomplexes were detected incubating the membranes for 1 hour with horseradish peroxidase conjugated anti mouse IgG (Amersham Pharmacia Biotech). Immunoblots were developed with enhanced chemiluminescence ECL Plus (GE Healthcare Amersham Biosciences).
All procedures involving mice brains were performed as previously described [54]. Brain slices were realized, in order to allow lesion profiling according to the standard method described by Fraser and Dickinson [55]. Briefly, samples were fixed in neutral-1 buffered 10% formalin (4% 2- formaldehyde) before being cut at determined levels and paraffin embedded. After deparaffinization, 2 µm-thick tissue sections were stained with haematoxylin and eosin.
Brains were rapidly removed from euthanised mice and frozen on dry ice. Thick 8–10 µm cryostat sections were cut, transferred onto Superfrost slides and kept at −20°C until use. Histoblot analyses were performed on 3 brains per infection at 2nd and 3rd passage, using the 3F4 anti-PrP antibody as previously described [35].
Tg340 mouse line expressing about 4-fold level of human PrP M129 on a mouse PrP null background has been generated following a similar procedure previously describe for the generation of other transgenic mouse line expressing different species PrP [51], [56]. Briefly, the open reading frame (ORF) of human PrP gene was isolated by PCR amplification from human genomic DNA encoding methionine at codon 129. The primers used created a XhoI restriction enzyme site adjacent to the translation start and stop sites of the human PrP ORF (5′ -CTCGAGATTATGGCGAACCTTGGCTGCTGG- 3′ and 5′- CTCGAGTCATCCCACTATCAGGAAGATGAG- 3′, respectively). The PCR fragments obtained were sub cloned into a pGEM-T Easy Vector System (Promega) following manufacturer instructions, and inserts were sequenced to confirm no differences in the inferred amino acid sequence with respect to previously sequenced human PrP genes (GenBank accession number NM_183079) and to confirm the presence of the consequent codon 129 nucleotide variant (MetATG). The human PrP ORF was excised from the cloning vector using the restriction enzyme XhoI and inserted into the expression vector MoPrP.Xho [51], [56]. This vector contains the murine PrP promoter (including exon 1, intron 1, exon 2 and 3′ untranslated sequences) flanked by two XhoI restriction sites but could be distinguished from the wild type murine PrP gene because of the absence of intron 2. The vector was also digested with XhoI to excise the murine PrP ORF and the correspondent human PrP ORF were inserted by ligation, obtaining the plasmid pMo-huPrP129M.Xho.
The human transgene was excised from the plasmid vector using the restriction endonuclease Not I leading to DNA fragments of approximately 12 Kb. Finally, the DNAs were purified and dissolved in TE at a final concentration of 2 to 6 µg/ml and microinjected into pronuclear stage ova collected from super-ovulated B6CBAf1 females mated with 129/Ola males carrying a null mutation in endogenous PrP [51], [56].
DNA from founders' tails biopsies was extracted using a Extract-N-Amp Tissue PCR kit (Sigma-Aldrich) following manufacturer instructions. The presence of the human transgene in these founders was identified by PCR amplification using specific primers for the mouse PrP exon 2 and human PrP open reading frame. The absence of the murine PrP ORF in the transgenic mice generated was confirmed by PCR amplification using the primers: 5′- TAGATGTCAAGGACCTTCAGCC- 3′ and 5′- GTTCCACTGATTATGGGTACC -3′.
Eight different lines (founders) of human PrPC (huPrP) and murine PrPC (muPrP) heterozygous transgenic mice (PrP mu+/− hu+/−) were obtained. The expression of human PrPC in brain of these mouse lines was analyzed and compared with PrPC content in human brain homogenate by western blot using mAb 3F4 which recognizes the 109MKHM112 epitope (numbered according to the human PrP sequence). Human PrPC was detected in 100% of the tested lines (data not shown). From the initial 8 different mouse lines heterozygous for both murine and human PrP genes (PrP mu+/− hu+/−), the mouse line named as tg340 was selected for further experiments on the basis of the level of PrPC expression.
Homozygous Tg340 mouse line was established backcrossing these animals with homozygous null animals MuPrP−/− (Prnp−/−) to obtain a null murine PrP background (PrP mu−/− hu+/−). Interbreeding within these animals was performed to obtain homozygosis for the human PrP transgen within a murine PrP background (PrP mu−/− hu+/+). The absence of murine PrP gene was determined by PCR using specific primers. Human PrPC expression levels, determined more accurately in brain from homozygous tg340 animals was about 4-fold higher than PrPC levels in human brain homogenates as determined by dilution experiments in western blot (Figure 8).
The GenBank accession number for the human Prnp gene used in this paper is NM_183079.
|
10.1371/journal.pbio.1002452 | Functional Sites Induce Long-Range Evolutionary Constraints in Enzymes | Functional residues in proteins tend to be highly conserved over evolutionary time. However, to what extent functional sites impose evolutionary constraints on nearby or even more distant residues is not known. Here, we report pervasive conservation gradients toward catalytic residues in a dataset of 524 distinct enzymes: evolutionary conservation decreases approximately linearly with increasing distance to the nearest catalytic residue in the protein structure. This trend encompasses, on average, 80% of the residues in any enzyme, and it is independent of known structural constraints on protein evolution such as residue packing or solvent accessibility. Further, the trend exists in both monomeric and multimeric enzymes and irrespective of enzyme size and/or location of the active site in the enzyme structure. By contrast, sites in protein–protein interfaces, unlike catalytic residues, are only weakly conserved and induce only minor rate gradients. In aggregate, these observations show that functional sites, and in particular catalytic residues, induce long-range evolutionary constraints in enzymes.
| The basic biochemical functions of life are carried out by large molecules called enzymes. Enzymes consist of long chains of amino acids folded into a three-dimensional structure. Within that structure, a specific cluster of amino acids, known as the active site, performs the biochemical function. Substituting one amino acid for another in the active site typically results in a defective, non-functional enzyme, and therefore mutations at or near enzyme active sites are often lethal. Moreover, even mutations far from the active site have been found to disrupt function. Nonetheless, as organisms evolve, enzymes accumulate random mutations. Where in enzymes’ structures do these mutations accumulate without causing harm? Here, we observe evidence for extensive interactions between active sites and distant regions of the enzyme structure, in a comprehensive set of over 500 enzymes. We show that active sites tightly control the substitutions that an enzyme can tolerate. This control extends far beyond regions of the enzyme immediately adjacent to the active site, covering over 80% of a typical enzyme structure. Our findings have broad implications for molecular evolution, for enzyme engineering, and for the computational prediction of active-site locations in novel enzymes.
| Enzymes facilitate the chemical reactions necessary for life. To function properly, enzymes must reconcile two competing demands: they must fold stably into the correct three-dimensional conformation, and they must display the correct catalytic residues in their active sites. As enzymes evolve, mutations that are functionally beneficial are often deleterious for stability, and vice versa [1–3]. Thus, the patterns of evolutionary divergence observed in enzyme evolution are shaped by the interplay of these two potentially conflicting constraints.
Mutations affecting fold stability can occur anywhere in the protein structure, though in general stability effects tend to be more pronounced in the interior, more densely packed regions of a structure than on the protein surface [4,5]. By contrast, where mutations affect function in a protein structure is less clear. Site-directed mutagenesis experiments demonstrate that mutations at catalytic residues, unsurprisingly, disable enzyme function [6,7]. Accordingly, residues directly involved in protein function tend to be more conserved over evolutionary time than other residues [8–10]. Less intuitively, however, mutations 20 Å or more from a catalytic residue can attenuate catalytic activity in enzymes such as glycosidase [11], TEM-lactamase [12], or copper nitrate reductase [13]. Similarly, a study of a small set of α/β-barrel enzymes has found that evolutionary conservation decays continuously with the distance to the nearest catalytic residue [14]. These results suggest that residues far from an active site may be functionally important, but that this importance may decline with distance in physical, three-dimensional space.
Here, we analyze a dataset of 524 distinct enzyme structures spanning the six major functional classes of enzymes. We systematically assess how site-specific evolutionary variation in these enzymes relates to the geometric location of residues relative to the nearest catalytic residue. We find that, across all six major classes of enzymes, the constraining effects of catalytic residues extend to most of an enzyme’s structure, irrespective of protein size. These effects exist regardless of whether an active site is located on the surface or in the core of a protein, and they remain even when controlling for other structural features predicting evolutionary variation. Finally, we find that we can use site-specific conservation gradients to accurately recover active sites in more than 50% of enzymes. In summary, these findings demonstrate that functional sites induce long-range evolutionary constraints in enzyme structures.
To systematically explore the relationship between site-specific evolutionary rates and distance to the nearest catalytic residue, we have analyzed 524 diverse enzyme structures. We have chosen these structures as a subset of enzymes analyzed previously for their relationship between protein structure and evolutionary variation [15]. The structures represent all six major classes of enzymes, and no two structures in the dataset share more than 25% of their respective amino-acid sequences. The dataset includes both single subunit proteins (monomers) and multi-subunit proteins (multimers), and annotations describing the biological assembly and the location of the catalytic residues are available for each structure (see Methods for details). For each enzyme, we have constructed alignments of up to 300 homologous sequences, selected from the UniRef90 database [16,17]. We estimate evolutionary variation at each site in each alignment by calculating a site-specific relative evolutionary rate, using the software Rate4Site [18]. The relative rates are normalized such that a value of one corresponds to the average rate in a given protein, and larger or smaller values represent proportionally larger or smaller rates. For brevity, we will also refer to the relative rates simply as “rates.” In mathematical expressions, rates will be denoted by the letter K.
We first ask whether there is an overall trend toward increased evolutionary conservation near active sites. To address this question, we pool all sites from all structures into one combined dataset and then calculate the mean evolutionary rate as a function of Euclidean distance to the nearest catalytic residue in the respective structure. As expected, we find that evolutionary rates are, on average, the lowest at or directly near catalytically active sites. Moreover, we find that rates increase approximately linearly with increasing distance to the nearest catalytic residue, up to a distance of approximately 27.5 Å (Fig 1A). Beyond this distance, rates level off. Importantly, 80% of all residues in our dataset fall within a distance of 27.5 Å to the nearest catalytic residue (Fig 1B). Thus, the vast majority of all residues in each protein appear to experience some amount of purifying selection mediated by catalytic residues.
We can think of sites in a protein as organized into shells according to their distance to the closest catalytic residue. Each shell is 5 Å in width, the approximate minimum distance between two amino acid side-chains. The boundaries between these discrete shells are indicated in Fig 1B with dashed lines, and we can see clear dips in the distribution at 2.5 Å and 7.5 Å, the boundaries between the 0th and 1st and the 1st and 2nd shells (the boundaries between shells become less precise for higher shell numbers). We can subdivide the sites of our dataset into these discrete shells and then plot the rate distribution within each shell (Fig 1C). We find that the mean rate for each shell increases up to shell 6 (32.5 Å) and then stabilizes. Similarly, the width of the distribution also increases up to shell 6. Thus, all shells include some proportion of conserved sites, but increasingly distant shells include an increasing fraction of moderately or highly variable sites.
The broad rate distributions that we observe within individual shells, in particular within shells distant from catalytic residues, highlight that there are other factors besides distance that also influence the extent and type of selection acting on individual sites. In fact, one important evolutionary constraint is the requirement for proteins to fold stably into their active conformation [19]. This constraint causes sites in the interior of the protein, shielded from the solvent and involved in many inter-residue contacts, to be more evolutionarily conserved than sites on the protein surface (for a recent review, see [30]) [4,15,20–29].
Two structural measures are commonly used to quantify this structural constraint: relative solvent accessibility (RSA) [31] and weighted contact number (WCN) [32]. RSA measures the exposure of a given residue to a hypothetical small solvent molecule, typically water. RSA is useful for determining if a residue is on the surface or the interior of a protein structure. WCN measures the local packing density of a given residue. WCN is high in the core of the protein, where residues are tightly packed. We have calculated both WCN and RSA for each site in each protein in our dataset. We have based this calculation on the published biological assembly of each protein, so that intra-chain contacts are properly accounted for in the case of enzymes that natively function in a multimeric state. As has been reported previously, on average WCN displays higher correlations with site-specific rate than RSA does, in particular when WCN is calculated with respect to the side-chain coordinates of each residue (see also S1 Fig) [33]. However, in our dataset, correlations of rate with WCN are only moderately higher than correlations with RSA, and there are proteins for which RSA outperforms WCN (S1 Fig). Therefore, throughout this work, we consider both WCN and RSA as measures of structural constraints acting on site-specific protein evolution. Importantly, neither WCN nor RSA make any assumptions about catalytic residues in proteins. Both quantities are purely geometric measures of protein structure. Conversely, the distance d to the closest catalytic residue does not explicitly contain information about packing density or solvent accessibility. Yet, in our dataset, the three quantities WCN, RSA, and d are all correlated with each other (S2 Fig). Therefore, we next ask to what extent the distance d captures an evolutionary constraint that is distinct from the constraints captured by WCN and RSA.
To address this question, we regress site-specific evolutionary rates K against WCN, RSA, and d, in all possible combinations, and separately for each enzyme in our dataset. We then record the R2 for each model and each enzyme (Fig 2A). We find that the best purely structural model, using both WCN and RSA as predictor variables, explains on average 39% of the variation in rate (Fig 2A). Adding distance as a third predictor to this model increases the average R2 to 44%. Thus, distance explains on average at least 5% of rate variation that cannot be attributed to purely structural factors, and possibly more than that; by itself, distance explains on average 25% of the variation in rate. Some of that variation may be accidentally captured by WCN or RSA, because active sites are frequently located closer to the interior than to the surface of the protein structure.
To further assess the independent contribution of distance to the pattern of site-specific rates, we compare model predictions and empirical rates as functions of distance to the active site. We compare rates predicted by the linear models K ∼ WCN + RSA and K ∼ WCN + RSA + d, which are fit for each protein individually. For visualization only, we average within shells, as explained above. We find that a linear model containing only WCN and RSA tends to overestimate site-specific evolutionary rates within the first three to four shells (green line in Fig 2B and 2C). Adding distance to this model removes nearly all of the overestimation (orange line in Fig 2B and 2C). These findings demonstrate that structural metrics alone are unable to accurately predict conservation patterns near active sites.
Enzymes often sequester substrates into a buried catalytic core. This sequestration allows them to facilitate chemistry that would otherwise be impossible in the broader cellular environment. For this reason, many enzymes tend to have active sites in the protein interior, where local packing density is high and solvent accessibility is low [19]. For those enzymes, we expect the distance metric d to correlate with WCN and/or RSA. By contrast, if the active site is located on the protein surface, then distance should correlate very little or not at all with either WCN or RSA.
To further disentangle active-site effects from WCN and RSA, we can identify individual structures from our dataset in which distance is sufficiently uncorrelated (defined as r < 0.25) from both WCN and RSA. Among these structures, we find four for which distance correlates strongly with evolutionary rate (defined as r ≥ 0.55) (see Fig 3). They correspond to the enzymes dihydrofolate reductase (DHFR, protein databank identifier [PDB ID]: 1DHF) [34], superoxide reductase (SOR, PDB ID: 1DO6) [35], anti-sigma factor SpollAB (PDB ID: 1L0O) [36], and the Serratia endonuclease (PDB ID: 1SMN) [37]. All of these enzymes perform different biological functions, and they are active in multimeric conformations. In these proteins, rate correlates more strongly with distance than it does with WCN or RSA (Fig 3A), and the mean rate increases linearly with distance throughout the entire structure (Fig 3B). In all four cases shown, the active sites are located near the protein surface (mean active-site RSA ranges from 0.19 to 0.25) and away from the protein center (Fig 3C).
To analyze the effect of active-site location on rate variation more systematically, we next subdivide our entire dataset into three categories based on active site location, measured by the mean RSA of all catalytic residues in the structure. We define these categories as active site in the protein interior (mean catalytic-residue RSA < 0.05), active site with intermediate solvent exposure (mean catalytic-residue RSA between 0.05 and 0.25), and active site on the protein surface (mean catalytic-residue RSA ≥ 0.25). Our dataset contains 98, 367, and 59 proteins in these three categories, respectively.
As before, we find that the purely structural metrics RSA and WCN tend to overestimate site-specific evolutionary rates near the active site in all three groups (Fig 4). Moreover, the structure-based models perform worse as the active site moves from the core of the enzyme to the surface. In all cases, incorporating distance into the model corrects most rate overestimation near the active site. Interestingly, all models perform better for active sites in the core than for active sites on the surface (Fig 4). We interpret this observation as follows: When the active site is located in the core of an enzyme, functional and structural constraints are aligned. The sites most conserved due to function are also the sites most conserved due to structure, and this overall trend is captured well in the linear models. By contrast, when the active site is located on the surface, functional and structural constraints are at odds with each other. The sites most conserved due to function are now the sites least conserved due to structure, and vice versa. In this case, since there are now two opposing trends within one structure, it is more difficult for any linear model to accurately capture rate variation throughout the structure.
We have previously seen that approximately 80% of all residues in our dataset fall within the 27.5 Å cutoff, inside of which evolutionary variation is reduced in proportion to distance to the nearest active site (Fig 1B). However, the 80% figure may be somewhat misleading, because in that analysis we have pooled all residues from all proteins. Our dataset comprises proteins of very different sizes, from 95 to 1,287 amino acids long, and for small proteins every residue falls within 27.5 Å of an active site, while for large proteins only one-half to two-thirds of the residues lie within the 27.5 Å distance cutoff.
To ascertain whether the relationship between functional sites and evolutionary rates depends on enzyme size, we can re-analyze our data by protein size. We define three evenly sized groups: small proteins (95–268 sites), medium-size proteins (270–385 sites), and large proteins (386–1,287 sites). Each group contains 175, 175, and 174 structures, respectively. We observe that as enzyme size increases, the rate–distance slope decreases (Fig 5A). Distance effects are weaker in larger proteins but also extend further out. The effect remains visible when we analyze the distance–rate relationship for individual proteins and in the context of WCN and RSA (Figs 5B and S5): purely structural models, which use only WCN and RSA to predict rate, overestimate rate up to shell 3 in small proteins, up to shell 4 in medium-sized proteins, and up to shell 5 in larger proteins.
In summary, we see that the leveling off of rate at shell 5, around 27.5 Å, in the pooled dataset, does not represent a universal cutoff but rather an average obtained from combining many different structures into one analysis. For any individual protein, there will generally be a distance effect, but it may extend only to shell 3 or 4 in small proteins while extending to shell 6 (and possibly beyond) for very large proteins.
As we have seen from the preceding analyses, active sites in enzymes impose a selection gradient that can be detected throughout the majority of the protein structure. This observation leads us to ask whether we can use this gradient to identify active sites when their location is not known. To answer this question, we blindly search for distance–rate gradients in our dataset. We systematically use one residue at a time as a reference point in the structure and fit a linear model of rate versus the distance to that reference point. We record the resulting R2 for each model, and we consider the reference point with the highest R2 as the putative active site in the structure.
We find that in 18% of the structures in our dataset, the putative active site coincides with a known catalytic residue (Fig 6). In an additional 37% of structures, the putative active site falls within 7.5 Å of a catalytic residue but is not a catalytic residue itself. A distance of 7.5 Å corresponds to one shell, i.e., it captures residues in direct contact with a catalytic residue. Note that the gap visible between 0 and 2.5 Å in Fig 6 corresponds to the closest distance that two side chains can physically contact each other. A putative active site either is a catalytic residue, in which case it has a distance of 0 Å to the nearest catalytic residue (i.e., itself), or alternatively it has to be at least a distance of 2.5 Å away from the catalytic residue. In summary, for more than half (55%) of the 524 enzymes in our dataset, we can use the existing selection gradient to identify either a catalytic residue or an immediate neighbor.
As a control, we have also considered a model that places the active site at the core of the protein, at the residue with the overall highest WCN (since, as stated above, the active site is located in the protein interior for many enzymes). We find that this control approach recovers catalytic residues or their immediate neighbors in 31% of enzymes (S6 Fig). Thus, while the control approach can recover active sites in a substantial fraction of enzymes, the selection-gradient-based method performs significantly better (odds ratio = 2.8, p < 1.7 x 10−15, Fisher’s Exact Test, S7 Fig, S1 Table).
Many enzymes function as components of multimeric protein complexes. In fact, more than half of the enzymes in our dataset contain multiple subunits in their biological assemblies. The arrangement of and interaction between these subunits could substantially modify how protein structure and protein function shape protein evolution, especially if the active site occurs at the interface of two subunits. In our dataset, we find that residues in protein–protein interfaces are, on average, only slightly more conserved than any other residues, whereas catalytic residues are much more conserved (Fig 7A): residues in interfaces evolve, on average, at a rate of 0.91 relative to the average residue, while catalytic residues evolve, on average, at a relative rate of 0.10. To verify that the little conservation we see in interface sites is not an artifact of our enzyme dataset, we have also analyzed rates in a set of 17 non-enzymatic protein–protein complexes, consisting of 30 individual proteins total (see Methods). Again, we find that residues in protein–protein interfaces show only moderate conservation relative to all other residues (relative rate of 0.82, Fig 7A). Moreover, consistent with their weak conservation, protein–protein interfaces induce only very minor gradients of conservation, if any, in both enzyme and non-enzyme proteins (compare Fig 7B–7D with Fig 1). Thus, protein–protein interactions impose much weaker evolutionary constraints than catalytic sites.
The structural metrics RSA and WCN are also sensitive to subunit arrangement, and subunit arrangement can be incorrectly annotated in the biological assembly. To assess whether subunit arrangement and/or annotation errors affect the distance–rate relationship, we re-analyze our enzyme data using three additional variations of analysis choices: (i) RSA and WCN are calculated using the biological assembly, and any residues at the interface between subunits are excluded; (ii) RSA and WCN are calculated using a single subunit, and no residues are excluded; and (iii) RSA and WCN are calculated using a single subunit, and all interface residues are excluded (S8–S34 Figs). In all three cases, our results remain qualitatively unchanged from our prior results: rate increases with increasing distance to a catalytic residue, up to about 27.5 Å; distance has an effect on rate variation that is independent from the purely structural metrics WCN and RSA. Thus, in summary, there is a positive distance–rate relationship that is independent of WCN or RSA, and it exists regardless of how we treat multi-subunit enzymes and interface residues.
We have shown that many enzymes exhibit a clear, nearly linear relationship between site-specific evolutionary rates and distance to the nearest catalytic residue. We have found this trend consistently throughout a large dataset of 524 diverse enzymes, and we have found that the relationship extends to most of the residues in any given enzyme structure. Using combined linear models containing RSA, WCN, and distance, we have found that distance explains at least 5% of the variance in rate after controlling for WCN and RSA, and potentially up to approximately 36% (Fig 3) in proteins in which the active site is located near the protein surface. Moreover, models containing only the structural predictors WCN and RSA consistently overestimate evolutionary variation near active sites, through shell 5 (27.5 Å) in large proteins. Finally, we have shown that in over half of the enzymes in our dataset, we can recover catalytic residues or their immediate neighbors from the evolutionary gradients they imprint throughout the protein structure.
For some enzymes in our dataset, we have found little evidence for functional or structural constraints on site-specific evolutionary rates. There are some proteins for which less than 10% of the variation in evolutionary rate can be accounted for with distance to a catalytic residue, RSA, or WCN. These low correlations suggest that either the rates themselves are uninformative, or that the available PDB structures are not reflective of protein structure in vivo. In the first case, the sequence alignments used to determine evolutionary rate could contain a mix of proteins with very different arrangements in vivo. We have no way of determining the biological assembly of every sequence in the alignment, so differences in corresponding subunit arrangement could bias the site-specific evolutionary rates. Additionally, the RCSB protein database may have conflicting biological assemblies. For example, the biological assembly for human dihydrofolate reductase (PDB ID: 1DHF) is classified as a homodimer, while the biological assembly for a separate structure (PDB ID: 1DRF) of the same protein is classified as a monomer. We have attempted to control for possible structural variability in the sequence alignments and biological assemblies by re-analyzing all structures as monomers and/or removing residues at the interface of subunits before computing correlations, and the overall trends observed remain the same (S8–S34 Figs). Regardless, all of the factors mentioned here could result in rates that correlate poorly with any structural predictors.
The field of molecular evolution has long sought to understand the relationship between protein structure, function, and sequence evolution [30]. Here, we have assessed this relationship by comparing distance to the nearest catalytic residue with site-specific evolutionary rates. Past work has employed covariation analyses to reveal clusters of co-evolving residues in protein structures, deemed “protein sectors” [38]. In specific cases, such as in serine proteases, these protein sectors also correspond to different functional biochemical regions of the structure. A recent reanalysis of the seminal protein sector work demonstrates that, in proteins with just one sector, sequence conservation recovers clusters of functional residues just as well as covariation analyses [39]. Our work demonstrates that not only are clusters of functional residues highly conserved, but that such residues induce gradients of conservation within a structure. This finding of long-range interactions between residues is consistent with the sector model of large regions of co-evolving residues.
We have found that active sites are among the most highly conserved sites in proteins, whereas residues involved in protein–protein interactions are only weakly conserved relative to the average site in a protein. Moreover, the gradients of conservation induced by protein–protein interfaces are much less marked than those induced by catalytic sites. This finding is consistent with prior work on protein–protein interactions. While several prior works have found increased conservation in interface regions [40–42], effect sizes have generally been found to be small. For example, [42] found that the reduction in evolutionary rate in a protein–protein interface was mostly (though not entirely) explained by the reduction in solvent accessibility induced by complex formation. Also, complementation assays and computer simulations suggest that protein–protein interfaces can experience extensive divergence without loss of function [43], again supporting the notion that protein–protein interfaces are frequently not under strong purifying selection. For these reasons, we believe that the rate gradients we have found toward active sites are not strongly confounded by protein–protein binding interfaces.
Long-range interactions between residues in a protein have historically been studied in the context of allostery. Initially proposed in 1961, allostery describes the process by which a small molecule (ligand) binds to one area of an enzyme (allosteric site) and induces a conformational change at a distant active site [44]. Studies of allosteric interactions shed light on two key aspects of our findings. First, biophysical models have been developed that explain long-range interactions. The Monod-Wyman-Changuex model, the most widely studied of allostery models, proposes that ligand binding stabilizes a biologically active or inactive quarternary structure [44]. Recent studies, however, demonstrate that some monomeric proteins also contain allosteric sites [44]. In G-protein coupled receptors, for example, a simplified model of conserved, physically connected amino-acid residues explains the long-range interactions between allosteric sites and active sites [45]. Our dataset contains a mix of monomeric and multimeric proteins, and we observe long-range interactions in both types of proteins. Thus, our findings suggest that allosteric-like couplings between active sites and distant residues may be more common than previously thought. The physical distance between allosteric ligand-binding sites and active sites ranges from 20 Å in hemoglobin to 60 Å in glycogen phosphorylase [46]. Therefore, the selection gradients we have observed here extend to distances well within the range of experimentally observed allosteric interactions. Second, while the observed selection gradients have allowed us to recover residues in close proximity to the active site in a little over half of the proteins in our dataset, in many proteins (45%) the selection gradient points toward a residue >7.5 Å from the active site. It is possible that these non-catalytic residues, which are highly predictive of the overall patterns of evolutionary rates in the structure, may be allosteric sites. Allosteric sites tend to be highly conserved, although typically not as conserved as active sites [47]. In summary, studies of allostery provide biophysical explanations for long-range interactions between residues and may explain why we failed to recover catalytic residues from selection gradients in some proteins in our dataset.
That selection gradients can recover active sites has potentially broad applications, even beyond enzymatic proteins. For example, some of us have previously used optimized distance to identify important functional sites in influenza A hemagglutinin (HA) [48]. HA, a viral surface protein, interacts directly with sialic acid found on the surface of human cells. Viral infection requires binding of HA to sialic acid, and antibodies bind near the sialic-acid binding region to inhibit viral infection. Residues in that region are thus under strong positive selection for immune escape, and consequently the selection gradient in HA revealed a rapidly evolving functional site. This finding suggests that selection gradients could effectively recover diverse types of functional sites, not only those that are well conserved. More broadly, evolutionary history is a useful predictor of active sites [8,9] and binding partners [10]. Assuming that a given structure has been crystalized, the rate gradients we have found here could improve computational predictions of active sites and binding sites.
We selected 524 of 554 previously characterized enzymes [15] to conduct our analysis. We removed 30 structures because they contained chains with no available catalytic residue information, or because the UniRef90 database did not contain enough homologous sequences to construct a diverse alignment. These enzymes consist of 204 monomers and 320 multimers, and no two enzymes in the dataset have more than 25% sequence identity. For each enzyme, we obtained catalytic residue information from the Catalytic Site Atlas [49].
We acquired PDB structures of the biological assemblies for these proteins from the RCSB protein database [50]. A biological assembly represents the functional form of a given enzyme in vivo based on the best experimental data available. When available, we used biological assemblies that are author-provided or both author-provided and software-supported (labeled “A” and “A+S,” respectively, in the RCSB protein database). If author-provided biological assemblies were not available, we used biological assemblies predicted by PISA (protein interfaces, surfaces, and assemblies, http://www.ebi.ac.uk/pdbe/pisa/) (labeled “S”). PISA biological assemblies are entirely predicted by software. In cases in which there were multiple author-provided biological assemblies, we chose the first of those assemblies listed in the RCSB protein database.
In addition to the enzyme dataset, we also compiled a non-enzyme dataset as a control. We selected 17 of 179 protein–protein complexes from the Protein–Protein Interaction Affinity Database 2.0 [51]. We selected only non-enzymatic proteins based on interaction classification, absence of enzyme comission (EC) number, and UniProt annotations. We also excluded complexes containing antibodies, since antibodies evolve on a different time-scale and by different mechanisms than other cellular proteins. We acquired structures of the protein–protein complexes from the RCSB protein database.
To calculate site-specific evolutionary rates, we first extracted the amino-acid sequences from the PDB structures. Using PSI-BLAST [52], we then queried the UniRef90 database [16,17] to retrieve homologous sequences for each enzyme. Among these homologous sequences for each enzyme, we removed sequences with less than 10% pairwise divergence to any other sequence, to reduce phylogenetic bias. Next, we randomly downsampled the homologous sequences to a maximum of 300 sequences per enzyme. Then, we performed a multiple sequence alignment (MSA) of the sequences with MAFFT 7.215 (Multiple Alignment using Fast Fourier Transform) [53] and generated phylogenetic trees with RAxML 7.2.8 (Randomized Axelerated Maximum Likelihood) [54] using the LG substitution matrix (named after Le and Gacuel) [55] and the PROTCAT model of rate heterogeneity [56]. We calculated site-specific evolutionary rates with the program Rate4Site 2.01 [18], using the MSAs and phylogenetic trees from the previous step as input. We used the empirical Bayes approach for rate estimation and the JTT (Jones, Taylor, and Thorton) model of amino acid replacement [57]. Lastly, we normalized the rates such that the rates for each protein have a mean of 1. Because these rates are measured relative to the average divergence rate of the entire protein, they are dimensionless. Throughout this work, we refer to these site-specific relative rates as K, or simply “rates.”
For each protein structure, we calculated several predictor variables at each site. First, we calculated the weighted contact number WCN for each residue i as follows:
WCNi=∑j≠i1rij2.
(1)
Here, rij is the distance between the geometric center of the side-chain atoms in residue i and the geometric center of side-chain atoms in residue j. To calculate these distances for residue pairs involving glycine, which has no side-chain, we used the location of the Cα in those residues instead. Unless noted otherwise, WCN was calculated using the complete biological assembly of the protein.
Next, we calculated the relative solvent accessibility (RSA) at each site. To this end, we first calculated the accessible surface area (ASA) using the software mkdssp [58,59]. We then normalized ASAs by the maximum solvent accessibility for each residue in a Gly-X-Gly tripeptide [31]. Peptide linkages across chains, typically disulfide bridges, were assigned an RSA of zero. Unless noted otherwise, RSA was calculated using the complete biological assembly of the protein.
Finally, we calculated the distance d to the nearest catalytic residue for each residue in each structure. Most enzymes have multiple catalytic residues, so we define d as distance to the nearest catalytic residue. As was the case for WCN, distances were measured from the geometric center of the side-chain of one residue to the the geometric center of the side-chain of another residue. And in the case of glycines, the position of Cα was again used in place of the side-chain geometric center. Any residue with d = 0 is therefore a catalytic residue, and conversely, all catalytic residues lie at d = 0.
We defined interface residues as residues for which RSA differed by a minimum of 10% when calculated for the full biological assembly or for a single chain. All interface residues were included in the analyses presented in the main body of the text, but we excluded interface residues in the analyses presented in S17–S34 Figs.
For each enzyme in the dataset, we fit the following linear models (represented in standard R notation): K ~ d, K ~ RSA, K ~ WCN, K ~ RSA + d, K ~ WCN + d, and K ~ RSA + WCN + d, where K is site-specific evolutionary rate, RSA is relative solvent accessibility, WCN is weighted contact number, and d is distance to the nearest catalytic residue.
All statistical analyses were carried out using the R software package [60]. Linear models are fit to each enzyme individually. After fitting the models, data are then binned for visualization purposes. Plots are generated with ggplot2 [61]. All code and data necessary to reproduce our analyses are available in a Github repository at: https://github.com/benjaminjack/enzyme_distance. Processed enzyme data are also provided as S1 Data. Processed data from the non-enzyme dataset are available as S2 Data. Parameter estimates for each linear model fitted to each enzyme are available in S3 Data.
|
10.1371/journal.ppat.1004406 | Persistence of Virus Reservoirs in ART-Treated SHIV-Infected Rhesus Macaques after Autologous Hematopoietic Stem Cell Transplant | Despite many advances in AIDS research, a cure for HIV infection remains elusive. Here, we performed autologous hematopoietic stem cell transplantation (HSCT) in three Simian/Human Immunodeficiency Virus (SHIV)-infected, antiretroviral therapy (ART)-treated rhesus macaques (RMs) using HSCs collected prior to infection and compared them to three SHIV-infected, ART-treated, untransplanted control animals to assess the effect of conditioning and autologous HSCT on viral persistence. As expected, ART drastically reduced virus replication, below 100 SHIV-RNA copies per ml of plasma in all animals. After several weeks on ART, experimental RMs received myeloablative total body irradiation (1080 cGy), which resulted in the depletion of 94–99% of circulating CD4+ T-cells, and low to undetectable SHIV-DNA levels in peripheral blood mononuclear cells. Following HSC infusion and successful engraftment, ART was interrupted (40–75 days post-transplant). Despite the observed dramatic reduction of the peripheral blood viral reservoir, rapid rebound of plasma viremia was observed in two out of three transplanted RMs. In the third transplanted animal, plasma SHIV-RNA and SHIV DNA in bulk PBMCs remained undetectable at week two post-ART interruption. No further time-points could be assessed as this animal was euthanized for clinical reasons; however, SHIV-DNA could be detected in this animal at necropsy in sorted circulating CD4+ T-cells, spleen and lymph nodes but not in the gastro-intestinal tract or tonsils. Furthermore, SIV DNA levels post-ART interruption were equivalent in several tissues in transplanted and control animals. While persistence of virus reservoir was observed despite myeloablation and HSCT in the setting of short term ART, this experiment demonstrates that autologous HSCT can be successfully performed in SIV-infected ART-treated RMs offering a new experimental in vivo platform to test innovative interventions aimed at curing HIV infection in humans.
| While antiretroviral therapy (ART) can reduce HIV replication, it does not eradicate the virus from an infected individual. Replication-competent viruses persist on ART and our incomplete understanding of these viral reservoirs greatly complicates the generation of a cure for HIV. In this study we performed, for the first time, hematopoietic stem cell transplant (HSCT) in the established model of SIV infection of rhesus macaques (RM). The HSC originating from the bone marrow were collected before SIV infection. After SIV infection, RM were treated with ART for several weeks to reduce viral replication before performing a total body irradiation and a transplant with their own, pre-infection, stem cells. The irradiation eliminated 94–99% of the circulating CD4+ T-cells, the main cell target of HIV/SIV infection. A successful engraftment of the HSC was observed and blood viral reservoirs were drastically reduced. However, when ART was interrupted, a rapid rebound of plasma viremia was observed in two out of three transplanted RM indicating that the massive reset of the hematopoietic compartment was not sufficient to eliminate the total-body virus reservoir in the setting of short term ART. This model of HSCT in SIV-infected RM provides a new platform to investigate HIV eradication strategies.
| The introduction of antiretroviral therapy (ART) has dramatically reduced the morbidity and mortality associated with HIV infection and AIDS. However, currently available ART requires life long treatment with significant potential side effects and a cost that places an inordinate burden on public health systems. While reduction of HIV viral loads below detectable limits is often achieved in ART-treated individuals, a treatment that can eradicate or functionally cure HIV infection remains elusive. Many studies indicate that the key obstacle to cure HIV infection is the presence of a persistent reservoir of latently infected cells that are not eliminated by ART [1], [2]. Thus, interruption of ART consistently results in a rebound of viremia to pre-treatment levels [3], [4]. Several biological aspects of this virus reservoir, including its exact cellular and anatomic origin as well as the mechanisms responsible for its establishment and persistence under ART remain poorly understood. This limited knowledge represents a fundamental barrier to a cure for HIV infection, and novel therapeutic strategies aimed at eliminating the reservoir will likely not be developed until we overcome this barrier.
In 2009 it was reported that an HIV-infected individual with acute myelogenous leukemia treated with myeloablative chemotherapy and allogeneic hematopoietic stem cell transplant (HSCT) from a Δ32ccr5 homozygous donor had remained without detectable HIV replication in the absence of ART for 1.8 years [5], [6]. This first demonstration of a functional cure in this patient was confirmed in 2013 in a follow-up study showing no signs of recrudescent HIV replication and waning of HIV-specific immune responses five years after interruption of ART [7]. More recently, two HIV-infected individuals have been described with prolonged (i.e., 3–8 months) suppression of viremia in absence of ART following allogeneic HSCT from donors homozygous for the wild type ccr5 allele [8], [9]. Similar to the “Berlin patient” described above, these two transplant recipients were themselves Δ32ccr5 heterozygotes. The factors involved in the lack of detectable virus replication after ART interruption in HIV-infected individuals undergoing HSCT are complex, and may include (i) the myeloablative regimen involving various combinations of chemotherapy, immunosuppression, and total body irradiation (TBI); (ii) the deficiency of CCR5 in the transplanted donor cells (in the first case); and (iii) a graft versus host effect that may target cells that are latently infected with HIV (i.e., graft versus reservoir effect). Assessing the relative contribution of these factors will likely provide useful information to define the clinical potential of HSCT as a cure for HIV infection.
SIV infection of non-human primates, such as rhesus macaques (RMs) has been used for over two decades as an in vivo model for studies of HIV pathogenesis, prevention, and treatment [10]. SIV-infected RMs show remarkable similarities to HIV-infected individuals in terms of mechanisms and markers of disease progression, and current ART regimens can fully suppress virus replication in these animals [11]–[14], thus making this model suitable for probing HIV eradication strategies. In this study, we conducted a controlled test of the contribution of pre-transplant myeloablative irradiation to clearance of the viral reservoir in a cohort of RMs infected with a chimeric simian-human immunodeficiency virus (SHIV) and treated with ART. To the best of our knowledge, this is the first time HSCT has been utilized in RMs to investigate viral persistence. The procedure was successfully performed after SHIV infection and ART-induced control of virus replication using HSCs collected prior to infection. While these recipients showed undetectable plasma viremia and low to absent SHIV-DNA in PBMCs after HSCT, interruption of ART resulted in a rapid rebound of virus replication in two out of three animals. The one transplanted RM who maintained undetectable viremia and SHIV-DNA PCR in PBMCs after ART interruption showed low but detectable levels of SHIV-DNA in sorted circulating CD4+ T-cells, spleen and lymph nodes but not in the gastro-intestinal tract or tonsils. Collectively, these results indicate that the massive reset of the lympho-hematopoietic compartment that follows TBI-induced myeloablation was not sufficient to eliminate the total-body virus reservoir in SHIV-infected RMs in the setting of short term ART. However, this study provides a critical foundation upon which to test other potential contributors to a transplant-mediated cure of HIV.
Six RMs were included in this study. All six RMs were males with an average age of 4.2 years (Table 1). Figure 1 shows an overview of the experimental design. Three rhesus macaques (T1, T2, T3) were treated with G-CSF for CD34+ stem cell mobilization followed by HSC collection by leukopheresis and cryopreservation of the collected cells. The six RMs were infected i.v with 104 TCID50 RT-SHIVTC. Starting at week four post-infection all six RMs were initiated on ART. The ART regimen consisted of two nucleotide/side reverse transcriptase inhibitors (PMPA/tenofovir and FTC/emtricitabine), one non-nucleoside reverse transcriptase inhibitor (efavirenz) and one integrase inhibitor (raltegravir). After five to eight weeks on ART, RMs T1-T3 received myeloablative TBI as pre-transplant conditioning. The leukopheresis products collected before infection were infused within 24 hours following the last dose of TBI. Recipients were given a total of 7.3×108+/−1.3×108 total nucleated cells (TNC)/kg which corresponded to 2.9×106+/−1.1×106 CD34+ cells/kg. After successful engraftment of donor cells (five to eleven weeks post-transplant), ART was interrupted in RMs T1-T3 as well as in the control RMs.
As shown in Figure 2A, following experimental infection with RT-SHIVTC the six RMs experienced a rapid, exponential increase in virus replication that peaked at week two post infection (105–107 SHIV-RNA copies/ml plasma). ART initiated at week four after infection drastically reduced plasma viral load to less than 100 copies of SHIV-RNA per ml of plasma. Consistent with prior studies of SIV/SHIV infection in RMs, the absolute number of peripheral CD4+ T-cells was decreased following infection and partially restored on ART (Figure 2B).
The myeloablative TBI resulted in a drastic reduction of the absolute count of blood cells including neutrophils, monocytes, lymphocytes and CD4+ T-cells (Figure 3A). The nadir was observed at day eleven post-TBI for neutrophils (41.6–78.2 neutrophils/µl), day seven post-TBI for monocytes (4.4–14.8 monocytes/µl), and between day one and day five post-TBI for lymphocytes and CD4+ T-cells (54–60 lymphocytes/µl and 6.7–45.5 CD4+ T-cells/µl). Of note, between 94.2 and 99.2% of circulating CD4+ T-cells were eliminated by the TBI (Figure 3A). Engraftment was demonstrated by increasing neutrophil and platelet counts unsupported by transfusion. Neutrophil engraftment was defined as an absolute neutrophil count (ANC) exceeding 500 cells/µl for three consecutive days. The first of these three consecutive days was considered the day of engraftment. As shown in Figure 3B, neutrophil engraftment was successfully achieved between day sixteen and day eighteen post-HSC infusion in the three transplanted animals. During HSCT, the three transplanted animals received platelet and whole blood transfusions for thrombocytopenia prior to platelet engraftment, as well as several antimicrobial prophylactic interventions (Figure S1). Platelet engraftment was defined as a blood platelet count exceeding 20,000 cells/µl in absence of transfusion support for seven consecutive days. According to this definition, platelet recovery was achieved at 42, 22 and 33 days post-transplant for T1, T2 and T3, respectively (Figure 3C).
Following transplantation and engraftment, we observed a rapid increase in the absolute leukocyte count and a slower reconstitution of the circulating CD4+ T-cells (Figure 4A and B). The peripheral reconstitution of CD4+ T-cells appeared to involve peripheral T-cell expansion as evidenced by the increased proportion of circulating CD4+ T-cells expressing the proliferation antigen Ki-67 (Figure S2A). In addition, HLA-DR and CCR5 were increased on CD4+ T-cells following HSCT (Figure S2B,C). Further immunophenotypic analyses revealed a significant increase in the proportion of memory CD4+ T-cells (including memory stem cells, central memory, and effector memory) following transplantation (p = 0.03, Figure S3), similar to previous reports of both autologous and allogeneic HSCT [6], [15]. These results are consistent with CD4+ T-cells recovery occurring primarily through the homeostatic proliferation of memory CD4+ T-cells post-transplant.
A few blips of transient low-level viremia were observed in the plasma of the three transplanted animals immediately after TBI and HSC infusion and while still on ART (Figure 5A). The origin of these transient increases in viral load is not clear, but it may represent release of virus from pre-existing reservoirs in the setting of events of CD4+ T-cell activation during conditioning and the peri-transplant period. With the exception of these transient episodes of viremia, the plasma viral load remained undetectable in all six animals on ART (Figure 5A). Of note, the ART regimen alone reduced the level of SHIV-DNA in PBMCs (i.e., the peripheral viral reservoir) by 1.0–1.5 log in the three control RMs (Figure 5B). In the transplanted animals, the reduction in cell-associated viral DNA was more pronounced, with two RMs showing levels of SHIV-DNA in PBMCs below the limit of detection and one RM (T1) close to this level (as low as 130 copies/million PBMC, Figure 5B). The normalization of the cell-associated SHIV-DNA level to the CD4+ T-cells counts suggest a decrease in the frequency of infection of these cells post-transplant (Figure 5B).
ART was interrupted after stem cell engraftment (between 78 and 128 days post-initiation, Table 1). As expected, a rapid viral rebound was observed in the plasma of the three control animals as early as one week post ART cessation (Figure 5A). Two out of the three transplanted animals also experienced a rapid plasma viral rebound post ART interruption. The remaining transplanted animal (T2) maintained an undetectable plasma viral load at two weeks post ART interruption (Figure 5A). Unfortunately, further time-points were not analyzed in this animal as he was euthanized due to progressive renal failure. As shown in Figure 5B, ART interruption led to an increase of the SHIV-DNA levels in the PBMCs of the two transplanted RMs who also experienced a plasma viral rebound. This rebound in PBMC SHIV-DNA was observed at the first assessment post-ART interruption in both animals (day 28 for T1 and day 15 for T3). Of note, no SHIV-DNA was detected in the PBMCs of RM T2 who also maintained undetectable plasma viral load at two weeks after ART interruption. However, further analyses of this animal revealed low but detectable levels of SHIV-DNA in sorted peripheral CD4+ T-cells obtained at the same time-point (i.e., two weeks after ART interruption at necropsy) (Figure 5C).
Several tissues were collected at necropsy including ileum, jejunum, colon, rectum, superficial and mesenteric lymph nodes as well as tonsils. SHIV-DNA levels in cell suspensions obtained from these tissues were quantified by PCR. As shown in Figure 6, low levels of SHIV-DNA were detected in the spleen and lymph nodes of the transplanted RM who maintained an undetectable peripheral viral load post ART interruption (T2) but not in the tonsils or gut compartments. Of note, we were able to detect SHIV-DNA in the gut and tonsils of the other two transplanted RMs (T1 and T3) who exhibited a rapid rebound of viremia after ART interruption.
The apparent cure of HIV infection in the “Berlin patient” [5]–[7] has energized efforts to understand the mechanisms of virus persistence despite ART-mediated suppression of virus replication. The factors thought to be involved in the favorable outcome of the Berlin patient following HSCT include (i) the myeloablative conditioning regimen; (ii) the donor's homozygosity for Δ32ccr5; and (iii) the graft versus host effect. In this test-of-concept study of autologous HSCT in SHIV-infected RMs we interrogated the relative contribution of a myeloablative conditioning regimen in eliminating the persistent reservoir of latently infected cells. To the best of our knowledge this is the first time that a study of similar design has been conducted.
The key findings of this study are the following: (i) autologous HSCT using apheresis products collected prior to infection is feasible in SHIV-infected RMs; (ii) as expected, the myeloablative TBI used for conditioning induced a massive reset of the lympho-hematopoietic compartment, consequently resulting in the depletion of 94.2–99.2% of circulating CD4+ T-cells; (iii) animals receiving autologous HSCT under ART exhibited a prompt and pronounced decline in the peripheral blood viral reservoir (with undetectable SHIV-DNA in PBMCs in two out of three RMs) and maintained undetectable SHIV-RNA viremia with the exception of a few minor blips; (iv) two of the three transplanted RMs showed a very rapid rebound of viremia after ART interruption; and (v) the third transplanted RM, who was sacrificed for clinical reasons at day fourteen post ART interruption, had no detectable virus in plasma, PBMCs, tonsils, and GI tract, low but detectable levels of SHIV-DNA in sorted peripheral CD4+ T-cells and lymph nodes, and moderate levels of SHIV-DNA in the spleen.
Due to many logistical challenges of this experiment we chose to conduct the study in a temporally compressed fashion, with 37–53 days of ART before autologous HSCT, and interruption of ART after hematopoietic reconstitution, rather than prolonged continuation of therapy. This study was therefore designed to determine the impact of myeloablative irradiation on the viral reservoir, rather than the impact of prolonged viral suppression in conjunction with myeloablation. It is therefore possible that a similarly designed study, in which ART is maintained for a significantly longer period both before and after autologous HSCT, would have a different outcome, possibly demonstrating a more dramatic effect of autologous HSCT on the persistent reservoir of latently infected cells. Moreover, we cannot rule our the possibility that the level of virus suppression achieved by the short-term ART regimen in this experiment might not be as complete as what is observed in HIV-infected individuals on long-term ART. In this model of SHIV-infected RM, 5 to 7 weeks on ART pre-transplant may have been insufficient to fully suppress viral replication and the transient low-level viremia observed immediately post-transplant could be attributed to an insufficient period of ART pre-transplant. However, similar viral blips were observed in one patient who received allogeneic stem cell transplant after many years on combined ART [8]. Although the origin of these transient blips is unknown, it may represent release of the virus from latently infected cells in the setting of cell activation during conditioning and the peri-transplant period. In keeping with this hypothesis, it should be noted that in our study the post-transplant period was characterized by an expansion of CD4+ T-cells expressing CCR5 as well as proliferation and activation markers. Together with the observed increased proportion of memory CD4+ T-cells post-transplant, these results suggest that the CD4+ T-cell compartment recovered primarily through homeostatic proliferation of memory CD4+ T-cells.
The myeloablative TBI used for conditioning resulted in the depletion of 94.2–99.2% of circulating CD4+ T-cells. Unfortunately, due to the clinical challenges of this innovative experiment, no tissue biopsies could be obtained immediately post-transplant to evaluate the TBI-induced CD4+ T-cell depletion in tissues. However, this study shows that myeloablative TBI and autologous HSCT did not prevent a rebound of viremia post-ART interruption in two out of three RMs despite relatively early ART initiation (day 28 post-infection). Moreover, while the SHIV-DNA level in PBMCs was undetectable or close to undetectable post autologous HSCT, it rapidly rebounded after ART interruption to levels that were similar or higher than those observed in the control animals at the same time-point. While in the third animal (T2) there was no sign of virus present in the plasma, PBMCs, and various tissues at the time of necropsy, this RM had to be sacrificed due to kidney failure at day fourteen after ART interruption making the interpretation of these data somewhat difficult. Of note, this study was not designed to identify the cellular and anatomic sources of the rapid plasma viral rebound observed in two transplanted RMs following ART interruption. Determining the relative contribution of tissue CD4+ T-cells, macrophages, and potentially other sources represents an important area for future investigation, amenable for interrogation with this model.
We acknowledge a number of limitations in our study including the small number of animals and the foreshortened time line involved. However, the demonstrated feasibility of this test-of-concept study in a non-human primate model of AIDS virus infection is per se an important result given the extreme complexity of the experimental protocol. The RMs included in this study underwent a series of procedures that have been only rarely, if ever, used in the same animal, including stem cell mobilization and harvesting by apheresis, RT-SHIV infection, daily four-drug ART administration, total body irradiation, re-infusion of HSCs, repeated platelet transfusions, and receipt of several antimicrobial prophylaxes. The feasibility of HSCT in SIV- or SHIV-infected RMs suggests, in our view, that further studies using this model in conjunction with longer term ART as well as additional interventions aimed at purging both the peripheral blood and lymphoid tissue-based viral reservoirs will provide critical information for the requirements to cure HIV infection in humans.
With respect to our understanding of the mechanisms responsible for “curing” HIV infection in the Berlin patient, our study supports the hypothesis that myeloablative TBI can cause a significant decrease in the viral reservoir in circulating PBMCs, even though it was not sufficient to eliminate all reservoirs. While the conditioning regimen in the Berlin patient also included antithymocyte globulin and chemotherapy, the use of a Δ32ccr5 homozygous donor and/or the presence of graft versus host disease likely played a significant role in that clinical context. The importance of graft versus host disease that effectively results in a “graft versus reservoir” effect is also emphasized by the recent observation of two HIV-infected patients in which a prolonged (i.e., 3–8 months) period of undetectable viremia in absence of ART was observed after allogeneic HSCT from donors with wild-type ccr5 alleles [9], although these patients did eventually develop rebound of viremia [16]. Future studies of allogeneic HSCT in SIV- or SHIV-infected RMs in the presence or absence of gene therapy interventions to knock out ccr5 would be very informative in this regard, and may elucidate the mechanism of the sustained cure seen in the Berlin patient but not the above mentioned recipients of donor cells wild type for ccr5.
In conclusion, we have conducted the first test-of-concept study of myeloablative irradiation and autologous HSCT in ART-treated SHIV-infected RMs. This experiment demonstrated that autologous HSCT is a feasible intervention that can lead to a marked reduction of the virus reservoir in the peripheral blood, and can be used as an experimental in vivo platform to test innovative interventions aimed at curing HIV infection in humans.
This study was conducted in strict accordance with USDA regulations and the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, and were approved by the Emory University Institutional Animal Care and Use Committee (Protocol # YER-20000373-061714). SIV-infected animals were housed in standard non-human primate cages, received standard primate feed as well as fresh fruit and enrichment daily, and had continual access to water. Cages also contained additional sources of animal enrichment including objects such as perching and other manipulanda. Animal welfare was monitored daily. Appropriate procedures were performed to ensure that potential distress, pain, or discomfort was alleviated. The sedatives Ketamine (10 mg/kg) or Telazol (4 mg/kg) were used for blood draws and biopsies. Euthanasia of RMs, using Pentobarbital (100 mg/kg) under anesthesia, was performed only when deemed clinically necessary by veterinary medical staff and according to IACUC endpoint guidelines.
Six Indian RMs (Macaca mulatta), with exclusion of Mamu B*08 and B*17 positive animals, were included in this study. All animals were housed at the Yerkes National Primate Research Center (Atlanta, GA) and treated in accordance with Emory University and Yerkes National Primate Research Center Institutional Animal Care and Use Committee regulations.
Autologous HSCs were harvested at two separate time points in each animal using our previously described apheresis procedure [17]. Animals were prepared for leukopheresis with epoeitin alfa (nine doses of 150 mg/kg, Amgen), given in the two months prior to leukopheresis to increase red cell mass and thus increase the safety of the leukopheresis procedure and filgastrim (G-CSF, 50 mg/kg intramuscularly daily to a maximum of 300 mg, Amgen) for six days prior to leukopheresis to mobilize HSCs as previously described [18]. The leukopheresis was analyzed for cell content and then cryopreserved in 10% DMSO using standard clinical techniques. Both apheresis units were infused into the transplant recipient within 24 hours of the completion of TBI.
The leukopheresis products were analyzed by flow cytometry prior to cryopreservation for the total nucleated cell dose, the CD34+ cell dose, CD3+ T-cell dose, CD4+ T-cell dose, CD8+ T-cell dose, and the CD20+ B-cell dose using the following antibodies; CD3 (clone SP34-2), CD34 (clone 563), CD45 (clone D058-1283), CD8 (clone RPA-T8) from BD Biosciences; CD20 (clone 2h7), CD4 (clone OKT4) from eBioscience.
The RMs were intravenously (i.v.) infected with 10,000 50% tissue culture infective doses (TCID50) of RT-SHIVTC. The virus stock was provided by Dr. Tom North (Emory University) and prepared as previously described [19], [20]. The RT-SHIV used for this study had the T-to-C substitution at position 8 of the SIV tRNA primer binding site which is necessary for high replication of RT-SHIV in vivo [21].
Efavirenz was provided by Bristol-Myers Squib, raltegravir was provided by Merck, and emtricitabine (FTC) and tenofovir (PMPA) were provided by Gilead Sciences. Efavirenz was fed at 200 mg per day by mixing the contents of a 200 mg capsule into food. Raltegravir was fed at 100 mg twice daily by mixing the drug into food. Stock solutions of FTC were prepared in phosphate-buffered saline (PBS, pH 7.4). PMPA was suspended in distilled water, with NaOH added to a final pH of 7.0. FTC and PMPA stocks were filter sterilized and stored at 4°C. These drugs were administered subcutaneously, at a dose of 30 mg/kg of body weight once daily. Drug dosages were adjusted weekly according to body weight.
The pre-transplant preparative regimen consisted of myeloablative TBI to a total dose of 10.8 Gy, given in three divided fractions of 3.6 Gy each (at a rate of 7.5 cGy/minute) using a Varian Clinac 23 EX (Varian). Irradiation took place on days −2, −1, and 0 (the day of transplant), with the final dose of irradiation given just prior to infusion of the first of two leukopheresis products.
Animals were treated with the following empiric antimicrobial agents in the peri-transplant period, as previously described [17], [22]. (Figure S1): (1) Polymixin B (1,000,000 units orally daily, Ben Venue Laboratories, Inc) and neomycin sulfate (500 mg orally daily, Teva Pharmaceuticals). Dosing of both agents was begun on day −7 and continued until neutrophil engraftment (Absolute neutrophil count >500 cells/µl for three consecutive days). (2) Enrofloxacin (7 mg/kg intramuscularly daily, Bayer Healthcare) starting on day −1 and continuing until neutrophil engraftment. (3) Fluconazole (5 mg/kg orally daily, Pfizer) starting on day −1 and continuing until neutrophil engraftment. (4) Cidofovir (5 mg/kg i.v., Gilead) starting on day +6 and continuing once weekly as clinically tolerated, to prevent CMV reactivation. Cidofovir was given to transplant recipients 1 and 2. However, because we observed significant increases in serum creatinine in these recipients, the third transplant recipient was treated with oral valganciclovir (60 mg twice daily, Genentech), which was begun after neutrophil engraftment was observed.
Transplanted animals received both platelet rich plasma and whole blood (irradiated at 2200 rad prior to transfusion) to treat thrombocytopenia (platelet count <50×106/ml) or anemia (hemoglobin <10 g/dl) or with the development of clinically significant bleeding. Blood product support adhered to ABO antigen matching principles.
EDTA-anticoagulated blood samples were collected regularly and used for a complete blood count, routine chemical analysis and immunostaining, with plasma separated by centrifugation within 1 h of phlebotomy. PBMCs were prepared by density gradient centrifugation. CD4+ T-cells were negatively selected from frozen PBMCs using magnetically labeled microbeads and subsequent column purification according to the manufacturer's protocol (Miltenyi Biotec). Tissue samples including ileum, jejunum, colon, tonsils and mesenteric and superficial lymph nodes were collected post-mortem. After two washes in RPMI and removal of connective and fat tissues, gut tissues were cut in small pieces and lymph nodes and tonsils were grinded using a 70-µm cell strainer. Gut cells were isolated by digestion with collagenase and DNase I for 2 h at 37°C and then passed through a 70-µm cell strainer. The cell suspensions obtained were washed and immediately used for immunostaining, cryopreserved or lysed in RLT+ buffer and stored at −80°C until use.
Plasma viral quantification was performed as described previously [23]. DNA was extracted from PBMCs, sorted peripheral CD4+ T-cells, and tissue cell suspensions using the Blood DNA Mini Kit (QIAGEN). Quantification of SIVmac gag DNA was performed as previously described on the extracted cell-associated DNA by quantitative PCR using the 5′ nuclease (TaqMan) assay with an ABI7500 system (PerkinElmer Life Sciences). The sequence of the forward primer for SIVmac gag was 5′-GCAGAGGAGGAAATTACCCAGTAC-3′; the reverse primer sequence was 5′-CAATTTTACCCAGGCATTTAATGTT-3′; and the probe sequence was 5′-6 FAM-TGTCCACCTGCCATTAAGCCCGA-TAMRA-3′. For cell number quantification, quantitative PCR was performed simultaneously for monkey albumin gene copy number. All PCR were performed in duplicate with 10,000 cell equivalent per reaction with a limit of detection of 1 copy per reaction.
Multicolor flow cytometric analysis was performed on whole blood or frozen PBMCs using predetermined optimal concentrations of the following fluorescently conjugated mAbs: CD3-PacBlue or -APC-Cy7 (clone SP34-2), CD95-PE-Cy5 (clone DX2), Ki-67-AF700 (clone B56), HLA-DR-PerCP-Cy5.5 (clone G46-6), CCR7-PE-Cy7 (clone 3D12), CCR5-PE or -APC (clone 3A9), CD45RA-FITC (clone L48), Biotin-CD122 (clone Mik-β3) from BD Biosciences; CD8-BV711 (clone RPA-T8), CD4-APC-Cy7 or -BV650 (clone OKT4), Streptavidin-PE from Biolegend, and CD28-ECD (clone CD28-2) from Beckman-Coulter. Flow cytometric acquisition and analysis of samples was performed on at least 100,000 events on an LSRII flow cytometer driven by the FACSDiva software package (BD Biosciences). Analyses of the acquired data were performed using FlowJo Version 10.0.4 software (TreeStar).
For the comparison of SHIV-DNA in sorted CD4+ T-cells in transplanted and control RMs the nonparametric Mann-Whitney U test was used. For the comparison of the proportion of memory CD4+ T-cells before and after transplant, a Wilcoxon matched-pairs signed rank test was used. Statistical significance was set at p<0.5. All analyses were performed using GraphPad Prism v4.0.
|
10.1371/journal.pntd.0001895 | Chrysomya putoria, a Putative Vector of Diarrheal Diseases | Chrysomya spp are common blowflies in Africa, Asia and parts of South America and some species can reproduce in prodigious numbers in pit latrines. Because of their strong association with human feces and their synanthropic nature, we examined whether these flies are likely to be vectors of diarrheal pathogens.
Flies were sampled using exit traps placed over the drop holes of latrines in Gambian villages. Odor-baited fly traps were used to determine the relative attractiveness of different breeding and feeding media. The presence of bacteria on flies was confirmed by culture and bacterial DNA identified using PCR. A median of 7.00 flies/latrine/day (IQR = 0.0–25.25) was collected, of which 95% were Chrysomya spp, and of these nearly all were Chrysomya putoria (99%). More flies were collected from traps with feces from young children (median = 3.0, IQR = 1.75–10.75) and dogs (median = 1.50, IQR = 0.0–13.25) than from herbivores (median = 0.0, IQR = 0.0–0.0; goat, horse, cow and calf; p<0.001). Flies were strongly attracted to raw meat (median = 44.5, IQR = 26.25–143.00) compared with fish (median = 0.0, IQR = 0.0–19.75, ns), cooked and uncooked rice, and mangoes (median = 0.0, IQR = 0.0–0.0; p<0.001). Escherichia coli were cultured from the surface of 21% (15/72 agar plates) of Chrysomya spp and 10% of these were enterotoxigenic. Enteroaggregative E. coli were identified by PCR in 2% of homogenized Chrysomya spp, Shigella spp in 1.4% and Salmonella spp in 0.6% of samples.
The large numbers of C. putoria that can emerge from pit latrines, the presence of enteric pathogens on flies, and their strong attraction to raw meat and fish suggests these flies may be common vectors of diarrheal diseases in Africa.
| While it is well recognized that the house fly can transmit enteric pathogens, here we show the common African latrine fly, Chrysomya putoria, is likely to be an important vector of these pathogens, since an average latrine can produce 100,000 latrine flies each year. Our behavioral studies of flies in The Gambia show that latrine flies are attracted strongly to human feces, raw beef and fish, providing a clear mechanism for faecal pathogens to be transferred from faeces to food. We used PCR techniques to demonstrate that these flies are carrying Shigella, Salmonella and E. coli, all important causes of diarrhea. Moreover our culture work shows that these pathogens are viable. Latrine flies are likely to be important vectors of diarrheal disease, although further research is required to determine what proportion of infections are due to this fly.
| Controlling diarrheal deaths is crucial to achieve Millennium Development Goal 4; reducing mortality in children under five years by two thirds between 1990 and 2015 [1]. Diarrhea is the second leading cause of death in this age group and is responsible for killing about 1.5 million children each year [2]. In sub-Saharan Africa, diarrhea is caused by a wide range of pathogens including; diarrheagenic (enterotoxigenic [ETEC], enteropathogenic [EPEC] and enteroaggregrative [EAEC]) E. coli, serovars of Salmonella enterica, Shigella spp., Campylobacter spp., Vibrio spp. and Aeromonas spp. One important route of infection is thought to be the mechanical transmission of diarrheal pathogens by flies [3]–[5]. However, despite the long association between flies and pathogens [4] the evidence incriminating flies as vectors of diarrheal pathogens remains weak simply because few adequately controlled studies have been conducted. Moreover, whilst human feces are a rich source of bacteria it is not guaranteed that flies breeding in feces or feeding on feces will be contaminated with high bacterial loads. For example most Salmonella spp are destroyed by passage through the acid midgut of the blowfly larva, Calliphora vicinia [6]. Bacteria can also be lost during metamorphosis from larva to pupa inside the puparium. At this stage bacteria in the fore and hindgut of the larva lie inside the puparium, outside of the pupa [7], and consequently the fly may emerge from the puparium sterile [4]. Moreover, the normal gut biota of a fly can eliminate potential pathogens as seen with Salmonella spp when it is eliminated in the prepupal stage [4].
Of the large number of historical studies investigating the role of flies in the transmission of enteric pathogens, the most convincing was carried out in Texas in communities with high levels of diarrhea [8]. In that context, DDT-sprayed communities had 60% fewer flies (mainly Musca domestica, the house fly), 46% lower incidence of diarrhea and a 49% reduction in deaths from diarrhea and enteritis in children under two years old, than unsprayed communities. When the treatment was crossed over, there was a corresponding decline in fly numbers and diarrhea incidence in sprayed communities.
There have been only two recent intervention trials investigating the role of flies in the transmission of diarrheal diseases. Control of M. domestica was evaluated in a cross-over study at two military camps in Israel, where fly traps supplemented with insecticidal spraying were used in one camp and not the other. Where the flies were controlled the incidence of diarrheal diseases dropped by 42% and shigellosis decreased by 85% [9]. A community-randomized trial in Pakistan showed that fly control using insecticide reduced the incidence of childhood diarrhea by 23% during periods of high fly densities, where nearly all flies were M. domestica [10]. These findings clearly incriminate M. domestica as a major vector of diarrheal diseases. In addition, given the cosmopolitan distribution of this fly and the high numbers of adult flies found around human habitation, research on fly-borne diarrheal diseases is dominated by this species.
Exploring the role of other synanthropic flies that may also act as vectors of diarrheal pathogens is also important. Species of Chrysomya may be vectors since they are strongly associated with human feces and food. Like M. domestica, Chrysomya putoria is found across Africa. It was found to be the major fly species in a variety of latrine types in Botswana and Tanzania [11]. In The Gambia, an average pit latrine produces over 100,000 flies each year [12]. Of these flies, 97.8% were identified as Chrysomya spp. and only 1.2% M. domestica. Since it is common in African markets to see large numbers of adult Chrysomya spp on fresh meat and fish we hypothesised that they may mechanically transfer diarrheal pathogens from human feces to food. We explored this hypothesis using odor-baited traps in The Gambia to determine the attractiveness to flies of a variety of breeding and feeding media common in rural villages and by searching for the presence of diarrheal pathogens on flies collected from pit latrines in rural villages. Understanding the routes of diarrhea-pathogen transmission is critically important since it may suggest new opportunities for the control of these life-threatening diseases.
This study combines a number of different research methodologies undertaken in the field at different times. We describe the numbers of flies emerging from latrines, their attractiveness to different baits and examine whether they were contaminated with fecal pathogens. These studies are cross-sectional studies, rather than longitudinal ones.
Collections were carried out in villages in the Upper River Region of The Gambia between June 2011 and February 2012. This is an area of open Sudanian savannah with a rainy season from June to October followed by a long dry season. Most people live in small rural villages in houses with mud or cement walls and thatched or metal roofs. Toilets are usually pit latrines, although open defecation also occurs. Adult and immature flies were collected from different experiments, locations and seasons to identify common Chrysomya spp.
Odor-baited fly traps were used to assess the relative attractiveness of different breeding and feeding media commonly found in rural Gambia. These studies were carried out at the Medical Research Council's Field Station at Basse Santa Su (13°18′57.96″N, 4°12′31.90″W). For the first experiment the potential attractants were feces from: (1) two Gambian child aged about two years old, (2) three local dogs, (3) a goat, (4) a horse, (5) a cow and (6) a calf. The same animals and children were used throughout. In the second experiment the potential attractants used were: (1) cooked rice, (2) uncooked rice, (3) cut mango, (4) raw beef, (5) raw fish (S. batensoda), and (6) human feces from two year old children served as a positive control.
In each experiment 50 g of each media was randomly allocated to six traps positioned along a straight line at 2 m intervals. Traps were rotated between locations on different trapping occasions using a Latin Square design. For the experiment using only feces, traps were left at each position for about seven hours and the entire procedure repeated on 12 occasions. The experiment using different foods was carried out six times for four hours each time. The experiments were carried out for different periods to adjust for differences in monthly fly numbers; for seven hours when fly numbers were low and for four hours when they were high. This sample size was sufficient to detect a 33% difference in fly numbers between traps, at the 5% level of significance and 80% power based on our pilot work (Epi Info version 7.9.7). This a priori power calculation was designed to detect large differences in attractiveness between baits since we wanted to identify the major fly attractants. We considered that differences of 33% or more would be sufficient for this purpose.
General linear modelling was used to account for the variation in fly numbers due to trap bait, position of trap and replicate using SPSS version 19.0. Comparisons between different baits were made using Bonferonni corrections to account for multiple comparisons. Comparisons between proportions were made using the Chi-square test in Epi Info version 7.9.7. Population prevalence rates of infection were estimated from pooled samples based on the bias-corrected maximum likelihood estimation, using a skewness-corrected score for estimating 95% confidence intervals. An excel add-in available from the West Nile Virus website of the Centers for Disease Control and Prevention was used to compute these values [23].
Written informed consent was provided by householders for fly collections from latrines and by the parents and guardians of children for donation of stool samples. Ethical approval for this study was provided by the Gambian Government/Medical Research Council's Laboratories Joint Ethics Committee in The Gambia (SCC 1234v and 21238v2) and the London School of Hygiene and Tropical Medicine's Ethics Committee (010/243 and 5984). An animal protocol was not necessary since no animals were handled during the study. Animal faeces were collected from the ground.
A total of 4,572 flies were collected from 62 pit latrines during July and November (Figure 1; median = 7.00 flies/latrine/day, interquartile range, IQR = 0.0–25.25). We found 13% (n = 8) of latrines sampled produced 85% of the total flies (n = 3689). 60% (n = 37) produced less than 10 flies. Of the 4,034 flies collected from 31 latrines in July, 94.72% were C. putoria, 5.21% Musca spp., 0.05% C. marginalis and 0.02% Sacrophaga spp. Of the two morphologically similar species of adult Chrysomya, C. putoria was the dominant species in both the wet season (99.2%, 891/898; Table 2) and dry season (99.1%, 1173/1183). Since nearly all flies collected were C. putoria we refer, for simplicity, only to this species hereafter. Overall 80.9% of the C. putoria were females, with a greater proportion of females emerging from latrines in the dry season (91.5%) than wet season (74.5%, χ2 = 6.017, p = 0.0142) and captured in fish-baited traps in the wet season (91.7%) compared with the dry season (77.1%, χ2 = 12.44, p<0.001).
The larvae of Chrysomya spp collected in pit latrines were all C. putoria. None of the samples included larvae of C. albiceps, which are highly distinctive due to their fleshy processes [18]. The percentage of Psychodidae compared to C. putoria was 7.75% (30/387) in the wet season and 61.47% (461/750) in the dry season. These flies were found mainly in different latrines from C. putoria, preferring latrines with relatively clear water compared to the more solid contents.
The number of C. putoria collected from traps baited with the feces of different species varied significantly (Figure 2, F = 18.04, p<0.001). There was a significant difference between human feces and goat, horse, cow and calf feces (p<0.001). More flies were collected from traps with feces from young children (median = 2.5, IQR = 1.0–8.5) and dogs (median = 1.0, IQR = 0.0–12.0) than from herbivores (median = 0.0, IQR = 0.0–0.0; goat, horse, cow and calf; p<0.001). There were significant differences in the number of C. putoria collected from different foods and human feces (Figure 3, F = 30.01 p<0.001). Most were collected in traps baited with raw beef, human feces and raw fish. There was no difference between human feces and beef (z = −0.943, P = 0.345). Flies were strongly attracted to raw meat (median = 44.5, IQR = 26.2–143.0) compared with fish, cooked and uncooked rice, and mangoes (median = 0.0, IQR = 0.0–0.0; p<0.001).
Our findings support the hypothesis that C. putoria is a putative vector of diarrheal diseases since 21% of this species emerging from latrines carried fecal bacteria, including pathogens that cause diarrhea. Moreover, since these flies are also strongly attracted to raw meat and fish, they are likely to contaminate foodstuffs with bacterial pathogens. Although meat and fish are not eaten raw in The Gambia, we speculate that pathogens could be transferred to the mouth and other foodstuffs after handling contaminated meat.
Chrysomya putoria is known as the tropical African latrine blowfly [19], and was the dominant fly species emerging from pit latrines in our study. They are common in sub-Saharan Africa [24], with recent incursions into South America [25]. Chrysomya putoria have been reported as the dominant fly from pit latrines in Dar es Salaam, Tanzania [26], Kinshasa, Democratic Republic of Congo [27], Sudan [28] and Zimbabwe [29]. The first, and only, previous record of C. putoria in The Gambia was from one ‘modern’ lavatory in Sukuta, on the coast, in July 1952 [30]. The adults of C. putoria are morphologically similar to C. albiceps, and we think that recent descriptions of C. albiceps emerging from pit latrines in The Gambia [12] are likely to be mistaken; meaning C. putoria is the dominant fly in the country, not C. albiceps. Chrysomya putoria are known to breed in wet feces and can liquefy large fecal masses [31], which help break down feces and increase the longevity of latrines. We found 13% of latrines produced 85% of the flies, which is roughly comparable with Pareto's principle, that 80% of disease transmission results from 20% of hosts, or in this case, latrines [32]. It may be that the large number of latrines that did not produce flies were too dry although this requires further investigation that may lead to identifying potential fly interventions.
Our choice of experiments demonstrates that C. putoria are attracted strongly to both human and dog feces. This attractiveness is graphically illustrated when finding human feces deposited in the open; these, and the nearby vegetation, are usually covered rapidly by C. putoria, (Figure 4). We also demonstrated that C. putoria is attracted to raw beef. Although our experiments did not find that flies were attracted strongly to fish, this was probably because in this choice experiment flies preferred feces and raw meat. We know from other work that fish can be used as bait for C. putoria. Both meat and fish are common and major sources of protein in rural Gambia. Our findings support the general view that C. putoria is a blowfly attracted to feces and carrion [24], and these sources of protein are essential for egg development [19].
We demonstrated by PCR that pools of Chrysomya collected emerging from pit latrines were contaminated with human fecal bacteria, including the important human pathogens: E. coli (27%), EAEC (2%), Salmonella spp (1.4%) and Shigella spp (0.6%). We also showed that these bacteria are likely to be viable since we were able to culture E. coli directly from 21% of plates where live flies had been introduced. These levels of infection are comparable to the 0.5% Chlamydia trachomatis infection rate found for Musca sorbens [33], which is responsible for 56% of trachoma cases in Gambian children [34]. We know of only one study reporting enteric pathogens (polio, coxsackievirus, E. coli and Salmonella spp.) from C. putoria and this was from Madagascar over 50 years ago [35].
There are limitations to our study design. We cannot exclude the possibility that cross-contamination occurred between flies collected in the same trap or from the trap itself thereby inflating the true bacterial carriage prevalence rate. Moreover, the carriage of pathogens does not prove conclusively that Chrysomya spp are mechanical vectors of diarrheal pathogens. This evidence must come from a randomised-controlled trial. Nonetheless, we demonstrated that only culture plates with flies collected from latrines were contaminated with fecal bacteria, not those exposed to flies grown without feces, nor those plates which were manipulated in a way to simulate fly introduction.
Although we suggest that for C. putoria the mechanism of transmission is indirect, via the contamination of foodstuffs, rather than direct transmission, as is the case for trachoma, the infection rates found in our study suggest that these flies are likely to be common mechanical vectors of diarrheal pathogens in The Gambia and other parts of sub-Saharan Africa. Moreover, they can be found emerging in prodigious numbers from latrines at a rate of 74/flies/latrine/day reported in our pilot study and 274/flies/latrine/day reported by Emerson and colleagues during a year-long survey of latrines in The Gambia [12]. If each latrine produces 100,000 flies each year [12], with most compounds containing at least one latrine, this represents an enormous capacity to transfer fecal pathogens to food. Calliphorids, like C. putoria, are generally long lived with laboratory colonies surviving an average of three to four weeks [36] and many may return to feed and lay eggs on feces during their life time increasing their opportunity to be contaminated with diarrheal pathogens. Coupled with E. coli's persistence outside a host of more than one month [37], these flies have the potential to be vectors for substantial periods. This is further enhanced by the long distance dispersal of these flies normally ranging from 0–6 km, with a maximum dispersal of >16 km in 12 days [38]. Our findings suggest that around 1,000 infected flies will land on an average piece of meat (assuming regular replacement) over one year, assuming a 2% carriage of enteric pathogens, with a median of 44.5 flies landing on 50 g of raw meat in 4 hours, assuming fly contact was at a similar rate for 12 h each day (i.e. [44.5×3×365]×0.02). Whilst we acknowledge that cooking will kill bacteria on meat and fish, we suggest that those handling contaminated food pass the bacteria on to others on their hands and common household items. Dirty hands are a well-known route of transmission and it has been shown that hand-washing can reduce the risk of diarrheal incidence by 48% [39].
How important might C. putoria be as a vector contributing to diarrheal diseases mortality? An examination of the seasonality of deaths from acute gastroenteritis [40] and the seasonality of numbers of C. albiceps (we suggest is C. putoria) collected from latrines in different parts of The Gambia in different years (Figure 5) shows that both fly numbers and diarrheal deaths rise and fall at the same time of year. There are two major qualifications with this analysis. Firstly, association does not prove causality, and there may be other reasons for the seasonality in diarrhea deaths. Secondly, the data may be biased since they were drawn from studies conducted in different places at different times. Nonetheless the shared seasonality of flies and diarrheal deaths, merits further investigation.
Crucially, if C. putoria are important vectors, their control may be relatively simple and targeted at the source of the problem; the pit latrine. Targeted control of the small number of most prolific latrines will dramatically reduce major breeding sites. The construction of Ventilated Improved Pit latrines (VIPs) [41], the use of insecticides [42] or using the odors of the latrine as a natural bait (unpublished data) may lead to dramatic reductions in fly numbers. Nonetheless this will only be effective if combined with health programmes that reduce open defecation. At present 5% of the Gambia's rural population still practice open defecation [43]. Moreover, in recent focus group discussions with village latrine users in Kundam Demba (unpublished data), it was revealed that children under five years old were not allowed to use their latrines. Mothers forbade their children using the latrine, fearing they would make it dirty “especially when they have diarrhea”. Another study in The Gambia suggested that it is the last-born child that is prevented from using the latrines, rather than the actual age of the child [44]. Young children were expected to openly defecate and the mothers to dispose of the feces in their latrines. This left a period between defecation and disposal when the feces would be exposed to the open air and flies. Open defecation is likely to increase feeding and breeding opportunities for Chrysomya spp therefore increasing their potential as diarrheal vectors. Any control programmes to reduce fly numbers should also tackle open defecation.
Here we suggest that C. putoria may be of major public health importance as mechanical vectors of diarrheal pathogens. Since this fly is widely distributed across sub-Saharan Africa and parts of South America this hypothesis should be of wider importance since diarrheal diseases are a major cause of childhood morbidity and mortality in these regions. Whilst transmission of diarrheal diseases by flies is typically associated with the house fly, M. domestica, here we suggest that the strong association between the blowfly C. putoria and human feces, combined with its strong attraction to raw meats, may make it a putative mechanical vector of enterovirulent pathogens. Intervention trials are needed to establish the role of C. putoria as a mechanical vector of diarrheal pathogens.
|
10.1371/journal.pcbi.1005914 | Model-driven discovery of long-chain fatty acid metabolic reprogramming in heterogeneous prostate cancer cells | Epithelial-mesenchymal-transition promotes intra-tumoral heterogeneity, by enhancing tumor cell invasiveness and promoting drug resistance. We integrated transcriptomic data for two clonal subpopulations from a prostate cancer cell line (PC-3) into a genome-scale metabolic network model to explore their metabolic differences and potential vulnerabilities. In this dual cell model, PC-3/S cells express Epithelial-mesenchymal-transition markers and display high invasiveness and low metastatic potential, while PC-3/M cells present the opposite phenotype and higher proliferative rate. Model-driven analysis and experimental validations unveiled a marked metabolic reprogramming in long-chain fatty acids metabolism. While PC-3/M cells showed an enhanced entry of long-chain fatty acids into the mitochondria, PC-3/S cells used long-chain fatty acids as precursors of eicosanoid metabolism. We suggest that this metabolic reprogramming endows PC-3/M cells with augmented energy metabolism for fast proliferation and PC-3/S cells with increased eicosanoid production impacting angiogenesis, cell adhesion and invasion. PC-3/S metabolism also promotes the accumulation of docosahexaenoic acid, a long-chain fatty acid with antiproliferative effects. The potential therapeutic significance of our model was supported by a differential sensitivity of PC-3/M cells to etomoxir, an inhibitor of long-chain fatty acid transport to the mitochondria.
| The coexistence within the same tumor of a variety of subpopulations, featuring different phenotypes (intra-tumoral heterogeneity) represents a challenge for diagnosis, prognosis and targeted therapies. In this work, we have explored the metabolic differences underlying tumor heterogeneity by building cell-type-specific genome-scale metabolic models that integrate transcriptome and metabolome data of two clonal subpopulations derived from the same prostate cancer cell line (PC-3). These subpopulations display either highly proliferative, cancer stem cell (PC-3/M) or highly invasive, epithelial-mesenchymal-transition-like phenotypes (PC-3/S). Our model-driven analysis and experimental validations have unveiled a differential utilization of the long-chain fatty acids pool in both subpopulations. More specifically, our findings show an enhanced entry of long-chain fatty acids into the mitochondria in PC-3/M cells, while in PC-3/S cells, long-chain fatty acids are used as precursors of eicosanoid metabolism. The different utilization of long-chain fatty acids between subpopulations endows PC-3/M cells with a highly proliferative phenotype while enhances PC-3/S invasive phenotype. The present work provides a tool to unveil key metabolic nodes associated with tumor heterogeneity and highlights potential subpopulation-specific targets with important therapeutic implications.
| Prostate cancer (PC) is the most commonly diagnosed non-cutaneous malignancy among Western men and accounts for the second leading cause of cancer-related death [1]. In the majority of cases, PC eventually becomes independent of androgens, resuming growth after androgen-deprivation therapies in a more aggressive and therapy-refractory form [2].
The coexistence within the same tumor of a variety of cell subpopulations, featuring different phenotypes (intra-tumoral heterogeneity) associated with tumor evolution and progression reflects extreme plasticity and adaptation capability of neoplastic cells. This diversity is reached through genetic evolution of neoplastic cells and epigenetic and metabolic reprogramming of neoplastic and non-neoplastic tumor components that enhance tumor progression and represent a challenge for targeted therapies [3,4].
A major driver of intra-tumor heterogeneity is Epithelial-Mesenchymal transition (EMT), which induces alterations in the intricate and large cancer cell gene regulatory and metabolic networks (metabolic reprogramming) [5]. However, although EMT-mediated molecular and cellular changes have been widely studied, the EMT-induced metabolic changes remain poorly understood. In this sense, it is widely accepted that metabolic reprogramming is one of the ten hallmarks of cancer [6] which endows cancer cells with a phenotype characterized by a rapid and continuous proliferation, metastasis, invasion, and treatment resistance. Thus, study of the metabolism in these heterogeneous cellular populations is of special interest and must be approached from a global perspective integrating global metabolism with consideration of different subpopulations.
In this context, integration of omics data from high-throughput technologies, such as transcriptomics, into a genome-scale metabolic network reconstruction analysis, has been successfully used to study the metabolic mechanisms underlying different cancer types [7,8]. However, the differences in metabolic physiology between intra-tumoral subpopulations have not yet been taken into account in these computational approaches.
Here, we have built comparative genome-scale metabolic network models based on transcriptomic data for two clonal sub-populations isolated and separated from an established prostate cancer cell line (PC-3): i) a Cancer Stem Cell subpopulation -CSC- with high metastatic potential, low invasiveness and a higher proliferation rate (PC-3/M cells) and ii) a non-CSC subpopulation expressing EMT markers with high invasiveness and low metastatic potential (PC-3/S cells) [9]. These neoplastic cell sub-populations, capturing extreme epithelial vs. mesenchymal phenotypes, were derived from the same tumor cell line and represent an excellent cellular model to study how intra-tumoral heterogeneity and the different phenotypes endowed by the different subpopulations provides advantages to the tumor in terms of metastatic capability and drug resistance.
Our computational analysis has unveiled several subpopulation-specific metabolic alterations associated with long-chain fatty acids (LCFA) metabolism. First, we have identified an increased transport activity of LCFA into the mitochondria via Carnitine palmitoyl transferase I (CPT1), suggesting an increased β-oxidation, which could enhance proliferation in the PC-3/M subpopulation. Second, PC-3/S cells are predicted to have enhanced conversion of LCFA to arachidonic acid (AA), the precursor of a variety of eicosanoids that enhance angiogenesis, cell adhesion and invasion. Finally, the lower CPT1 activity predicted in PC-3/S cells leads to Docosahexaenoic acid (DHA) accumulation, a LCFA with antiproliferative effects [10]. The latter prediction is consistent with the lower proliferation rate observed in PC-3/S cells.
Next, using targeted metabolomics measurements, we experimentally confirmed these predictions and demonstrated that: i) the low proliferative rate of PC-3/S cells compared with PC-3/M subpopulation is consistent with higher intracellular concentrations of DHA ii) PC-3/M cells presents a higher CPT1 and β-oxidation activity that can be associated with their observed higher proliferative rate and iii) in PC-3/S, the reported cell adhesion and enhanced invasive capability can be explained by higher levels of AA and eicosanoids, PGE2 and 12S-HETE. Finally, we experimentally showed that the low efficacy of etomoxir (a CPT1 inhibitor) in metastatic PC tumors is conferred by the low sensitivity of non-metastatic subpopulation (PC-3/S) towards this drug in contrast with the high sensitivity showed by metastatic subpopulation (PC-3/M) and can be explained by altered LCFA transport activity facilitated by CPT1. The approach presented hereby provides a tool to unveil key metabolic nodes and vulnerabilities specific to distinct cancer cell subpopulations and opens new avenues in the development of more specific and efficient anti-tumoral therapies.
To infer the activity states of the metabolic networks of PC-3/S and PC-3/M subpopulations, we used previously generated transcriptomic data for PC-3/S and PC-3/M cells based on microarray technology [9] which was integrated into a genome-scale reconstruction of the human metabolic network [11]. In brief, this integrative method defines an upper threshold above which the genes are considered highly expressed and a lower threshold below which the genes are considered lowly expressed and seeks a network activity state in which the number of active reactions associated with highly expressed genes and the number of inactive reactions associated with lowly expressed genes are maximized [12,13]. In other words, this approach defines an objective function intended to maximize the similarity between gene expression and the activity state of the metabolic network rather than predefine an objective function that may not properly describe the cellular phenotype (i.e. biomass maximization). Next, we identified a set of reactions whose activity state was unambiguously different between subpopulations using sensitivity analysis (see Methods and Supplementary information S1 File and S1 Text).
Finally, our computational analysis permitted to infer, for each subpopulation, a set of active reactions, either intracellular or nutrients uptake/secretion between cell and media (see Fig 1 and Supplementary information S1 Text). The predicted metabolic uptake/secretion rates were in accordance with experimental measurements (true positive rate of 70% with an associated p-value < 0.001; [14]). For instance, our model-driven analysis successfully predicted the consumption / production patterns of most of the bioactive amino-acids, as well as glucose consumption and lactate production in both subpopulations (Supplementary information S1 Text).
This model-driven analysis revealed two major differences between the two subpopulations. First, the activity of fatty acid oxidation predicted by the model is higher in PC-3/M than in PC-3/S cells. The oxidation of fatty acids in the mitochondria produces NADH, FADH2 and acetyl-CoA that fuels the production of energy via TCA cycle and the electron transport chain. Most of the reactions differentially activated in this pathway involve carnitine palmitoyl transferase 1 (CPT1). This mitochondrial membrane protein actively transports LCFA from cytosol into the mitochondria [15]. Our analysis provided a set of eight cytosolic LCFAs that were predicted to be substrates of CPT1 exclusively in PC-3/M cells. Interestingly, it has been reported that five of these eight LCFAs, including DHA, have antiproliferative effects [15–20]. Second, the analysis predicted an increased activity of eicosanoid metabolism in PC-3/S cells that is associated with angiogenesis, cell invasion and adhesion [21–23]. Arachidoic acid metabolism (AA) is the main precursor of this pathway that in turn is fueled by a fraction of the LCFA previously mentioned. Finally, our analysis revealed other significant differences between the metabolic activities of the two PC-3 subpopulations that are consistent with published evidences. For example, Vitamin D3 metabolism was predicted to be more active in the PC-3/S subpopulation. This molecule controls proliferation in prostate cells [24] and has antiproliferative effects on a number of cancer cell lines, being PC-3 cells (parental cell line) one of the few cell lines insensitive to this drug [25]. Taken together, the in silico analysis suggests the occurrence of metabolic alterations that correlate with a more proliferative phenotype in PC-3/M cells and with a less proliferative and more invasive phenotype in PC-3/S cells. These predictions are consistent with the reported phenotypes of both PC3 subpopulations [9].
Our computational analysis predicts both a higher LCFA entry into the mitochondria via CPT1 and a more active LCFA β-oxidation in PC-3/M cells. LCFA must be imported into mitochondria to be degraded via β-oxidation and CPT1, a mitochondrial membrane enzyme that plays a critical role in its transport into the mitochondrial matrix [15,26]. To experimentally verify that CPT1 protein levels differ between PC-3/M and PC-3/S cells, we compared CPT1 levels in PC-3/M and PC-3/S subpopulations by western blotting (Fig 2C). In line with the computational inference, we found that PC-3M cells express 30% higher levels of CPT1 than PC-3/S cells. As CPT1 mRNA levels are not significantly different between the two subpopulations (7.68±0.22 A.U in PC-3/M cells and 7.13±0.033 A.U. in PC-3/S cells and p-value > 0.01 using T-test), this highlights the power of the data integration approach to infer metabolic alterations even if mRNA and protein-level changes do no match owing to post-transcriptional regulation [13].
To experimentally determine whether β-oxidation activity is higher in PC-3/M, we measured acylcarnitine levels. These molecules are CPT1 activity intermediates in the entry of acyl‐CoAs into the mitochondrial matrix and are used to experimentally infer the β-oxidation activity [27]. Here we found that acylcarnitine levels were significantly higher in PC-3/M compared with PC-3/S cells (Fig 2E and Supplementary information S6 File), which is in accordance with our computational predictions and supports the hypothesis that PC-3/M cells present a more active β-oxidation. Etomoxir is a CPT1 inhibitor that consequently inhibits β-oxidation and its associated oxygen consumption [15, 26]. Since in many cancer types, tumor onset and progression relies more on lipid fuel than on aerobic glycolysis, this compound is widely used in cancer research [15] but not in clinical practice due to its hepatotoxicity. However, as stated above, its antiproliferative efficacy on PC-3 cells is the lowest among the different PC cell lines studied [15]. Based on these evidences and the results of our analysis, we hypothesized that this could be explained by a low activity of CPT1 in the PC-3/S subpopulation.
To experimentally test this hypothesis, we measured the oxygen consumption rate (OCR) before and after exposure to etomoxir of both subpopulations (Fig 2B, see Methods). We found that PC3/M cells show a 30% higher sensitivity to CPT1 inhibition than PC-3/S cells, implying that CPT1 and hence β-oxidation is more rate-limiting in the PC-3/M subpopulation.
In addition, we studied the dose-effect relation between etomoxir and the proliferation of PC-3/M, PC-3/S and parental PC-3 cells. This experiment also showed a higher sensitivity of PC-3/M cells towards the antiproliferative effects of etomoxir and the difference in the proliferative rate between subpopulations increased with higher concentrations of etomoxir (Fig 2F and Supplementary information S5 File). These experimental observations are in accordance with our computational predictions. Furthermore, PC-3 cells that represent a heterogeneous population containing cells with “PC-3/M-like” and “PC-3/S-like” phenotypes showed an intermediate sensitivity which supports our hypothesis that the low sensitivity of PC-3 cells to etomoxir may be conferred by tumor cell subpopulations with PC-3/S-like metabolic features.
We also hypothesized that, as a side effect of the relatively low CPT1 activity in PC-3/S cells, the levels of some LCFAs would be higher in this subpopulation compared to PC-3/M cells. More specifically, we focused on determining the levels of docosahexaenoic acid (DHA), a compound with antiproliferative effects in cancer [28–31]. To this aim, we used a targeted approach based on the Biocrates platform Assay [32] to quantitatively measure the concentration of DHA in both PC-3 subpopulations. We found that the concentration of DHA was 262% higher in PC-3/S cells compared to PC-3/M cells (Fig 2D, see Methods), which supports our model-driven predictions and is consistent with the lower proliferation rate reported in the PC-3/S subpopulation.
A further key prediction of our model-driven analysis is that the eicosanoid metabolism is more active in PC-3/S cells. Arachidonic acid (AA) is the precursor of this pathway, which in turn is metabolized from some of the eight LCFAs that are predicted to be transported by CPT1 into the mitochondria exclusively in PC-3/M cells. Based on these results, we hypothesized that the levels of AA and other eicosanoids were higher in PC-3/S than in PC-3/M cells. To validate this hypothesis, we applied a targeted approach using the Biocrates platform Assay [32], which allows quantitative measurements of arachidonic acid, eicosanoids and other oxidation products of polyunsaturated Fatty Acids (PUFAs). Among all the metabolites measured by this platform (Supplementary information S4 File), AA, Prostaglandin E2 (PGE2) and 12(S)-hydroxy-5Z,8Z,10E,14Z-eicosatetraenoic acid (12S-HETE) were significantly higher in PC-3/S than in PC3-M cells (Fig 3). 12-HETE is the dominant AA metabolite in PC3 cells and its levels in human prostate cancer tissues exceed by > 9-fold its levels in normal human prostate tissue [34]. Furthermore, in PC3 cells, 12(S)-HETE increases the expression of ITGA3V gene, which is associated with cell adhesion [20] and promotes PGE2 metabolism in cultured PC3 cells [35]. High PGE2 levels are associated with cancer [36–38] and affects different mechanisms that have been shown to play a role in carcinogenesis such as cell invasion via the Akt signaling pathway [22] or angiogenesis by over-expressing the VEGF gene [23,39]. Consistent with these regulatory effects of AA metabolites, the expression level of ITGA3V, VEGF and AKT genes were significantly higher in PC-3/S than in PC-3/M cells (Log2 FC of 1.24 ± 0.27, 0.79 ± 0.25, 1.46 ± 0.4 respectively). Taken together, these results indicate a higher activity of eicosanoid metabolism in PC-3/S cells leading to higher levels of eicosanoid pathway intermediates and the upregulation of genes associated with angiogenesis, cell adhesion and invasion, all of which are consistent with the phenotype observed in this subpopulation.
Intratumoral heterogeneity is key to understanding the hierarchical and functional relationships between different neoplastic cell populations within a given tumor, with direct implications on tumor dynamics and progression [40]. Here we have focused on the study of the metabolic profiles of two clonal cell sub-populations isolated from an established prostate cancer cell line (PC-3): PC-3/M and PC-3/S cells [9]. These sub-populations were derived from the same PC cell line and thus they can be assumed to coexist within the same tumor, representing an excellent model to study how intra-tumoral heterogeneity benefits the tumor in terms of invasiveness, metastatic capability and drug resistance. This fact also enables us to investigate the relationships between gene expression and metabolism with tumor-initiating cell or mesenchymal-like attributes in neoplastic cells.
Here, we inferred the metabolic activity states of these PC-3 subpopulations by integrating transcriptomic data with a genome-scale metabolic network reconstruction. The applied constraint-based method treats gene expression levels merely as cues for the likelihood that their associated reactions carry metabolic fluxes and hence allowing for potential post-transcriptional regulatory effects. For example, a reaction associated with a highly expressed gene does not necessarily entail a high flux. As a consequence, the method allows us to infer metabolic activity patterns that go beyond conventional gene expression analysis. Indeed, some of the inferred metabolic differences between PC-3/M and PC-3/S cells do not correlate with the expression patterns of the underlying genes. One example is provided by the expression levels of genes associated with the fatty-acyl-ACP hydrolase reaction that participates in the oxidation of fatty acids. The transcriptomic profiles suggest that these genes are more active in PC-3/S cells, which is in contrast to our computational analysis identifying the corresponding reaction to be active only in the PC-3/M subpopulation (See Supplementary information S3 File). Importantly, we have provided experimental support for this prediction by observing a significantly higher sensitivity of PC-3/M cells to CPT1 and β-oxidation inhibition by etomoxir compared to PC-3/S cells, thereby demonstrating the importance of considering the network context when inferring metabolic changes from transcriptomic data. Overall, our approach has revealed two major metabolic differences at the level of LCFA utilization with relevance for tumor proliferation, invasiveness and metastasis in our dual cell model.
First, it has unveiled an increased CPT1 activity in PC-3/M cells. CPT1 has numerous cytosolic substrates, including cervonyl coenzyme A (DHA precursor), eicosatetranoyl coenzyme A, arachidyl coenzyme A (arachidonic acid precursor), trans-2-octadecenoyl-CoA(4-), palmitate, Malonyl-CoA, linoelaidyl coenzyme A (linoleic acid precursor) and vaccenyl coenzyme A. Interestingly, it has been reported that five of these eight LCFAs have antiproliferative effects [16–20]. Thus, the higher CPT1 and β-oxidation activities in PC-3/M cells may have two roles: i) first, and probably the most evident, is to maintain the energetic requirements imposed by the high proliferation rates of PC-3/M cells and ii) to eliminate LCFAs with antiproliferative effects. The predicted differences in CPT1 activity were supported experimentally by the finding of 33% higher CPT1 protein levels in PC-3/M than PC-3/S cells. Further, we demonstrated that β-oxidation activity was more sensitive to the inhibition of CPT1with etomoxir in PC-3/M than in PC-3/S cells (see Fig 2B) which highlights the importance of this enzyme in the energy metabolism of this subpopulation. This is of special interest since fatty acid oxidation plays a key role as source of NADH, NADPH, ATP and FADH2, all providing survival advantage to cancer cells [41]. Finally, we showed that the concentration of DHA is significantly higher in PC-3/S than in PC-3/M cells (Fig 2D). Several studies have reported anti-proliferative effects of DHA in tumors, consistent with the high proliferative rate observed in PC-3/M cells and support the hypothesis that an increased activity of CPT1 is also necessary to eliminate anti-proliferative molecules in PC-3/M cells. Taken together, our findings supports the key role played by CPT1 to sustain the high proliferation rate of PC-3/M cells by degrading LCFAs through energy metabolism while avoiding their antiproliferative effects.
Secondly, our analysis predicted a higher activity of the eicosanoid metabolism in PC-3/S cells. Most of the LCFAs described in this study are AA precursors that in turn fuel this pathway. Eicosanoid metabolism produces a variety of molecules with reported tumorogenic activity in prostate cancer [42]. Importantly, these processes are predicted to occur in the lysosome, in which is reported that may produce pro-oncogenic alterations [43]. Here we validated this prediction by using metabolomic measurements which revealed higher levels of AA in PC-3/S (see methods and Fig 3). In addition, in line with the computational predictions, the concentrations of several products of eicosanoid metabolism, such as 12S-HETEand PGE2, were significantly higher in the PC-3/S subpopulation. Prior evidence suggests that these compounds are associated with the upregulation of ITGA3V, VEGF and AKT [21–23, 39], which promote cell adhesion, angiogenesis and cell invasion. Importantly, we have found that these genes are indeed upregulated in PC-3/S cells. Thus, our findings suggest that higher levels of AA, 12S-HETE and PGE2, associated with a more active eicosanoid metabolism in PC-3/S cells, contribute to increased angiogenesis, cell adhesion and invasion potentials of these cells.
Overall, our findings support the view that the relatively high activity of CPT1 in PC-3/M cells increases the entry of LCFAs into the mitochondria to be oxidized and to produce energy to sustain a high proliferation rate (Fig 4). This process decreases the levels of LCFAs such as DHA, thus preventing their antiproliferative effects. In contrast, the lower CPT1 activity in PC-3/S cells would lead to an accumulation of anti-proliferative LCFAs in the cytosol, thereby reducing the growth rate in PC-3/S cells and increasing the availability of substrates for eicosanoid synthesis (Fig 4A). Summing up, we propose that the metabolic reprogramming involving LCFA utilization enhances the metastatic potential and proliferation in PC-3/M cells while in PC-3/S subpopulation increases cell adhesion, invasion and angiogenic capability and promotes DHA accumulation which reduces its proliferation.
The model-driven analysis employed here has provided additional insights into metabolic changes linked to cancer phenotypes. For example, our analysis further predicted acid ceramidase (ASAH1) to be active predominantly in PC-3/M cells, a prediction consistent with experimental evidences showing a higher ASAH1 enzymatic activity in this subpopulation [44]. Our analysis also predicts that calcitriol metabolism is mainly active in PC-3/S cells. This molecule has antiproliferative activity in a variety of human cancer cells [25] which is consistent with the low proliferative rate of PC-3/S cells compared with PC-3/M cells [9]. Our model prediction also suggests that the reported low sensitivity of PC-3 cells towards Vitamin D3 [24] could be conferred by the low Vitamin D metabolism activity in the PC-3/M subpopulation. In addition, this prediction is consistent with the lower proliferative rate observed in PC-3/S cells, which would be more sensitive to Vitamin D3 anti-proliferative effects.
Our analysis of a dual-cell model representing distinct and opposing neoplastic phenotypes allows us to propose subpopulation-specific and complementary therapeutic interventions. The results of the experiment determining the dose-effect relation between etomoxir concentration and cell proliferation showed that PC-3/M cells are more sensitive to etomoxir than PC-3/S cells, and that the parental PC-3 cell line presents an intermediate sensitivity. Thus, the poor performance of etomoxir at inhibiting the growth of PC-3 cells compared to other prostate cell lines may be explained by the low metabolic dependence of the PC-3/S subpopulation on CPT1. In other words, androgen-independent prostate cancer cells with CSC attributes similar to PC-3/M cells would be sensitive to etomoxir, while this drug would be less efficient in tumor cell subpopulations with EMT attributes similar to PC-3/S cells with a phenotype characterized by high cell invasion and adhesion and angiogenic capability.
Finally, it has been reported that the cox-2 reaction, which produces PGE2 and is over-expressed in prostate cancer [45], is activated by 12S-HETE [46] which is in turn metabolized by the 12-LOX reaction. Interestingly, a number of drugs such as cinnamyl-3,4-dihydroxy-alpha-cyanocinnamate (CDC) or baicalein, that inhibit 12-LOX activity, have been shown to present strong anti-tumoral effects in prostate cancer [47].
Our study represents a novel approach to discern metabolic vulnerabilities associated with heterogeneous tumor cell populations. However, future studies measuring the effects of single and combinatorial drug treatments affecting subpopulation-specific targets on heterogeneous co-culture of non-CSC (PC-3/S) and CSC (PC-3/M) subpopulations are needed to determine the significance of these findings. For instance, the combinatorial effect of CDC or baicalein with drugs such as oxfenicine or perhexiline (CPT1 inhibitors without the hepatotoxicity of etomoxir–[48]) could be tested as potential anti-tumoral drug treatments targeting the key metabolic processes preferentially active in PC-3/S or PC-3/M cells, respectively. Additionally, as gene networks associated with progression and metastasis in our PC-3 dual model is significantly correlated with those in other tumor types [14], the metabolic reprogramming proposed here could be extrapolated to different cancer types. Our findings will facilitate a better understanding of the EMT-induced metabolic changes and their role in tumor heterogeneity and opens new avenues for the development of new subpopulation-specific anti-cancer therapies.
Transcriptomic data: Gene expression levels of each cell subpopulation (three replicates per subpopulation GSE24868, [9]) by microarray analysis (Affymetrix genechip u133a 2.0) and normalized by RMA [49]. Transcriptomic data was integrated into a genome-scale metabolic network reconstruction analysis to infer the activity state of the metabolic reactions in both subpopulations.
Consumption and production of metabolites: Additionally, we used the measured consumption and production of some metabolites [14] to assess the reliability of model predictions (Supplementary information S1 Text). These metabolites were: glucose, lactate, pyruvate, glutamate and aminoacids.
To obtain accurate cell-specific genome-scale metabolic models of the PC-3 subpopulations, we performed a subpopulation-specific genome-scale network reconstruction analysis by integrating the transcriptomic data into the most recent reconstruction of human metabolism (Recon2) [11]. Recon2 is a genome-scale stoichiometric model that represents the entire network of human metabolic reactions. This generic genome-scale metabolic model provides the appropriate transcript-protein-reaction associations that permit the integration of the previously mentioned transcriptomic data for which we used a widely tested constraint-based method [12].
In order to reduce the computational time necessary to perform the analysis, the metabolic model (Recon2) [11] was reduced. The reduction was done by removing the blocked reactions from the model. These reactions are those incapable of carrying any metabolic flux in steady state [50]. To this aim we first performed a Flux Variability Analysis (FVA) [51–53] using Fasimu software [54]. This analysis computes minimal and maximal flux in each reaction. Each analysis evaluates the feasibility of the simulation. The reactions in which their maximization and minimization simulations were not obtained any feasible solution were considered as blocked reactions. In order to ensure that the reduced model was able to consume/produce the experimentally measured extracellular metabolites, we forced the corresponding exchange reactions to be always active. It was achieved by splitting all the exchange reactions in a forward and a backward reaction and the lower/upper bounds of the reactions associated to the experimentally measured metabolites were fixed at 0.001/1000 in the forward reactions and at -1000/-0.001 in the backward reactions. Once determined, the blocked reactions were removed from the model, as well as those metabolites that were neither products nor substrates of any reaction.
We integrated the transcriptomic data into Recon2 [11] by using the gene-protein-reaction (GPR) associations included in the model. These associations are “and/or” logical sentences that establish a relation between the metabolic reactions and the genes encoding the enzymes that catalyze them. GPR associations include information related with isoenzymes (using the logical “or”), complexes (using the logical “and”) or direct gene-reaction relations (i.e. the activity of Reaction1 depends on: “(geneA and geneB) or (geneC and geneD)”). To integrate the gene expression data from PC-3/M and PC-3/S subpopulations into Recon2 and determine the gene expression level associated to the metabolic reactions in each subpopulation, we substituted the logical “and” and “or” by “minimum” and “maximum”. Thus, for example, if the activity of a given reaction depends on the expression of different genes and it is defined by the following logical expression “(geneA and geneB) or (geneC and geneD)”, and the expression of the gene A, B, C and D are 0.5, 3, 1 and 0.1 respectively. Then, by integrating the gene expression levels into the logical sentence and replacing the logical operators by “minimum” and “maximum” we obtained the following expression: “max(min(0.5,3),min(1,0.1))”. Thus, based on the transcriptomic data and the GPR association, the gene expression associated with the reaction is 0.5. Finally, we obtained a numerical value for each reaction indicating the level of expression of their corresponding associated genes.
We used gene expression levels associated with the metabolic reactions to infer the activity states of reactions in the network by using a recently developed constraint-based method [12]. This method solves a mixed integer linear programming (MILP) problem to obtain a flux distribution in which the number of reactions associated with highly expressed genes is maximized (RH), and the number of reactions associated with lowly expressed genes is minimized (RL) while satisfying the thermodynamic and stoichiometric constrains imposed by the model:
maxv,y+,y−=(∑i∈RH(yi++yi−)+∑i∈RLyi+)
S∈v=0
(1)
The mass balance constraint: where v is the flux vector and S is a n x m stoichiometric matrix, in which n is the number of metabolites and m is the number of reactions.
Thermodynamic constraints, that restrict flow direction, are imposed by setting vmin and vmax as lower and upper bounds respectively.
The Boolean variables y+ and y–. In RH reactions represent whether the reaction is active or not respectively. In RL y+ represents the reaction is not active.
A highly expressed reaction is considered to be active if it carries a significant positive flux that is greater than a positive threshold Ɛ. In our study Ɛ = 1. Consequently the ith reaction is active if: vi ≥ 1
vi+yi−(vmin,i+ε)≤vmax,i,i∈RH
or has a significant negative flux <–Ɛ (as our model didn't consider reversible reactions it cannot occur)
vmin(1−yi+)≤vi≤vmax,i(1−yi+),i∈RL
(5)
Lowly expressed reactions are considered to be inactive if they carry zero metabolic flux, though changing Eq (5) to enable these reactions to carry a low metabolite flux (that is, with an upper bound lower than Ɛ) and still be considered inactive provides qualitatively similar results. The Fig 5 illustrates the process.
This method defines an upper threshold above which the expression of a given gene is considered high and another threshold below which gene expression is considered low. In our study, the chosen upper and lower thresholds were those symmetric percentiles that maximize the cases where the number of reactions associated with highly expressed genes in one subpopulation were associated with lowly expressed gene in the other subpopulation and vice versa. Thus, we defined the upper threshold at the 66th percentile and the lower threshold at the 33th percentile. The method also uses the parameter that represents the flux above which a given reaction is considered to carry a significant metabolic flux. As is defined in [12] we gave to Ɛ a value of 1. Once the thresholds were fixed, we performed the expression-based activity prediction analysis with Fasimu software by applying “compute-FBA–xs” option (See Fasimu tutorial [54]).
In the Expression-based activity prediction analysis we found an optimal solution in terms of the objective function maximization, although this solution may not be unique. A space may exist of alternative optimal solutions that represent alternative steady-state flux distributions yielding the same similarity with the gene expression data (the same objective function value).
To account for these alternative solutions, we employed Sensitivity analysis [12].
This is performed by solving two MILP problems (as is described in Expression-based activity prediction) for each reaction to find the maximal attainable similarity with the expression data when the reaction is: (i) forced to be activated and (ii) forced to be inactivated.
Thus, a reaction is considered to be active if a higher similarity with the expression data is achieved when the reaction is forced active than when it is inactive (the objective function is higher when the reaction is active). Conversely, it is considered to be inactive if the similarity is higher when the reaction is forced to be inactive. If the similarities with the experimental data are equal in both cases the activity state of the reaction is considered to be undetermined. From this analysis we could infer which pathways are more active in each subpopulation.
By analyzing the predicted activity state of the exchange reaction we can infer which metabolites are consumed and/or produced. In order to determine the goodness of our model predictions we compared qualitatively the consumption and production of some experimentally measured metabolites [14] with the corresponding model predictions (Supplementary information S1 Text). This comparison was done by constructing a 2x2 contingency matrix and the levels of significance were determined using Fisher exact test (Supplementary information S1 Text).
The algorithm used to integrate the information from gene expression levels into a Genome-scale metabolic network reconstruction defines a threshold above which gene expression levels are considered high and a second threshold below which they are considered low. It calls for the performance of a robustness analysis in order to demonstrate the lack of dependency of our predictions on the thresholds used in the analysis. In order to determine the robustness of our prediction, we performed the analysis previously described in sensitivity analysis defining different sets of thresholds:
Thereby, we defined a set of reactions that were predicted to be active, inactive or undetermined (the method cannot predict their activity state) independently of the thresholds.
Cells were seeded in XF24-well microplates (Seahorse Bioscience) at 4.5·104 cells/well and 9.0·104 cells/well, respectively, in 100 μL of growth medium, adding 100 μL more after 3–5 h, and then incubated at 37°C with 5% CO2 overnight. After overnight incubation and 1 h before the assay, growth media was replaced by basal media (unbuffered DMEM; Sigma-Aldrich) with 3 mM glucose and 5 mM carnitine. The sensor cartridge was loaded with etomoxir and calibrated prior to the start of the experiment. Determinations were performed on a XF24 Extracellular Flux Analyzer (Seahorse Bioscience). Responses to etomoxir (Signma-Aldrich) treatment (final concentration 30 μM) were expressed as LOG2 to indicate the fold change comparing the measured point immediately after and before the corresponding injection.
To determine eicosanoids and oxidation products of polyunsaturate fatty acids levels in PC-3/M and PC-3/S cells we used Biocrates triple quadrupole MS-based platforms [32]. This platform enables the systematic quantification of relevant biological metabolites. The method is a quantitative screen of selected metabolites detected with multiple reaction monitoring, neutral loss and precursor ion scans. Metabolites are then quantified by comparison to structurally similar molecules labeled with stable isotopes added to the samples in defined concentrations as internal standards. The process is controlled by MetIDQ Software which controls sample management, data collection, data validation, and analysis.
Cell extracts were obtained from frozen cell pellets using RIPA buffer (50 mM Tris,pH 8.0, 150 mM NaCl, 0.1% SDS, 1% Triton X-100 and 0.5% sodium deoxycholate) supplemented with protease inhibitor cocktail (Sigma-Aldrich). Protein concentrations from the supernatant were determined by the BCA assay. Thirty-five mg of protein per sample were loaded and separated by 10% SDS-PAGE and transferred to PVDF membranes. Membranes were blocked by incubation with PBS-Tween (0.1% (v/v)) containing 5% non-fat dried milk for 1 hour at room temperature. Then, membranes were incubated with CPT1 primary antibody (Sigma-Aldrich, SAB1410234, 1/200), rinsed with PBS-Tween (0.1% (v/v)) and finally incubated with the secondary antibody anti-rabbit (Amersham Biosciences, NA934V, 1/3000) for 1 hour at room temperature. Blots were treated with the Immobilon ECL Western Blotting Detection Kit Reagent (Millipore) and developed after exposure to Fujifilm X-ray film.
For sample acquisition and processing, 5 106 cells of PC-3M and PC-3S cell lines were tripsinized and washed twice with ice-cold PBS prior to snap-freezing in liquid nitrogen. Cell pellets were stored at -80°C until measure. Right before measuring, cell pellets were thawed at room temperature and resuspended in 70 μL of 85:15 EtOH:PBS solution. Cells were disrupted by two sonication/freezing/defreezing cycles using titanium probe (VibraCell, Sonics & Materials Inc., Tune: 50, Output: 25), liquid N2 and a 95°C heat block. Cell lysates were then centrifuged at 20,000 rcf for 5 minutes at 4°C. Supernatants were collected into new tubes and total protein content was determined by Bichinconinic acid (BCA) assay (Thermo Fisher Scientific, Waltham, MA USA).
Then, standards, internal standards, quality controls (10 μL of each), and samples (30 μL) were loaded into the Biocrates AbsoluteIDQ® p180 plates (BIOCRATES Life Sciences AG, Innsbruck, Austria), processed according to manufacturer instructions and measured by FIA-MS/MS using a SCIEX 4000 QTRAP mass spectrometer.
Concentrations for metabolites were determined using the MetIDQ™ software package, which is an integral part of the AbsoluteIDQ® kit. The obtained metabolite concentrations were corrected considering the loaded volume of cell lysates and normalized by protein content.
|
10.1371/journal.pbio.1002288 | Phosphatidylthreonine and Lipid-Mediated Control of Parasite Virulence | The major membrane phospholipid classes, described thus far, include phosphatidylcholine (PtdCho), phosphatidylethanolamine (PtdEtn), phosphatidylserine (PtdSer), and phosphatidylinositol (PtdIns). Here, we demonstrate the natural occurrence and genetic origin of an exclusive and rather abundant lipid, phosphatidylthreonine (PtdThr), in a common eukaryotic model parasite, Toxoplasma gondii. The parasite expresses a novel enzyme PtdThr synthase (TgPTS) to produce this lipid in its endoplasmic reticulum. Genetic disruption of TgPTS abrogates de novo synthesis of PtdThr and impairs the lytic cycle and virulence of T. gondii. The observed phenotype is caused by a reduced gliding motility, which blights the parasite egress and ensuing host cell invasion. Notably, the PTS mutant can prevent acute as well as yet-incurable chronic toxoplasmosis in a mouse model, which endorses its potential clinical utility as a metabolically attenuated vaccine. Together, the work also illustrates the functional speciation of two evolutionarily related membrane phospholipids, i.e., PtdThr and PtdSer.
| Lipids are essential constituents of biological membranes, and most organisms across the tree of life use a relatively limited repertoire of lipids in their membranes. This work reveals the natural and abundant presence of an exclusive lipid phosphatidylthreonine (PtdThr) in Toxoplasma gondii, a ubiquitous protozoan parasite of humans and animals. PtdThr is made by a novel parasite-specific enzyme, PtdThr synthase, which has evolved from the widespread enzyme phosphatidylserine synthase. The study shows that PtdThr is required for asexual reproduction and virulence of the parasite in vivo, and a metabolically attenuated mutant strain of Toxoplasma lacking PtdThr can protect vaccinated mice against acute and currently incurable chronic infection. This discovery demonstrates adaptive “speciation” of PtdThr from an otherwise near-universal membrane lipid phosphatidylserine and reveals de novo PtdThr synthesis in T. gondii as a potential drug target.
| Intracellular protozoan parasites impose a substantial threat to human and animal health. Toxoplasma gondii is one of the most prevalent protozoan parasites, infecting nearly all warm-blooded vertebrates, including humans [1]. Over the last two decades, T. gondii has also become a popular model organism to understand the biology of parasitic and free-living protozoans alike. The parasite causes debilitating opportunistic infections in immunocompromised individuals and neonates. The disease occurs by the multiplication and persistence of its acute and chronic stages, the latter of which is impervious to host immunity and existing drugs. Acute infection, hallmarked by tissue necrosis, is caused by successive rounds of lytic cycles, comprising host cell invasion, intracellular replication, and egression [1]. The entry and exit of T. gondii into and from host cells is dependent on calcium-regulated gliding motility and exocytosis of specialized secretory organelles [2,3].
Parasites proliferating within their host cells oblige a substantial biogenesis of organelle membranes, which are composed of mainly phospholipids and neutral lipids. The typical and natural phospholipids characterized so far include phosphatidylcholine (PtdCho), phosphatidylethanolamine (PtdEtn), phosphatidylserine (PtdSer), phosphatidylinositol (PtdIns), phosphatidylglycerol, and phosphatidate [4]. Others and we have shown that T. gondii contains common eukaryotic phospholipids as well as the pathways for autonomous synthesis [5–8]. Physiological functions of phospholipids in the parasite are poorly understood however, and most of the underlying enzymes have not been characterized as yet. Moreover, despite a steadily rising interest in roles of lipids in host–pathogen interactions [9], the existence and biogenesis of divergent pathogen-specific lipids remain very much underappreciated.
In our expedition to characterize membrane biogenesis in T. gondii, we fractionated the parasite lipids by high-performance liquid chromatography (HPLC) and observed a major lipid peak X1 eluting next to PtdSer (Fig 1A). Other major lipids were PtdCho, PtdEtn, PtdIns, PtdSer, and phosphoethanolamine-ceramide (PEtn-Cer), confirming previous reports [5,7]. To determine the precise identity of X1 fraction, we executed mass spectrometry (MS) analysis, which revealed certain PEtn-Cer and PtdSer species, as expected (Fig 1B). The most prominent peak in this fraction with an m/z of 850.5, however, did not correspond to a PEtn-Cer or PtdSer species. Tandem MS of the indicated peak showed a neutral loss of 101 atomic mass units (m/z, 749.6) contrary to the expected 87 for serine, or 141 for ethanolamine (Fig 1C). The m/z profile matched to threonine as the polar head group instead, which was also independently confirmed by HPLC analysis of amino acid derived from lipid hydrolysis (S1A Fig). The fatty acyl chains of this particular lipid, phosphatidylthreonine (PtdThr henceforth), were identified as 20:1 and 20:4. Other detectable, but evidently minor, PtdThr species also contained comparably polyunsaturated and long acyl chains (Fig 1B).
Next, we resolved the parasite lipids by two-dimensional thin layer chromatography (TLC). As apparent (S1B Fig), and also shown elsewhere [5], PtdCho, PtdEtn, PtdIns and PtdSer (besides PtdThr) were the major parasite lipids visualized by iodine-vapor staining. PtdThr (X1), detected again near PtdSer, was authenticated by MS analysis (S1C Fig). PtdThr accounted for ≈20 nmol/108 parasites by lipid phosphorus quantification. It is noteworthy that PtdThr has been previously reported as a rare and notably minor PtdSer analog in certain mammalian cells and selected prokaryotes [10–13]. It was also shown that the base-exchange-type PtdSer synthase activity located in the ER and its mitochondria-associated membranes in mammalian cells normally uses serine as its primary substrate [14,15]; but it can produce PtdThr as a by-product under serine-deprived condition [10]. In contrast, our results reveal a surprisingly abundant and natural occurrence of PtdThr in a widespread protist.
PtdThr species were absent in uninfected human fibroblasts used to culture parasites (S2 Fig), which implied their de novo synthesis in T. gondii. Our in silico and PCR (polymerase chain reaction) analyses aimed at establishing the genetic origin of PtdThr identified two putative base-exchange-type PtdSer synthases in the parasite database (www.ToxoDB.org; TGGT1_273540, TGGT1_261480) encoding for 614 and 540 residues, which we designated as TgPTS (PtdThr synthase) and TgPSS (PtdSer synthase), respectively, based on the results described in this work. Unlike PSS occurring across the phyla, orthologs of PTS could only be found in selected parasitic (Neospora, Eimeria, Phytophtora) and free-living (Perkinsus) chromalveolates (S3 Fig). Of note is the fact that distinct asparagine, histidine, and cysteine residues are conserved in all PSS orthologs, but not in TgPTS, which contains substitutions to glutamate, tryptophan, and serine at the equivalent positions (S4 Fig). Phylogeny supported the variability in the substrate-binding pocket of PSS [16] with that of PTS sequences and indicated a loss of latter enzyme in other related parasites.
Ectopic expression of epitope-tagged TgPTS-HA and TgPSS-HA showed a marked distribution in the endoplasmic reticulum (ER) of the parasite (Fig 2A). Because overexpression under the control of a foreign promoter may cause localization artifacts, we detected endogenous levels of PSS and PTS in transgenic parasite lines, in which the corresponding genes had been tagged with HA-epitope at the 3’-ends. As discussed below (S10B and S11 Figs), PSS fusion protein regulated by its promoter localized mainly in the parasite ER/mitochondrion intersecting with each other, and to some extent in acidocalcisomes/plant-like vacuole. The native expression of PTS was too low to be visualized (not shown). We nonetheless tested potential localization of PTS in other organelles using the parasites overexpressing TgPTS-HA; however, we found no apparent signal in micronemes, rhoptries, dense granules, mitochondrion, apicoplast, and acidocalcisomes/plant-like vacuole (S5 Fig). To evaluate the enzymatic function of both enzymes, we expressed them in Eschericia coli and assessed their catalytic activity in the presence of serine or threonine (Fig 2B). Lipid analyses of bacterial strains harboring empty vector (negative control), TgPTS, TgPSS, or Arabidopsis thaliana PSS (positive control [17]) showed synthesis of PtdSer by AtPSS and TgPSS as well as by TgPTS when using serine as substrate. Unlike AtPSS and TgPSS, however, TgPTS also produced PtdThr in presence of threonine, indicating that TgPSS is indeed a PtdSer synthase, whereas TgPTS can synthesize both PtdThr and PtdSer.
To endorse the function of TgPTS in T. gondii, we disrupted the gene in the parasite genome (Fig 3A). The Δtgpts strain was isolated by recombination-specific PCR screening, which confirmed an efficient disruption of the PTS gene locus (Fig 3B). Accordingly, the ORF-specific primers amplified a band of 4.2 kb in the Δtgpts strain in lieu of the expected 1.8 kb in the parental parasites (Fig 3C), which corroborated the targeted insertion of the selection marker and deletion of the predicted catalytic site (342ECWWD346; S4 Fig) [16]. Moreover, expression of adjacent transcripts flanking the PTS gene was unaffected, further confirming the specificity of transgenic manipulation (S6 Fig).
Synthesis of PtdThr was abrogated in the Δtgpts strain, as shown by TLC and lipid phosphorus assays (Fig 3D, S7 Fig). Concurrently, we observed a 3-fold gain in PtdSer level that was proportionate to the content of PtdThr in the parental strain (S7 Fig). This observed increase in PtdSer amount was due to an increased de novo synthesis of lipid, as shown by metabolic labeling with radioactive serine (S8 Fig). All these effects were entirely reversible as complementation of the mutant with a functional TgPTS recovered PtdThr (Fig 3D), as well as restored a normal PtdSer synthesis and lipid content (S7 and S8 Figs). Consistent with these results, the MS analyses revealed the absence of all PtdThr species and a parallel increase in PtdSer species in the mutant, which were reverted in the PTS-complemented strain (Fig 4). Taken together, these data show an autonomous synthesis of PtdThr in T. gondii and its abolition in the Δtgpts mutant. They also indicate a mutual regulation of PSS and PTS enzymes.
We next assessed the physiological impact of TgPTS ablation on the parasite growth by plaque assays. Compared to the parental strain, the Δtgpts strain formed noticeably smaller (−70%) and considerably fewer (−80%) plaques (Fig 5A and 5B). Ectopic expression of wild-type TgPTS largely rescued the parasite growth. In contrast, the catalytically-dead isoform of TgPTS(ΔECWWD), which was incapable of restoring PtdThr level in the Δtgpts strain (Fig 3D), could not amend the growth defect (S9 Fig), confirming the physiological need of the PTS activity for the parasite. It should be mentioned that the Δtgpts strain expressing TgPTS(ΔECWWD)-myc showed an accentuated growth defect when compared to the mutant (S9A Fig), which prevented its prolonged culture and detailed biochemical analyses.
In-depth phenotyping of the parental, Δtgpts mutant, and PTS-complemented parasite strains revealed a normal replication in the mutant (Fig 5C). Surprisingly, however, a complete lysis of host cells by the mutant was markedly delayed up to 96 hr, as opposed to 72 hr in host fibroblasts infected with the parental and complemented strains (dotted and solid arrows, Fig 5C). In accord, the mutant displayed a much slower natural egress than the two control strains (Fig 5D). For example, only about 27% and 46% of the mutant vacuoles were disrupted after 60 and 72 hr of infection as opposed to 67% and 94% of the parental vacuoles. Noticeably, the mutant was also impaired in invading fresh host cells (Fig 5E). Egression and invasion events require gliding motility in T. gondii, which drives the parasite’s exit from dilapidated host cells and ensuing infection of neighboring cells [2]. Indeed, the Δtgpts strain displayed an evidently reduced motility, as determined by a lower motile fraction and shorter trails compared to the two reference strains (Fig 6). These assays demonstrate the mandatory requirement of PTS activity for an effective functioning of the lytic cycle in T. gondii.
To examine whether an elevated level of PtdSer underlies the observed growth phenotype in the PTS mutant, we created a double mutant (Δtgpts/TgPSS-2HA-DD; S10A Fig). The TgPSS gene was fused with a Shield1-regulated degradation domain (DD) and 2HA epitopes at 3’-end [18] to achieve a conditional expression of PSS protein. The PSS-2HA-DD fusion protein showed a predominant fluorescent signal in the ER (S10B Fig). We also observed apparent staining of PSS with the markers of mitochondrion (F1B) [19] and acidocalcisomes/plant-like vacuole (vacuolar proton pyrophosphatase 1; VP1) [20], whereas other organelles, micronemes, rhoptries, dense granules, and apicoplast did not show evident PSS staining (S11 Fig). Expression of PSS-2HA-DD could be regulated by exposure to Shield1 (S10B and S10C Fig). Metabolic labeling of parasite lipids with serine (PtdSer and nascent decarboxylated product PtdEtn) also confirmed that PSS activity was restored to the parental level in the absence of Shield1 (Fig 7A, S12A Fig). A knockdown of PSS activity reinstated PtdSer content in the Δtgpts strain (S12B Fig). Even a normal PtdSer pool, however, was unable to rectify the growth defect in the Δtgpts/TgPSS-2HA-DD double mutant, which mirrored plaques formed by the Δtgpts strain (Fig 7B and 7C). These results exclude the impact of amplified PtdSer in disrupting the lytic cycle, while strengthening the physiological importance of PtdThr for T. gondii.
We also explored the prophylactic potential of the Δtgpts mutant. Examination of virulence in a mouse model demonstrated that nearly all animals infected with the Δtgpts mutant survived as opposed to the parental and PTS-complemented strains, both of which were explicitly lethal (Fig 8A). Importantly, all mice enduring the mutant infection became categorically resistant to a subsequent lethal challenge by a hypervirulent type I strain of T. gondii causing acute toxoplasmosis (Fig 8A). To further expand the therapeutic utility of our strain as a potential vaccine against chronic infection, we challenged the Δtgpts-infected animals with the cyst-forming type II strain. Remarkably, in contrast to naïve animals, the mutant-vaccinated mice showed no signs of chronic stage cysts in their brain tissue (Fig 8B and 8C). In accord, unlike the naïve control mice, we did not observe any inflammatory lesions in the cortex or meninges of the Δtgpts-immunized animals infected with the type II strain (Fig 8D). In brief, these results demonstrate a requirement of PtdThr for the parasite virulence and illustrate the prophylactic potential of a metabolically attenuated whole-cell “vaccine” against acute and chronic toxoplasmosis.
Our data reveal a natural and fairly abundant expression of PtdThr in a widespread pathogen. We also identified a novel enzyme realizing de novo synthesis of PtdThr in T. gondii. In addition, the work signifies functional speciation of two closely related lipids, i.e., PtdThr and PtdSer. Last but not least, we show a vital physiological role of PTS and PtdThr for the lytic cycle and virulence of T. gondii, which can be exploited to develop a vaccine against acute as well as chronic toxoplasmosis.
Besides being the building blocks of biological membranes, phospholipids are involved in many other cellular functions. For example, one of the several roles of PtdSer is to regulate calcium signaling and exocytosis that has been recognized for more than three decades in mammalian cells [21,22]. PtdSer controls Ca2+-triggered exocytosis by multiple mechanisms, which involve facilitating the binding of membrane-fusion protein machinery, altering the energy for membrane bending, as well as modulation of PLC-mediated IP3-dependent Ca2+ channels in the ER [23–25]. Further, anionic phospholipids, such as PtdSer, can also restrict Ca2+ slippage into the cytosol by sarcolemmal Ca2+-ATPase, which in turn increases the ion capture into the ER [26]. In T. gondii, calcium signaling is well-known to govern the consecutive events of motility, egression, and invasion by regulating exocytosis of specialized parasite organelles, notably micronemes [27,28]. PtdThr as one of the most abundant anionic lipids regulating Ca2+ homeostasis is therefore quite conceivable. Indeed, chemically-synthesized PtdThr derivatives are much more potent inducers of mast cell secretion than PtdSer, and the presence of defined acyl chains exerts a maximal exocytosis [29]—both of these findings are consistent with the natural and dominant existence of selected PtdThr species in T. gondii. It remains also possible that a lack of PtdThr induces adaptive changes in the parasite ER, which consequently impairs the lytic cycle.
The PTS mutant lacking PtdThr showed a balanced increment in PtdSer, which is reversed by genetic complementation. In line, we observed an apparent increase in the level of another major anionic lipid, PtdIns; however, only when PtdSer content was restored to normal in the double mutant deficient in PtdThr (Δtgpts/TgPSS-2HA-DD without Shield1), but not in the Δtgpts strain regardless of Shield1 in cultures (S12B Fig). Such a specific, reversible, and proportionate amplification of two other anionic lipids appears to maintain the net charge and membrane biogenesis but was entirely unable to mend the lytic cycle. It is therefore plausible that parasite has invented or selected PtdThr for realizing the lytic cycle, while satisfying the customary role of lipids in membrane biogenesis. In this context, it is worth stating that the parasite harbors a putative plant-like pathway to make threonine (www.ToxoDB.org), an amino acid otherwise essential for mammalian host cells. Our assays using stable 13C isotope of threonine demonstrated de novo synthesis of PtdThr in replicating T. gondii (S13 Fig). The isotope-labeled lipid accounted for only about 5% of the total PtdThr in the parasite, which implies a rather inefficient import of threonine by intracellular parasites and a dependence on autonomous synthesis to produce this exclusive lipid. A modest labeling of intracellular parasites with 13C-threonine resonates with a rather inefficient incorporation of radioactive precursor by extracellular parasites (not shown). Hence, it appears as though T. gondii has evolved a serine-threonine homeostasis that is quite distinct from its mammalian host.
Going forward, it will be important to define biochemical features of PtdThr-deprived and PtdSer-enriched mutant membranes. It will also be critical to characterize the biophysical properties of PtdThr species and perform high-resolution imaging of fluorescent analogs to determine its distribution in the parasite organelles. Likewise, knowing the exact sites of lipid synthesis using the antibodies against endogenous PSS and PTS proteins should help define the trafficking of PtdSer and PtdThr and their relative importance for calcium homeostasis in T. gondii. Most such studies, however, demand pure preparations of PtdThr species, fluorescent lipid derivatives and antibodies, which are not available at this point. Nonetheless, having established the genetic origin and functional relevance of PtdThr in a model pathogen provides a strong basis for future research on mechanism, evolution and therapeutic potential of PTS and PtdThr. Curative importance of a metabolically-attenuated strain has also been exemplified before using an uracil-auxotroph strain [30,31]. This work should therefore enable prospective vaccination studies using the attenuated PTS-disrupted strain, particularly against the yet-incurable and more prevalent chronic infections.
In summary, our research demonstrates the natural and abundant synthesis of an exclusive lipid class, PtdThr, in a widespread protozoan parasite, which is synthesized by a unique enzyme evolved from an otherwise universal protein. We also reveal a lipid-mediated regulation of parasite-specific functions, while illustrating an evolutionary paradigm, i.e., adaptive divergence of the related phospholipids. The physiological need of PTS for the parasite makes it an attractive therapeutic and vaccine target.
All in vivo assays were performed in compliance with the German animal protection laws directed by Landesverwaltungsamt Sachsen-Anhalt, Germany. The RHΔku80-hxgprt- strain of T. gondii and the pLIC-DHFR-2HA-DD vector were provided by Vern Carruthers (University of Michigan, United States) [32,33]. The plasmid pTgTUB8-TgDer1-GFP was obtained from Boris Striepen (University of Georgia, Athens, US). Primary antibodies for localization studies, mouse anti-TgMic2, mouse anti-TgRop1, mouse anti-TgGra5, mouse anti-TgF1B, rabbit anti-TgVP1 and rabbit anti-TgFd, were provided by David Sibley (Washington University, St. Louis, US), John Boothroyd (Stanford University School of Medicine, Stanford, US), Marie-France Cesbron-Delauw (CNRS-Université Joseph Fourier, Grenoble, France), Peter Bradley (University of California, Los Angeles, US), Silvia Moreno (University of Georgia, Athens, US), and Frank Seeber (Robert-Koch Institute, Berlin, Germany), respectively. Other primary antibodies used in this work, such as mouse anti-TgSag1, rabbit anti-TgGap45, and rabbit anti-TgHsp90 were offered by Jean-Francois Dubremetz (University of Montpellier, France), Dominique Soldati-Favre (University of Geneva, Switzerland), and Sergio Angel (IIB-INTECH, San Martin, Argentina), respectively. Oligonucleotides were purchased from Life Technologies. C57BL/6J mice were acquired from Janvier Labs (Saint Berthevin, France).
Tachyzoites of the RHΔku80-hxgprt- strain were propagated in human foreskin fibroblast (HFF) cells in a humidified incubator (37°C, 5% CO2). Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with fetal bovine serum (10%), glutamine (2 mM), MEM nonessential amino acids (100 μM each, glycine, alanine, asparagine, aspartic acid, glutamic acid, proline, serine), sodium pyruvate (1 mM), penicillin (100 U/ml), and streptomycin (100 μg/ml). Parasites were usually cultured at a multiplicity of infection (MOI) of 3 every 2–3 d unless stated otherwise. HFF were harvested by trypsinization and grown to confluence in fresh flasks, dishes, or plates as per experimental needs.
Parasite RNA was isolated using Trizol-based extraction method and subsequently reverse-transcribed into first-strand cDNA (Life Technologies). The complete ORFs of TgPSS and TgPTS were amplified from tachyzoite cDNA using PfuUltra II Fusion polymerase (Agilent Technologies, primers in S1 Table). TgPSS and TgPTS with a C-terminal HA-tag were cloned into the pTgSAG1-UPKO or pTgGRA1-UPKO plasmid at EcoRV/PacI or NsiI/PacI sites, respectively and transformed into E. coli XL-1b (Stratagene) for cloning and vector amplification. The plasmid constructs were transfected into fresh tachyzoites of the RHΔku80-hxgprt- or its derivative strains (50 μg DNA, ~107 parasites, 2 kV, 50Ω, 25 μF, 250 μs) using a BTX electroporation instrument. Extracellular parasites were suspended in filter-sterile Cytomix (120 mM KCl, 0.15 mM CaCl2, 10 mM K2HPO4/KH2PO4, 25 mM HEPES, 2 mM EGTA, 5 mM MgCl2, pH 7.6) for transfection and selected for drug resistance.
All UPKO-based plasmids allowed a targeted insertion of the PSS or PTS expression cassettes at the uracil phosphoribosyltransferase (UPRT) locus. Tachyzoites with a disrupted UPRT gene locus were selected using 5 μM of 5-fluorodeoxyuridine (FUDR) [34]. Stable transgenic parasites expressing PSS or PTS were transiently transfected with the pTgTUB8-TgDer1-GFP for colocalization studies. Der1 (degradation in ER 1) protein has long been shown to mediate the ER-associated protein degradation in yeast [35], and more recently in T. gondii [36]. TgDer1-GFP has also been used before to ascertain the localization of other ER proteins [37]. The PTS deletion construct pTKO-5’COS-HXGPRT-3’COS contained COS flanking a hypoxanthine-xanthine-guanine-phosphoribosyltransferase (HXGPRT) expression cassette, which enabled transgenic selection by MPA (25 μg/ml) and xanthine (50 μg/ml) [38]. 5’COS (0.9 kb) and 3’COS (0.8 kb) were amplified using the genomic DNA isolated from host-free tachyzoites and cloned at NotI/EcoRI and HpaI/HpaI sites of the pTKO vector, respectively. Plasmid was linearized with ApaI prior to transfection. Clonal drug-resistant parasites were isolated by limiting dilution and screened for 5’- and 3’-recombination events at the TgPTS gene locus using applicable primers (S1 Table). The PTS-complemented strain (Δtgpts/TgPTS-HA) was created by transforming the Δtgpts mutant with the pTgGRA1-UPKO-TgPTS-HA plasmid using FUDR selection [34].
To make the Δtgpts/TgPSS-2HA-DD strain, the TgPSS gene in the Δtgpts mutant was tagged with a C-terminal 2HA-DD epitope by 3’-insertional tagging (3’-IT). A 1.1-kb COS targeting the 3’-end of the TgPSS gene was amplified from tachyzoite gDNA and cloned into the PacI-digested vector (pLIC-DHFR-2HA-DD) by ligation-independent cloning (LIC, Clontech). The pLIC-DHFR-TgPSS-2HA-DD-3’IT construct was linearized by NsiI in the first half of the COS, transfected into the Δtgpts strain and selected by pyrimethamine (1 μM) [39]. The Δtgpts/TgPSS-2HA-DD strain expressed TgPSS-2HA-DD under the control of endogenous promoter and TgTUB-3’UTR.
All experiments were set up using fresh syringe-released extracellular tachyzoites. For plaque assays, 100–200 parasites of each strain were used to infect HFF monolayers in six-well plates. Parasitized cells were incubated for 7 d, fixed with cold methanol, and then stained with crystal violet. Plaques were imaged and scored for their sizes and numbers using the ImageJ software (NIH, US). To test the gliding motility, parasites were incubated on BSA (0.01%)-coated coverslips in Hanks Balanced Salt Solution (HBSS) for 15 min at 37°C. Samples were fixed in 4% paraformaldehyde and 0.05% glutaraldehyde (10 min), and then stained with anti-TgSag1 and Alexa488 antibodies. Motile fractions and trail lengths were quantified using the ImageJ software.
For invasion and egress assays, HFF monolayers cultured on glass coverslips were infected with fresh parasites for 1 hr (MOI, 10) or for 40–72 hr (MOI, 1), respectively [40]. Samples were subsequently fixed with 4% paraformaldehyde and 0.05% glutaraldehyde in PBS (2 min), neutralized by 0.1 M glycine/PBS (5 min), and then blocked in 3% BSA/PBS (30 min). Noninvasive parasites or egressed vacuoles were stained with anti-TgSag1 antibody (1:1,500, 1 hr) prior to detergent permeabilization. Cells were washed 3x with PBS, permeabilized with 0.2% triton X 100/PBS (20 min), and stained with anti-TgGap45 antibody (1:3,000, 1 hr) to visualize intracellular parasites. Samples were washed and immunostained with Alexa488 and Alexa594-conjugated antibodies (1:3,000, 1 hr). The number of invaded parasites was deduced by immunostaining with anti-TgGap45/Alexa594 (red), but not with anti-TgSag1/Alexa488 (green). The egressed vacuoles were scored directly from the number of vacuoles with TgSag1-stained parasites.
Localization of epitope-tagged proteins was performed by immunofluorescence assays. The method was essentially the same as described for invasion assays except for that samples were permeabilized prior to incubation with antibodies. A panel of organelle-specific antibodies (TgMic2 for micronemes, 1:1,000; TgRop1 for rhoptries, 1:1,000; TgGra5 for dense granules, 1:500; TgF1B for mitochondrion, 1:1,000; TgFd for apicoplast, 1:500; TgVP1 for acidocalcisomes/plant-like vacuole, 1:500) was used together with anti-HA antibody (1:5,000; Sigma-Aldrich, Germany) to assess localizations of epitope-tagged PSS and PTS proteins. Images were acquired using ApoTome microscope (Zeiss, Germany).
The M15/pREP4 strain was transformed with the empty pQE60 expression vector (Qiagen), pQE60-TgPTS, pQE60-TgPSS, or pQE60-AtPSS [17] constructs and cultured in Luria-Broth medium supplied with ampicillin (100 mg/L) and kanamycin (50 mg/L). Protein expression was induced by 1 mM IPTG at 25°C in overnight cultures containing 5 mM threonine or serine, followed by a 4 hr incubation at 37°C. Lipids were isolated and separated by one-dimensional TLC in chloroform/methanol/acetate (130:50:20) and visualized by ninhydrin staining.
Parasites were syringe-released from infected HFF (MOI, 3; 42–48 hrs of infection) and passed twice through 23G and 27G needles. Host debris was removed by filtering the parasite suspension through a 5 μm filter (Merck Millipore, Germany). Cell pellets (0.5-1x108 parasites) were resuspended in 0.4 ml of PBS and lipids were extracted according to Bligh-Dyer [41]. Briefly, 0.5 ml chloroform and 1 ml methanol were mixed to the samples, which were allowed to stand for 30 min and centrifuged (2,000 g, 5 min). The supernatant was transferred to a glass tube followed by addition of chloroform and 0.9% KCl (1 ml each). Samples were mixed, centrifuged and the lower chloroform phase containing lipids was transferred to a conical glass tube. Samples were stored at −20°C in the airtight glass tubes flushed with nitrogen gas. Lipids were resolved by two-dimensional TLC on silica gel 60 plates (Merck) using chloroform/methanol/ammonium hydroxide (65:35:5) and chloroform/acetic acid/methanol/water (75:25:5:2.2) as the solvents for the first and second dimensions, respectively. They were visualized by staining with iodine vapors and identified based on their migration with authentic standards (Avanti Lipids). The major iodine-stained phospholipid bands were scraped off the silica plate, and quantified by chemical phosphorus assay, as described elsewhere [42].
Total lipids (0.5–1 x 108 tachyzoites) were fractionated on chloroform-equilibrated silica 60 columns. Neutral lipids were eluted by acetone washing of the column. Phospholipids were subsequently eluted in 5 column-volumes of chloroform/methanol/water (1:9:1). Each lipid fraction was collected, dried under nitrogen stream at 37°C, and stored at −20°C for downstream assays. Internal standard PtdCho (44:2) was mixed with extracted lipids to calibrate the recovery yield of major lipids. 10−20 μl aliquots of phospholipid extract in chloroform/methanol (1:1) were introduced onto a HILIC column (Kinetex, 2.6 μm) at a flow rate of 1 ml/min to resolve different phospholipid classes, essentially as described elsewhere [43]. Column effluent was introduced into either a 4,000 Q-TRAP mass spectrometer (AB Sciex, Framingham, MA) or LTQ-XL (Thermo Scientific, Waltham, MA), and analyzed in the negative ion mode using electrospray ionization. Data were processed using the proprietary software of the respective instrument manufacturers. Lipidomics data reported in this work have been deposited in the Dryad repository [44]: http://dx.doi.org/10.5061/dryad.564sc
C57BL/6J mice were infected with extracellular tachyzoites of the RHΔku80-hxgprt- (parental), Δtgpts or Δtgpts/TgPTS-HA strains. Parasites for in vivo infections were propagated in HFF cells. Fresh host-free tachyzoites were syringe-released after 40 hr of infection, filtered (5 μm), and then injected via intraperitoneal (i.p.) route (50 parasites of the parental and Δtgpts/TgPTS-HA strains; 5 x 102 or 5 x 103 of Δtgpts strain). Animals were monitored for mortality and morbidity 3 times a day over a period of 4 wk. An inoculum of 50 parental tachyzoites (type I) was used to challenge the Δtgpts-immunized animals, which were monitored for additional 4 wk.
Cysts were harvested from the brains of female NMRI mice infected with T. gondii of the ME49 strain 5 to 6 months earlier (i.p.), as described before [45]. The Δtgpts-vaccinated mice (500 parasites) were challenged with the type II parasites (ME49, 3 cysts i.p. in 200 μl) 4 wk after the primary infection. A control group of naïve animals was also included. Parasite burden in the mouse brain was estimated by counting cysts and semiquantitative real-time PCR following another 4 wk of infection with the ME49 strain. Brain tissue was mechanically homogenized in 1 ml sterile PBS and cysts were counted using a light microscope. For qPCR, perfused brain tissue samples were snap-frozen and stored at −80°C [46]. 30 mg tissue was used to purify nucleic acids (QIAgen kit). FastStart Essential DNA Green Master (Roche, Germany) was mixed with genomic DNA (90 ng) in triplicate reactions, which were developed in a LightCycler 480 Instrument II (Roche, Germany). The parasite burden (target: TgB1 gene) was estimated relative to mouse (reference: argininosuccinate lyase, MmASL). Primers for the TgB1 and MmASL genes are listed in S1 Table. For cerebral histopathology, brain tissues isolated from infected animals were immersed in 4% paraformaldehyde for several days. Samples were embedded in paraffin, sliced into 4-μm thick sections, deparaffinized and then stained with hematoxylin-eosin stain, as described elsewhere [47]. Slides were developed using the Bond polymer refine detection kit (Menarini/Leica, Germany). Tissue sections were scanned at 230 nm resolution using a MiraxMidi Scanner (Zeiss MicroImaging GmbH, Germany) [48].
|
10.1371/journal.pgen.1000175 | Segmental Duplications Arise from Pol32-Dependent Repair of Broken Forks through Two Alternative Replication-Based Mechanisms | The propensity of segmental duplications (SDs) to promote genomic instability is of increasing interest since their involvement in numerous human genomic diseases and cancers was revealed. However, the mechanism(s) responsible for their appearance remain mostly speculative. Here, we show that in budding yeast, replication accidents, which are most likely transformed into broken forks, play a causal role in the formation of SDs. The Pol32 subunit of the major replicative polymerase Polδ is required for all SD formation, demonstrating that SDs result from untimely DNA synthesis rather than from unequal crossing-over. Although Pol32 is known to be required for classical (Rad52-dependant) break-induced replication, only half of the SDs can be attributed to this mechanism. The remaining SDs are generated through a Rad52-independent mechanism of template switching between microsatellites or microhomologous sequences. This new mechanism, named microhomology/microsatellite-induced replication (MMIR), differs from all known DNA double-strand break repair pathways, as MMIR-mediated duplications still occur in the combined absence of homologous recombination, microhomology-mediated, and nonhomologous end joining machineries. The interplay between these two replication-based pathways explains important features of higher eukaryotic genomes, such as the strong, but not strict, association between SDs and transposable elements, as well as the frequent formation of oncogenic fusion genes generating protein innovations at SD junctions.
| Duplications of long segments of chromosomes are frequently observed in multicellular organisms (∼5% of our genome, for instance). They appear as a fundamental trait of the recent genome evolution in great apes and are often associated with chromosomal instability, capable of increasing genetic polymorphism among individuals, but also having dramatic consequences as a source of diseases and cancer. Despite their importance, the molecular mechanisms of formation of segmental duplications remain unclear. Using a specifically designed experimental system in the baker's yeast Saccharomyces cerevisiae, hundreds of naturally occurring segmental duplications encompassing dozens of genes were selected. With the help of modern molecular methods coupled to detailed genetic analysis, we show that such duplication events are frequent and result from untimely DNA synthesis accidents produced by two distinct molecular mechanisms: the well-known break-induced replication and a novel mechanism of template switching between low-complexity or microhomologous sequences. These two mechanisms, rather than unequal recombination events, contribute in comparable proportions to duplication formation, the latter being prone to create novel gene fusions at chromosomal junctions. The mechanisms identified in yeast could explain the origin of a variety of genetic diseases in human, such as hemophilia A, Pelizaeus-Merzbacher disease, or some neurological disorders.
| In humans, segmental duplications (SD) cover up to 5.2% of the genome [1] and are responsible for numerous gene-dosage imbalances [2], gene fusions and disruption events 3,4,5. Together with large insertions/deletions, SDs lead to gene copy number variations (CNVs) which represent a major source of polymorphism between individuals [6]. They have been associated with the development and evolution of both cancers [7],[8],[9],[10],[11] and genetically complex phenotypes such as predisposition to autism [12], epilepsy [13], Alzheimer disease [14], glomerulonephritis [15], systemic autoimmunity [16] and susceptibility to HIV/AIDS infections [17]. A specific mapping of CNVs on human chromosome 22 revealed that more than 2/3 of the breakpoints intersect with SDs [18]. This strong correlation reflects the similar nature of CNVs and SDs and suggests tightly coupled co-evolution mechanisms [19].
We previously designed a gene dosage assay in Saccharomyces cerevisiae to screen for the spontaneous duplication of a single gene, RPL20B [20]. Although the size of this gene is relatively small (1.6 kb), no single gene duplication was ever found. Instead, only intra- and inter-chromosomal duplications of large DNA segments, encompassing dozens of neighboring genes, were recovered (88% and 12%, respectively, Figure 1A) [20]. These findings showed that spontaneous SDs can compensate for gene dosage imbalance by altering gene copy number in the yeast genome and that CNVs can encompass numerous genes. Approximately half of the SD junctions involved dispersed repeats such as Long Terminal Repeats (LTRs) from Ty retroposons, while the other half consisted of low complexity DNA sequences (poly A/T, trinucleotide repeats), as well as microhomologous sequences whose identity spans only over a few nucleotides in length. The location and the type of sequences found at the breakpoints suggested that SDs might result from replication accidents improperly repaired through both homologous and non-homologous recombination events [20]. In order to explore the mechanisms of SD formation, we deciphered how perturbations of the replication process and of double strand break (DSB) repair pathways affect rates, types, sizes and breakpoint sequences of duplications. Providing the largest set of experimentally generated de novo duplications, the present study describes 338 independent SDs recovered in different mutant backgrounds and culture conditions. We show that replication-generated DNA ends are converted into large SDs through both homology-dependent and -independent replication-based mechanisms.
Two highly similar paralogous genes, RPL20A (YMR242c) and RPL20B (YOR312c), encode the Rpl20 yeast ribosomal protein. The deletion of RPL20A results in a marked slow-growing phenotype which can be compensated by the spontaneous duplication of RPL20B [20]. Slow growing parental strains (rpl20AΔ) are propagated through serial transfer into rich medium. Rapidly growing revertants among slow growing populations are isolated by regularly plating aliquots of the cultures at each transfer step. Using this assay, we re-estimated the spontaneous duplication rate of RPL20B to be 1×10−7 SD/cell/division (Luria-Delbruck fluctuation tests using the 0 term of the Poisson law (p = 1−elnf0/ndiv; see Methods; Table 1). This value is higher than previously estimated (between 2×10−9 and 10−10 SD/cell/division [20],[21]) due to an initial underestimation of the time needed for a duplication-carrying cell to overtake the population of the slow growing parental cells (see Methods).
To confirm this surprisingly high value, an independent estimation of the duplication rate was achieved by designing a new selection assay based on the recovery of uracil prototrophy instead of growth recovery. In this system, RPL20A is not deleted and therefore both parental and duplicated strains show the same growth rate. Two truncated copies of the URA3 gene, overlapping by only 58 bp, were introduced in place of the two Ty3 LTRs, YORsigma3 and YORsigma4 located on either side of RPL20B and separated by 115 kb (YKFB614, Figure 1B). In the original growth-assay, approximately half of all SDs (48%, [20]), corresponds to an intra-chromosomal 115 kb direct tandem duplication between these two LTRs (Figure 1B). The size of the URA3 overlapping sequences (58 bp) is comparable to the largest identity region shared by the two LTRs (44 bp). Thus, recovery of a functional URA3 gene at the duplication breakpoints is indicative of direct tandem ura3-mediated SDs, mimicking the 115 kb LTR-mediated SDs. In this system, the duplication rate was evaluated to 0.9×10−7 event/cell/division (using the median method [22], Table 1). To further test this rate, we created a rpl20AΔ derivative of YKFB614 (YKFB605, Table S1) and examined its duplication rate using the growth recovery assay. We found a rate of 1.7×10−7, consistent with the fact that the rate derived from the URA3 assay represents only half of real duplication rate and close to our present estimate of 1×10−7.
This rate only accounts for duplications encompassing the RPL20B reporter gene, located on the right arm of chromosome XV. Therefore, extrapolation to the whole genome would lead to a much higher rate, suggesting that spontaneous SD events must be extremely common in yeast populations. For instance, a very high rate of histone gene amplification, compensating for decreased level of histones, was shown to result from recombination events between two Ty1 retroelements leading to supernumerary circular chromosomes [23]. However, our present estimate of SD rate is several orders of magnitude higher than that of other types of chromosomal rearrangements characterized in different studies using native yeast chromosomes [24],[25]. This discrepancy could be explained by the absence of spatial constraints imposed on the boundaries of the SDs in our screen while in the other studies, the location of one end of the rearrangements is restricted within a narrow chromosomal region.
To investigate the molecular mechanisms involved in SD formation, we used our selection system in conditions where replication is altered. Clb5 is a B-type cyclin known to activate late replication origin: in a clb5Δ strain S-phase duration is increased and the replication pattern modified [26],[27],[28]. The rate of SD formation is greatly increased in clb5Δ (730x compared to the control strain, Table 1), unveiling the broad genomic instability induced by the perturbation of replication origin firing. Interestingly, the relative proportions of intra- versus inter-chromosomal SDs are conserved compared to the wild-type (WT) strain (Table 1). Although this is at the limit of statistical significance, the proportion of the 115 kb LTR-mediated duplications (between the two Ty3 LTRs, YORWsigma3 and YORWsigma4, Figure 1B) is slightly increased in clb5Δ (62% compared to 48% in the WT, P = 0.05 Fisher's exact test, Table 1). It is noteworthy that these LTR sequences lie next to tRNA genes whose transcription by PolIII is known to stall the progression of replication forks [29]. The size distribution of the intra-chromosomal SDs in clb5Δ remains globally similar to that of the WT (see Figure 1C, in which WT and clb5Δ strains have radically different SD distributions as compared to rad52 and rad1 mutants).
The breakpoint sequence of a non-LTR mediated SD was characterized through comparative genomic hybridization (CGH) and PCR amplification, revealing the presence of microhomologies at the junction (Figure 2). This junction is identical to the one found in the strain YKF1080 strain isolated in our original control screen [20]: the same two copies of a 9 nucleotide microhomologous sequence (ACTTTTTTT) have been involved in the formation of two independent SDs, recovered in two different genetic backgrounds. There are 2367 copies of this sequence in the genome, 47 of which interspersed between the two recombining sequences. It is unlikely that this repetitive use occurred by chance and therefore must be indicative of a chromosomal rearrangement hotspot. Interestingly, the centromere-proximal sequence lies next to an autonomous replicating sequence (ARS1524) and the centromere-distal one corresponds to a replication termination site (Figure 2), which could explain their recurrent use in SD formation (below).
Altogether these results suggest that the mechanisms of SDs formation are similar in WT and clb5 strains. In addition, the dramatic SD rate increase associated with the clb5 mutation could be related either directly to the perturbed S-phase origin firing and/or to indirect effects of this perturbation onto replication. In this regard, the reported Rad9-dependent activation of the replication checkpoint key protein Rad53, by late S-phase, strongly suggests that a CLB5 deletion results in the formation of replication-induced DNA breaks [30]. Such breaks could therefore represent the precursor lesions leading to SDs.
In order to test whether broken replication forks could correspond to these precursor lesions, we monitored SD formation in cells treated with camptothecin (CPT), a topoisomerase I inhibitor. CPT stabilizes the covalent intermediate that forms during the catalytic DNA nicking-closing cycle of Top1, and CPT cytotoxicity results from the conversion of single strand nicks into double-stranded DNA ends when a moving replication fork collides with a CPT-Top1 complex [31]. The rate of SD formation is strongly increased in an exponential culture treated for 3 hours with 10 µg/ml of CPT (x 320, Table 1). This observation could be explained if the precursor lesions leading to SDs were indeed double-strand DNA ends that in standard conditions would result from replication accidents. Several other lines of evidence support this hypothesis. First, SD breakpoints often correspond to sequences known to interfere with the replication forks progression (Figure 2 and [20]). Moreover, replication-induced DNA damages in a clb5Δ strain [30] would explain the massive increase in SD formation observed in the absence of this cyclin.
These lesions are likely to impede fork progression and trigger the activation of the replication checkpoint. Besides preventing fork collapse and the subsequent formation of DNA breaks, the replication checkpoint also regulates a large variety of cellular events including repression of late-replicating origins, inhibition of mitosis and induction of DNA repair genes [32]. This trans-acting branch of the replication checkpoint relies on the hyperphosphorylation of Rad53 that can be specifically abrogated with a mrc1AQ allele [33]. To determine whether SDs result primarily from S-phase induced DSBs rather than being secondary byproducts of the checkpoint activation, we characterized SD formation in a mrc1AQ mutant in presence of hydroxyurea (HU). By inhibiting the ribonucleotide reductase activity, HU slows down replication fork progression and promotes the formation of ssDNA at the forks, which is sufficient to activate the checkpoint in normal cells [32],[34]. In a mrc1AQ strain and in the presence of HU (100 mM for 3 hours; Methods), the integrity of stalled replication forks is maintained while the trans-acting branch of the replication checkpoint is suppressed. In these conditions we found no significant differences between HU-treated and untreated cultures in either checkpoint competent or deficient cells (2 to 5 fold increase, Table 1). These results demonstrate that neither stalled replication forks nor the Rad53 hyperphosphorylation-mediated functions of the replication checkpoint are sufficient to stimulate SD formation. Altogether, the above findings strongly suggest that broken forks are the precursor lesions that are directly processed into SDs.
Free DNA ends generated at broken forks are thought to be repaired primarily by strand invasion of the sister-chromatid, followed by the assembly of a new fork and subsequent replication up to the chromosome end (or to the next replication fork). This break-induced replication (BIR) mechanism can occur through successive rounds of strand invasion and dissociation, and lead to chromosomal rearrangements if reinvasion occurs within ectopic repeated sequences [35]. We explored a potential role for BIR-related mechanisms by investigating SD formation in a pol32Δ strain. Pol32 is a non-essential subunit of S. cerevisiae major replicative DNA polymerase Polδ and is required for the replication fork assembly that initiates the BIR reaction [36]. Absence of Pol32 completely abolishes the formation of SDs. No mutant carrying duplications of any type were isolated out of 184 independent pol32Δ cultures. Thus, although the true duplication rate cannot be calculated, the occurrence of a single event, out of the 184 cultures, would have lead to a reversion rate of 6.9×10−9. Although this value is an overestimate of the true duplication rate, it represents a 14-fold reduction compared to the WT control (<0.07, Table 1). These data reveal the crucial role played by Pol32 in the generation of all types of SDs. Given that Pol32 is not required for repair by gene conversion (GC) events, SDs must therefore result principally from replication-based mechanisms rather than from unequal crossing-overs (UCO) between sister-chromatids. In addition, since only half of SDs contain repeated homologous sequences at their junctions [20], classical BIR mechanism involving Rad52-mediated interactions between large sequences of homology could only account for half of all of the events: the other half might result from a Pol32-dependent replication-based mechanism involving microhomologous or low complexity sequences at the site of strand invasion (see below).
These two Pol32-dependant replication-based mechanisms must rely on an initial step of ectopic strand invasion. The Rad1/Rad10 complex possesses an endonuclease activity required for the removal of non-homologous tails during GC events [37]. This complex is also essential for the Rad52-independent microhomology mediated end-joining (MMEJ) DNA repair pathway [38] and was shown to promote the production of gross chromosomal rearrangements (GCRs) [39]. A deletion of the RAD1 gene results in a 5-fold reduction of SD formation (x 0.2 as compared to WT, Table 1), suggesting that the endonuclease activity is required to generate duplications. We also noted a substantial (although not highly significant) decrease in the proportion of LTR-mediated SDs compared to WT (14% vs. 48%, respectively; P = 0,06, Table 1), suggesting that Rad1 is directly involved in the generation of BIR-mediated SDs rather than of microhomology-related ones. Consistently, the proportion of small (<115 kb) intra-SDs is increased and reminiscent of the distribution of rad52-independent duplications (Figure 1C; below).
One SD breakpoint from a rad1Δ mutant was sequenced, revealing an eight-nucleotide homology at the junction (Figure 2) and implying that, despite its predominant role in MMEJ, Rad1 is not required for the generation of microhomology-mediated SDs. Similar microhomologies were reported at the junction of GCRs recovered in rad1Δ and rad10Δ strains [39]. Interestingly the centromere-proximal microhomologous sequence involved in this rearrangement lies within the tRNA (tA (UGC)O; Figure 2) that flanks YORWsigma3 (the LTR recurrently used in the 115 kb intra-chromosomal SDs). Given that tRNAs transcription is able to stall incoming replication forks, these sequences were proposed to exhibit spontaneous fragility and thus promote chromosomal instability [40]. The eight-nucleotide microhomology sequence could therefore represent the recurrent breaking site which initiates the formation of the common 115 kb LTR-mediated SD: in the presence of Rad1, the 3′ flap sequence between this break site and the LTR sequence would be excised so that a BIR-mediated SD could occur.
It is generally believed that most SDs must result from non-allelic recombination events between dispersed repeats, but so far no demonstration for the involvement of the homologous recombination (HR) pathway in SD formation has been clearly established. In a rad52Δ strain where HR is abolished the class of 115 kb LTR-mediated SDs is completely suppressed (0 out of 71 independent events compared to 23 out of 48 duplications in the WT, P<10−6, Table 1). This result clearly demonstrates that this class of SDs results from Rad52-dependent recombination events between interspersed repeats. These duplication events are most likely resulting from a BIR reaction, since they are also dependent on the presence of Pol32 (see above). Furthermore, while pol32Δ exhibits a limited reduction in GC efficiency [36], absence of Rad51 restricts both BIR and GC events, although BIR occurs more frequently than GC among the remaining events [41],[42],[43]. In a rad51Δ strain, the rate of SD formation is increased (x 7.7, Table 1). This increase suggests that the lesions that were repaired in wild type through gene conversion or allelic BIR are channeled into non-allelic BIR in rad51 mutants. In addition, the proportion of inter-chromosomal SDs increases up to 32% (10 out of 31 events) as compared to 12% in the WT (6 out of 48, P = 0.02). All types of LTR-mediated SDs are favored in the absence of Rad51 (71% vs. 48% for the control, P = 0.02, Table 1). These findings suggests that Rad51 prevents recombination events between diverged sequences, such as the two LTR repeats YORWsigma3 and YORWsigma4 which share only 76% identity over 319 bp (largest identical domain: 44 bp). This is consistent with the fact that Rad51-independent BIR requires shorter identical regions to achieve strand invasion than Rad51-dependent repair (∼30 bp vs. ∼100 bp, respectively [44]). Therefore, it might be that RAD51 does not simply suppress recombination between diverged sequenced, but normally promotes gene conversion (and allelic BIR) that usually outcompetes ectopic BIR.
Altogether, the above results strongly suggest the following scenario for the formation of the class of 115 kb LTR-mediated SDs: (i) a DNA free end would arose from a broken replication fork in the vicinity of LTR YORWSigma3 (potentially stalled within the tA (UGC)O tRNA gene), (ii) repair of the DSB occurs through a Rad51-independent strand invasion of the non-allelic LTR sequence, YORWSigma4, (iii) followed by a Rad1-dependent 3′ flap removal and (iv) a Pol32-dependent conversion of this strand annealing intermediate into a replication fork generating a large intra-chromosomal SD through BIR (Figure 1D).
To further explore the contribution of homologous recombination to SD formation, a system where SDs result principally from HR events was designed. The two LTR sequences, YORWsigma3 and YORWsigma4, were replaced in this strain YKFB608 by two truncated copies of the URA3 gene, overlapping with a 401 bp region of perfect identity, such that a URA3-mediated intra-chromosomal duplication would restore uracil prototrophy (Figure 1B, Table S1). As expected, all growth revertants isolated in an rpl20AΔ background resulted from duplication events corresponding to URA3-mediated SDs (data not shown). Although the size of the URA3 overlapping sequences is similar to the size of the LTRs (401 and 319 bp, respectively), the rate of SD formation showed a 56 time increase compared to the original strain with intact LTRs (5.6×10−6 vs. 1×10−7, respectively) and a 62 time increase compared to a strain carrying only a 58 bp overlap (5.6×10−6 vs. 0.9×10−7, Table 1). These results confirm that the accumulation of divergence between dispersed repeats suppresses genome rearrangements, while increasing the length of sequence identity between these repeats promotes genomic instability. Indeed, the mismatch repair system can trigger an anti-recombination activity thereby limiting chromosome rearrangements between diverged repeats [45]. In addition, we monitored the effect of the POL32 deletion in this HR-based assay. In the absence of Pol32, and in the absence of mismatches between repeated sequences, only a 23-fold increase is observed, as compared to the 62-fold increase characterized in the presence of this protein (Table 1). This corresponds to a 2.7-fold decrease (63/23) in the rate of uracil-prototroph formation in pol32Δ, a lesser effect that the >14-fold decrease observed in the growth recovery assay (above). It also shows that in the absence of mismatches between repeated sequences, not all SDs require Pol32. These Pol32-independent SDs likely result from UCOs between the repeated identical URA3 sequences. In the original assay, similar UCO events involving the flanking LTRs are probably suppressed due to divergence between the sequences.
The rate of SD formation in a rad52Δ strain is slightly higher than in WT (2.8-fold increase, Table 1), revealing that duplications can form even when HR is abolished (as suggested previously in [46], although using a very different system). The SDs recovered in a rad52Δ background appear radically different from those obtained in the WT strain. First, there is a significant decrease in the proportion of inter-chromosomal events, since all 71 SDs but one correspond to intra-chromosomal duplications (versus 6 out of 48 events in the WT, P = 0.02, Table 1). Second, the size distribution of intra-chromosomal SDs is significantly biased towards smaller segments as most of them (57 out of 71) are smaller than 115 kb (Figure 1C). Third, sequencing of the breakpoints revealed that only microhomologous (between 8 and 9 nt) and low complexity sequences (polyT) are now used to generate SDs (Figure 2). Interestingly, a recent report proposed that the large SDs in the human genome that cause the dysmyelinating PMD disease might result from replication fork stalling followed by homology-independent template switching, relying instead on the presence of microhomologies [47]. Our sequenced breakpoints once again coincide with replication-related elements, such as ARS, termination sites and tRNAs (Figure 2). Given the location and the nature of the initiating lesions, as well as the strict dependency to Pol32 (see above), we conclude that the non-HR mediated SDs result from a new mechanism that would rely on an initial Rad52-independent recombination event, occurring between 5 to 10 bp of microhomology or stretches of low-complexity DNA sequences such as microsatellites, followed by a Pol32-dependent fork assembly initiating DNA synthesis (Figure 1D). Therefore, we propose to designate this new mechanism MMIR for microhomology/microsatellite-induced replication.
All of the above data clearly show that spontaneous SDs result from replication-based mechanisms. Nevertheless, the putative contribution of NHEJ to SD formation was addressed. NHEJ is strictly dependent on the activity of the ATP-dependent DNA ligase, Dnl4 (also named Lig4), as well as that of the Yku70/Yku80 DNA binding complex [48],[49] When DNL4 is deleted, SDs arise at a slightly lower frequency (x 0.8, Table 1), and present a similar proportion of LTR-mediated events (Table 1). Among the non-LTR mediated events, two junctions were sequenced. One lies next to a microsatellite (GTT)14 identical to the one found in the WT strain YKF1057 [20], again corresponding to the recurrent use of a particular sequences at SD boundaries. The other corresponded to a 10 bp-long sequence of microhomology (TGACGCAAAT), repeated 109 times in the genome, in which the two recombining copies lie next to a tRNA gene and a replication termination site (Figure 2). Although all of these characteristics are very similar to SDs generated in the WT strain, there is, however, a significant decrease in inter-chromosomal duplications (0 out of 51 in dln4Δ versus 6 out of 48 in WT, P = 0.01, Table 1), suggesting that Dnl4 is required for inter-chromosomal SD formation. However, the junction sequences of the 6 inter-chromosomal events in WT were indicative of either LTR-mediated or microsatellite-mediated events (3 occurrences each, respectively) [20],[21]. These sequences differ strongly from those usually found at NHEJ-mediation junctions (1–4 nucleotides complementary sequences, [50]), suggesting that, in addition to its well-described role in NHEJ, Dnl4 might participate in the replication-based mechanisms of inter-chromosomal SD formation.
In the double mutant rad52Δ dnl4Δ the rate of SDs formation is moderately increased (x 4.3) compared to WT (Table 1). It is noteworthy that in the GCR assay, developed by Kolodner and collaborators, the concomitant deletion of RAD52 and DNL4 completely abolished the formation of non-reciprocal translocations since all GCRs observed resulted from telomere additions [51]. This discrepancy underlines the differential genetic requirements between SD and other GCR mechanisms. Deletion of RAD1 in the rad52Δ dnl4Δ strain reduced the SD rate to a level similar to that of the WT (Table 1), as expected since Rad1 promotes SDs formation (above). The type of SDs, the size distribution as well as the breakpoint sequences isolated in the progenies of these double and the triple mutants strains, are similar to the ones characterized in the rad52 single mutant (Figures 1C, Table 1 and Figure 2). Therefore, when both HR and NHEJ are abolished and when MMEJ is, at least, severely compromised (as in rad52Δ dnl4Δ rad1Δ strain), SDs still occur at a WT rate. Since SDs would mainly result from the replicative-repair of a one-ended DSB generated at a broken fork, the concomitant mutations of the 3 major DSB repair pathways should severely reduce if not abolish SD formation. The maintenance of a rate of formation similar to WT and the physical characteristics of SDs in this background suggest that MMIR could represent a new DSB repair pathway. Alternatively, these SDs could be formed by template switching, in the absence of any DSB, as suggested for the formation of PLP1-encompassing SDs in the human genome [47].
Altogether, 26 SD breakpoints were sequenced (this work and [20],[21]) allowing the identification of 13 different chimerical Open Reading Frames (ORF) containing either microhomologies or trinucleotide repeats at their junctions (Figure 2). Microhomologies were found at breakpoint junctions in rad52Δ, dnl4Δ and rad1Δ backgrounds, where HR, NHEJ and MMEJ are impaired, respectively (Figure 2). This shows that these sequences can be used in the absence of all known DSB repair pathways. Because of their extremely high genomic density, the impact of microhomologies in SD formation, and more generally in genome dynamics, is likely to be important. For instance, the 8 to 10 nucleotide breakpoint sequences characterized in the rad52Δ, dnl4Δ and rad1Δ backgrounds are found in the S. cerevisiae genome from 109 times for the less frequent (TGACGCAAAT), and up to 793 times for the most common (TAGAGGA, Figure 2). Chimerical genes arise either from in- or out-of-frame ORF fusions (3 occurrences each), from 3′ or 5′ ORF truncations (1 and 5 occurrences, respectively) or from the fusion between an ORF and a tRNA (Figure 2). These fusions can generate new proteins and thus represent a potential mechanism of protein evolution. Whereas chimerical ORFs resulting from translocation and inversion events are associated with the concomitant lost of the original gene integrity, SD-mediated chimerical genes formation leave intact the original copies of the genes involved at the breakpoint. For instance, in addition to the original full-length gene a truncated copy of SGS1 (homolog of human BLM) has been found in the pathogen yeast species Candida glabrata [52]. This powerful mechanism allows SD-mediated chimerical genes to explore new combinations that might be counter-selected for in the cases of classical translocation- or inversion-mediated events. In-frame ORF fusions (3 cases) might result in new protein architectures by combining previously existing domains. In addition, SD-mediated frameshift fusions and ORF truncations may result in true protein innovations at the junctions by promoting the transcription of otherwise non-coding sequences. The corresponding transcripts would encode entirely new amino acid combinations. For instance, the frameshift chimerical ORF generated in strain YKF1114 comprises a coding sequence whose last 47 amino acids (from the breakpoint to the stop codon) represent a truly new protein segment that shows no similarity to the rest of the yeast proteome. Such peptides were found in 5 out of the 13 chimerical ORFs characterized (Figure 2). Although relatively small (average size of 28 amino acids), these peptides are new genomic features and may generate new protein domains. Despite their known association with diseases and genome rearrangements, it has been proposed that SDs have been fixed in the human genome to increase copy number of fusion genes originating from initial duplications of gene-rich core regions, eventually leading to the emergence of new gene families that are either unique to hominoids or considerably diverged when compared with other mammalian species [53].
Given their close association with various genomic disorders and cancers and their broad evolutionary impact, SDs and CNVs represent one of the most important discoveries that stem from the human genome project. Careful computational characterization of SD breakpoints in the genomes of human and other primates has suggested an important role for Alu-mediated recombination in the production of intra- and inter-chromosomal SDs [54]. However, Alu elements are found in only 30% of the SD breakpoints and sequences presenting the physicochemical properties of “fragile sites” were shown to play an important role as well [55]. In addition, recent studies have proposed that SDs and other complex rearrangements associated with genomic disorders would result from replication-based mechanisms rather than from more classically invoked recombination-based models such as non-allelic homologous recombination between dispersed repeats [47],[56],57. Although essentially based on breakpoint analyses, these studies reach conclusions similar to those drawn here from experimental evidences.
We found a massive SD rate increase both in a clb5Δ strain where origin firing is perturbed, S-phase is lengthened and DNA damages are detected by late S-phase [26],[27],[28],[30] and in CPT-treated cultures in which single-strand nicks are converted into broken forks [31]. The recurrent use of genetic elements known to interfere with replication forks progression at SD breakpoints (tRNA, microsatellites, ARS, replication slow zones and termination regions, Figure 2) also points towards the involvement of replication and the use of broken forks as the initiating lesions in the pathways leading to SDs. In addition, the finding that all SD formation requires the nonessential Polδ subunit Pol32 shows that duplications results from replication-based mechanisms rather than from UCOs, which are suppressed by the natural DNA divergence between dispersed repeats such as LTRs. It also suggests that BIR, which also requires Pol32 to initiate new DNA synthesis [36] would be the mechanism by which SDs are formed. However, BIR is a homologous recombination process which implies an initial Rad52-dependent invasion step necessitating large sequences of homology between the recombining molecules (reviewed in [58]). These requirements imply that BIR cannot be the unique pathway leading to SDs, because only half of the SDs are generated through a Rad52-dependent recombination event between homologous sequences (Table 1). The remaining SDs occur independently from both Rad52 and large homologous regions and are generated through recombination between short identical/low complexity sequences. A Rad52-independent half-crossover pathway was previously described [59],[60] and unequal half-crossovers in G2 could also generate tandem duplications. However, the class of Rad52-independent SDs described here involves only microhomology/microsatellite sequences at breakpoints and requires Pol32, two characteristics that are hardly compatible with the half-crossover pathway. Given its unique substrate and genetic requirements, this new mechanism of SD formation has been called microhomology/microsatellite-induced replication, or MMIR, because it brings together characteristics from both MMEJ (ie. recombination between microhomologies in a Rad52-independent manner, [38]) and BIR (ie. a Pol32-dependent DNA synthesis step, [36]). In addition, we show that MMIR-mediated SDs still form in the absence of all known DSB repair pathways (HR, NHEJ and MMEJ) suggesting that MMIR could represent a new repair pathway. Alternatively, one cannot exclude that MMIR-mediated SDs would arise in the absence of any DSB as a result of template switching events as it has been suggested for the large PLP1 duplications that cause the dysmyelinating PMD disease in human [47].
Altogether, our results provide the first experimental deciphering of the molecular pathways leading to SDs, demonstrating that two alternative replication-based mechanisms, BIR and MMIR, are responsible for the spontaneous SD formation in the yeast genome (Figure 1D). While these two pathways probably use similar precursor DNA lesions and share the Pol32 requirement, they differ from one another by their recombination substrate and their dependency to HR proteins (Rad52, Rad51 and Rad1). To our surprise, the Dnl4 ligase seems to contribute to the formation of inter-chromosomal SDs resulting from either BIR or MMIR. A similar Dnl4 requirement has been described for the formation of non-reciprocal translocations in S. cerevisiae [61]. Dnl4 has a preponderant role in NHEJ and also participates in MMEJ [38],[48],[49]. However, the sequences characterized at inter-chromosomal SD breakpoints (LTRs and microsatellites) are very different from typical signatures of either NHEJ or MMEJ events [38],[50]. These results suggest that the role played by Dnl4 in inter-chromosomal SD formation would be different from the other known functions of this protein.
Discrete microhomology/microsatellite sequences are recurrently used at SD breakpoints although hundreds, even thousands, of other identical copies are dispersed within the genome. These particular regions thus behave as duplication hotspots. Interestingly, they often correspond to genetic elements linked to replication initiation, progression and termination (e.g. ARS, termination regions, tRNAs, replication slow zones; Figure 2). Such correlation suggests that genomic architectural constraints may favor interactions between specific loci, for instance through promoting spatial proximity during replication. In yeast, two replication forks originating from the same replicon co-localize in the nucleus within a replication factory, a spatial location likely to harbor other forks as well [62]. The tight link between replication and SD formation raises interesting questions with regard to the influence of these factories on eukaryotic genome stability (Figure 3). A single broken fork could be repaired either in a Rad52-dependent or -independent manner (Figure 3i or ii, respectively). The invading broken strand would presumably correspond to the lagging strand template where more ssDNA is exposed at the forks [32]. Given that SD formation requires Pol32, the displacement of the lagging strand would also be compatible with the recent finding that lagging strand replication is performed by Polδ [63]. SDs recovered in the absence of Rad52 present a relatively smaller size (median = 60 kb), reminiscent of the size of a replicon bubble in yeast. This may proceed from the possibility for a DNA free-end to interact spontaneously in a Rad52-independent manner with a sequence present in its vicinity within in the same replication factory (Figure 3ii). In contrast, in a WT background where Rad52 is present, homology search would promote strand invasion between more distant sequences possibly located in different replication bubbles/factories, and thus generate larger duplications.
Interestingly, in highly aggressive cases of neuroblastoma, an heterogeneous pediatric cancer, segmental chromosome instability results in unbalanced chromosome translocations, sometimes associated with additional aneuploidies [64]. These genomic profiles are formally similar to the different classes of inter-chromosomal duplications characterized in S. cerevisiae [20]. Whereas BIR is the mechanism usually invoked to account for the development of such chromosomal alterations [65], the absence of repeated sequences at the breakpoints of many of these rearrangements suggests that MMIR may be an important path towards development of cancer.
All strains are derivatives of S. cerevisiae BY4743 (MATa/α, his3Δ1/his3Δ1, leu2Δ/leu2Δ, met15Δ/MET15, lys2Δ/LYS2, ura3Δ/ura3Δ) [66]. Strain names and their corresponding genotypes and origins are summarized in Table S1. Mutations were obtained either directly through a PCR-based deletion strategy or from EUROSCARF strains where the original geneticin resistance cassette KanMX4 was replaced by another resistance cassette. All constructions were verified by PCR and Southern blot analysis. For each mutation monitored, a diploid parental strain heterozygous for both the YMR242c (RPL20A) deletion and the deletion of the tested gene(s) was constructed then sporulated. Spores from the progeny carrying both the YMR242c deletion and the tested deletions were analyzed.
In the growth-recovery assay, duplication rates were calculated from Luria-Delbruck fluctuation tests, either by using the 0 term of the Poisson law (p = 1−elnf0/ndiv) when a small subset of all cultures contained revertant cells (see [20] for details), or using the median method when most of the cultures were overtaken by revertants [22]. In previous studies, the doubling time of a revertant culture was estimated to be twice as fast as the slow growing parental strain [20],[21]. However, in the culture conditions where the selection assay was performed (serial dilutions in 6 ml YPD in 24-wells plates), careful measurements revealed that the time needed for revertant cells to overtake slow growing populations was longer than predicted and was strain dependant: the doubling time of a duplicated strain is actually 1.3 to 1.4 times smaller than that of the slow growing parent, depending on the mutant background. This discrepancy resulted in a strong effect on the duplication rate estimation compared to our former studies (from 2×10−9 to 1×10−7 per cell per division in control strain).
In the strains used for the uracil-prototrophy recovery assay, the RPL20A gene is not deleted (see Table S1) and both parental and duplicated strains show the same growth rate. Two truncated copies of the URA3 gene, covering either the 5′ or 3′ half of the gene and overlapping by either 58 bp or 401 bp, were introduced in place of YORWsigma3 and YORWsigma4 (strains YKFB614 and YKFB608, respectively, Figure 1B). The rate of appearance of uracil autotrophic colonies was determined by a fluctuation test analysis using the median method [22]. Briefly, ten independent YPD cultures, inoculated with ∼200 cells, were grown at 30°C to ∼3×108 cells/ml. Cells were plated on uracil lacking medium, incubated at 30°C for 2 days and [ura+] colonies were counted. The breakpoint junction indicative of a 115 kb ura3-mediated direct tandem duplication was sought through PCR amplification of the region. All [ura+] colonies analyzed carried such duplications, resulting from the fusion of the two URA3 overlapping sequences.
Independent colonies (2×107 cells) from strains YKF120c and YBaG398 were inoculated in 24 wells plates containing 6ml YPD, and cultivated under agitation during 6 generations at 30°C. Approximately 2×106 cells from each well were then inoculated into either fresh YPD medium, YPD supplemented by 100 mM Hydroxyurea (HU, Sigma) or YPD supplemented by 10 µg/ml Camptothecin (CPT, Sigma), and incubated for 3 hours. After wash approximately 2×106 cells from each well were inoculated into fresh YPD medium. Every 10–11 generations, similar aliquots from each well were re-inoculated into fresh YPD medium. Between every cycle, a sample of the culture was plated onto YPD plates at a density of ∼2×102 to 5×103 cells/plate and incubated at 30°C (above; [20]).
Electrophoretic karyotypes of parental and revertant strains, as well as genomic DNA extraction and labelling, were performed as described [20]. Labelled DNA was hybridized against either PCR product-based (Ecole Normale Superieure, Paris France and MWG Biotech) or oligo-based yeast whole-genome arrays (Affymetrix, YG-S98). Arrays were analyzed with the GenePix Pro5.0 or with the Affymetrix GeneChip software, respectively. A genomic ratio for each ORF was defined as the ratio between normalized spot intensity of the revertant and parental strains, from which the mean of all spot intensities ratios was subtracted. SD junctions were PCR amplified. Products were purified using gel extraction columns (NucleoSpin, Macherey Nagel) and sequenced by the Genome Express company (Cogenics).
|
10.1371/journal.pgen.1006956 | Recruitment of Armitage and Yb to a transcript triggers its phased processing into primary piRNAs in Drosophila ovaries | Small RNAs called PIWI -interacting RNAs (piRNAs) are essential for transposon control and fertility in animals. Primary processing is the small RNA biogenesis pathway that uses long single-stranded RNA precursors to generate millions of individual piRNAs, but the molecular mechanisms that identify a transcript as a precursor are poorly understood. Here we demonstrate that artificial tethering of the piRNA biogenesis factor, Armi, to a transcript is sufficient to direct it into primary processing in Drosophila ovaries and in an ovarian cell culture model. In the fly ovarian somatic follicle cells, the transcript becomes cleaved in a stepwise manner, with a 5′→3′ directionality, liberating U1-containing ~24 nt piRNAs that are loaded into Piwi. Although uridines are preferred for generation of piRNA 5′ ends, processing takes place even in their absence, albeit at a lower efficiency. We show that recombinant Armi has 5′→3′ helicase activity, and mutations that abolish this activity also reduce piRNA processing in vivo. Another somatic piRNA pathway factor Yb, an interactor of Armi, is also able to trigger piRNA biogenesis when tethered to a transcript. Tethering-mediated primary piRNA biogenesis is also functional in the fly ovarian germline and loads all the three PIWI proteins present in this environment. Our study finds a broad correlation between piRNA processing and localization of the tethered factors to the cytoplasmic perinuclear ribonucleoprotein granules called germline nuage or somatic Yb bodies. We conclude that transcripts bound by Armi and Yb are identified as piRNA precursors, resulting in localization to cytoplasmic processing granules and their subsequent engagement by the resident piRNA biogenesis machinery.
| PIWI-interacting RNAs (piRNAs) are 24–30 nucleotide (nt) small RNAs that are exclusively expressed in animal germlines and are essential for suppression of transposable elements or ‘jumping genes’. Millions of piRNAs are produced from single-stranded transcripts that arise from large RNA polymerase II transcription units called piRNA clusters. Since they resemble other RNA pol II products like protein-coding mRNAs or non-coding RNAs, how piRNA precursors are selectively recruited to the biogenesis machinery is a big mystery. Here we demonstrate that artificial tethering of specific piRNA biogenesis factors to a reporter transcript is sufficient to identify it as a piRNA precursor in Drosophila ovaries and in an ovarian cell culture model. This results in fragmentation of the transcript into thousands of piRNAs, which are generated as a series of non-overlapping (phased) fragments that are loaded into PIWI proteins. Our study indicates that localization of the tethered factors to cytoplasmic perinuclear granules called nuage/Yb bodies is necessary for processing. Any mutations in the tethered factors that disrupt this sub-cellular localization, reduce processing. We believe that our strategy will allow generation of designer small RNAs to target almost any germline gene, and also provides a valuable tool to dissect the molecular mechanism of piRNA biogenesis.
| Bulk of the eukaryotic genomes are composed of genetic material derived from mobile genetic elements called transposons. Their mobility within the genome can cause mutations or deletions, impacting genome integrity [1]. Given the diversity of transposable elements within any genome, small RNAs are used to sequence-specifically identify and repress transposons in organisms ranging from plants to animals. In animals, this task is entrusted with a set of gonad-specific 24–30 nucleotide (nt) small RNAs called PIWI-interacting RNAs (piRNAs) that associate with the PIWI clade Argonaute proteins [2]. The basic functional unit of the pathway consists of a small RNA bound to a PIWI protein, with the piRNA acting as a guide for the protein [3]. Some PIWI proteins function as small RNA-guided endonucleases (slicers), while others recruit histone or DNA methylation machineries to mediate transcriptional repression of target genomic loci [2]. Any impairment of the piRNA pathway results in de-repression of transposons and failure of germ cell development, causing infertility.
Biogenesis of piRNAs is a cytoplasmic event. Single-stranded transcripts that arise from large (50–100 kilobases) genomic regions called piRNA clusters [4] or transposon transcripts and some genic mRNAs are substrates for piRNA production. They are transcribed by RNA polymerase II [5] and exported [6] to the cytoplasm where piRNA biogenesis factors are enriched in cytoplasmic perinuclear granules called nuage [7, 8]. Primary processing is the default pathway that converts the piRNA precursors into thousands of piRNAs having a preference for a uridine at the 5′ end (U1-bias). Since piRNA precursors resemble other cellular transcripts in having features like a 5′ cap and a 3′ poly A tail [5], how they are specifically targeted to the primary processing pathway is largely unknown.
Studies conducted in the Drosophila ovarian system indicate a role for sequences within the precursor transcripts in the recruitment process. In the fly ovarian germline, presence of complementary binding sites for piRNAs in transcripts, and the resultant slicing by PIWI proteins Aubergine and Ago3 identifies it as a precursor. This enables the entry of one of the cleavage fragments into piRNA biogenesis that generates a series of phased/non-overlapping piRNAs [9–11]. In contrast, Piwi slicing is demonstrated to be not essential for piRNA biogenesis in the fly ovarian soma [12], and primary processing has to create piRNAs de novo from the precursors. Studies reveal that sequences termed piRNA-trigger sequences (PTSs) present at the 5′ end of piRNA-producing regions are necessary and sufficient for recruiting a transcript into the somatic piRNA biogenesis machinery [9, 13]. Deletion of such sequences from an endogenous precursor transcript impacts piRNA biogenesis from the locus [13]. Precise nature of these poorly conserved sequences, and how they work are currently not understood, but they are assumed to provide binding sites for piRNA biogenesis factor(s) to initiate primary piRNA processing.
In this study, we recruited piRNA biogenesis factors to a reporter transcript by artificial tethering and demonstrate its entry into primary processing using transgenic fly lines and an ovarian somatic cell culture model. Tethering of the conserved piRNA biogenesis factor Armitage (Armi) [14, 15] or the somatic piRNA biogenesis factor Yb [16, 17] to a transcript results in its identification as a piRNA precursor in the fly ovarian soma and a somatic cell culture model. This results in non-overlapping/phased conversion of the transcript into ~24 nt U1-containing primary piRNAs. A similar effect is seen when Armi is tethered to a transcript in the fly ovarian germline, with generated piRNAs entering all the three PIWI proteins present in this environment. We find that this ability to induce piRNA generation is broadly correlated with localization of the factors to cytoplasmic processing granules called nuage in germ cells [7] or Yb bodies in the soma [17, 18]. Our study reveals a strategy for generating artificial piRNAs capable of targeting any germline gene, and provides a useful tool for dissecting the molecular mechanisms of primary piRNA biogenesis.
The Drosophila female germline is a widely used model for piRNA research. The fly ovaries are organized into a series of egg chambers, each of which is composed of a single layer of somatic follicle cells that enclose the germline (nurse cells and the developing egg) (Fig 1A). While the germline expresses all three PIWI proteins (Piwi, Aubergine and Ago3), the soma is a simple system operating a primary piRNA pathway that loads Piwi. The flamenco cluster is the largest source of piRNAs in this environment, and a fragment consisting of the 1st exon of the flamenco, termed piRNA-trigger sequence (PTS), when fused to any heterologous transcript is capable of initiating piRNA biogenesis [9, 13]. We reproduced these results using a reporter consisting of the flamenco PTS placed between luciferase and LacZ sequences (Fig 1B). Expression of the reporter in the ovarian somatic cell (OSC) culture model [19] results in the directional production of piRNAs from the downstream LacZ sequences, which are loaded into Piwi (Fig 1B and S1A Fig). Negligible levels of piRNAs are produced in the absence of the PTS element. The same reporter background (two independent constructs), but carrying a perfectly complementary binding site for abundant Piwi-loaded piRNAs (instead of the PTS element) did not result in piRNA production in the OSC system (S1B Fig). Thus we can rule out any role for Piwi slicing in somatic piRNA biogenesis, as already demonstrated [12]. So we hypothesize that the PTS recruits piRNA biogenesis factors to initiate primary processing.
We tested this possibility by directly recruiting piRNA biogenesis factors to a transcript in the ovaries of transgenic flies. To this end, we replaced the PTS sequence in the above reporter with five BoxB (5BoxB) hairpins (Fig 1C). When co-expressed with λ-N peptide-fusion proteins, the BoxB/N-peptide interaction [20] will artificially tether the protein at a central location within the transcript. We first tested Armitage (Armi) which is a highly conserved RNA helicase that is essential for production of all piRNAs in flies [15, 21]. Its orthologue MOV10L1 has a similar role in mice [22–25]. Transgenic fly lines co-expressing both the BoxB reporter and NHA-Armi (with the N-peptide and an HA-tag), specifically in the fly ovarian soma [under control of the traffic jam-GAL4 driver (tj-GAL4)] were generated. Entry of the reporter into the piRNA pathway was monitored by Piwi immunoprecipitations with fly ovaries and deep sequencing analysis of associated small RNAs (Fig 1A).
Tethering of NHA-Armi triggers piRNA production from the reporter, with most of the reads originating from the BoxB site and the LacZ region downstream (Fig 1C). When HA-Armi (that is unable to tether to the reporter) is expressed, the BoxB reporter produces only low background levels of piRNAs (Fig 1C). Although the reporter sequence has no particular nucleotide bias, the generated artificial piRNAs display a strong bias for having a uridine at the 5′ end (U1-bias), a primary piRNA feature (Fig 1D). Production of piRNAs from the upstream luciferase region is also increased upon the tethering (S1D Fig) but absolute levels remain low. This is not due to any particular features of the sequence, as the same stretch is used for piRNA production in other contexts [9] and as shown below (S5A Fig). However, due to the very low levels of luciferase piRNAs triggered by Armi tethering, we limit the analysis only on piRNAs produced from LacZ region.
To study how the transcript becomes cleaved during primary processing, we calculated the distances between neighbouring piRNAs. When 5′-to-5′ end distances were plotted, we observed peaks at positions ~25, 50 and 75 nt, which correspond to multiples of the approximate length (~24 nt) of a Piwi-bound piRNA (Fig 1E). Measurement of 3′-to-5′ end distances reveals a major peak at position 1 and another one at ~25 nt (Fig 1E). These likely correspond to the distance between 3′ end of a piRNA and the 5′ end of the one immediately downstream (distance of 1 nt) or to the piRNA even further downstream (distance of ~25 nt). These observations indicate a phased/non-overlapping primary piRNA biogenesis mechanism [9–11], where the primary processing machinery moves along the transcript in a stepwise/phased manner to introduce cleavages that simultaneously create the 5′ end of a piRNA and the 3′ end of the preceding one, liberating individual ~24 nt piRNAs (Fig 1F). These phased cleavages are not always precise, but closely spaced (1 nt), giving rise to the additional 5′-to-5′ end distance peak at position 1 (Fig 1E). Note that even in the absence of tethered Armi (when co-expressing HA-Armi), the residual levels of reporter-derived piRNAs generated have a phasing signature (S1C Fig). Taken together, direct binding of Armi to a transcript in the fly ovarian somatic follicle cells identifies it as a primary piRNA precursor, leading to phased piRNA production.
Armi is a putative RNA helicase that has conserved sequence motifs essential for ATP binding and ATP hydrolysis (Fig 2A). We directly tested this activity using recombinant Drosophila Armi (Fig 2A and S2A Fig). We annealed a 5′-end labelled short single-stranded RNA (ssRNA) with a longer unlabelled complementary sequence to prepare double-stranded RNAs (dsRNAs) with either 5′ or 3′ single-strand overhangs. These RNAs were then incubated with Armi, either in the presence or absence of ATP, and reactions were resolved by 15% native polyacrylamide gel electrophoresis (PAGE). Incubations with Armi, in the presence of ATP, resulted in the appearance of a fast-migrating short ssRNA band, indicative of RNA unwinding activity (Fig 2B and S2B Fig). Interestingly, only the dsRNA with a 5′ single-stranded overhang was used by Armi as a substrate. RNA helicase activity was not observed in the absence of ATP or when the dsRNA has a 3′ single-stranded overhang. Importantly, this activity was abolished when a single amino acid mutation (E863Q) was introduced into the catalytic motif (DEAG→DQAG) of Armi (Fig 2B and S2A Fig). This indicates that Armi is a 5′→3′ RNA helicase and is consistent with the known 5′→3′ RNA helicase activity of its mouse orthologue MOV10L1 [24].
Next, we wished to examine whether the RNA helicase activity is required for tethering-driven piRNA biogenesis. We created transgenic flies co-expressing the NHA-tagged catalytic-dead ArmiDQAG mutant protein and the BoxB reporter transcript in the somatic follicle cells of fly ovaries. When tethered to the reporter, overall piRNA production was reduced (2.5-fold) compared to that driven by NHA-Armi (Fig 2C). Examination of piRNA generation across the reporter transcript indicates a dramatic reduction in piRNA levels from transcript, except for those arising from the site of tethering (BoxB sequences) (Fig 2D). A similar reduction (4-fold) in overall piRNA levels is noted when we tethered a second Armi mutant (NHA-ArmiGNT) that carries a point mutation (K729N) in the ATP binding motif (GKT→GNT) (Fig 2C). Again, piRNA levels decreased across the transcript, except from the site of tethering (Fig 2D). Albeit at reduced levels, piRNAs initiated by Armi helicase mutants display a dominant bias for having a 5′ uridine, indicating genuine primary processing (Fig 2E). In conclusion, helicase activity of Armi is essential for robust piRNA production from the tethered transcript.
In addition to Armi, other factors are shown to be essential for piRNA biogenesis in the fly ovarian soma. These include the putative RNA helicase Yb [17, 18, 26] and the Hsp90 co-chaperon Shutdown (Shu) [27–29], both of which we tested in our tethering assay using transgenic fly lines. When tethered to the reporter in the fly ovarian somatic follicle cells, Yb led to robust piRNA production from the reporter (Fig 2C). The features of the generated piRNAs mirror that produced by Armi tethering. The sequences have a prominent U1-bias (Fig 2E), and arise in absolute terms mostly from the BoxB sequences and the downstream LacZ region (Fig 2D). Furthermore, measurement of piRNA-end distances reveals that Yb binding triggers phased primary piRNA processing of the transcript (Fig 2F), as demonstrated above for Armi.
In contrast, flies co-expressing NHA-Shu with the reporter revealed only background levels of piRNAs (Fig 2C and 2D). We confirmed by Western analysis that NHA-Shutdown is indeed expressed in fly ovary lysates (S2C Fig). These results indicate that Armi and Yb, but not Shu, when individually tethered to a transcript have the ability to identify it as a primary piRNA precursor in the fly ovarian somatic follicle cells.
Most piRNA biogenesis factors are cytoplasmic, where they accumulate in perinuclear granules called nuage [7]. In the fly ovarian somatic follicle cells, this is represented by the Yb body [18, 26]. To examine the localization of the various tethered factors, we carried out anti-HA staining of fly ovaries expressing fusion proteins under control of the soma-specific tj-GAL4 driver (Fig 3). Both HA- and NHA-tagged Armi were found in 1–2 granules/cell, which also contain endogenous Yb, identifying their presence within the Yb body (Fig 3A and 3B). This was also true for NHA-Yb, which was co-localized with endogenous Armi (Fig 3C and S2D Fig). In contrast, the ArmiDQAG and ArmiGNT mutant proteins were more dispersed and accumulated in numerous (up to 10) cytoplasmic granules (Fig 3A and 3B). Although most are non-overlapping with the Yb body, some do overlap (Fig 3B). This mislocalization is not due to any impact on structural integrity of the protein, as the point mutant behaves similar to the wildtype during size-exclusion chromatography (S2A Fig). Thus, loss of RNA helicase activity is directly responsible for failure of the mutants in accumulating in the Yb bodies. Interestingly, NHA-Shu is diffusely present throughout the cytoplasm of ovarian follicle cells, with no co-localization with the Yb body (Fig 3C and S2D Fig). Thus, Armi mutants and Shu that fail to support robust tethering-initiated primary processing are found not to be co-localizing with the Yb body in the fly ovarian follicle cells. It is expected that localization in the Yb body might facilitate association with other piRNA processing factors, for example, like the biochemical association of Armi-Piwi that we demonstrate here (Fig 2D). Taken together, we propose that tethering-induced piRNA production from the reporter transcripts is likely a consequence of the reporter transcript accumulating in the Yb bodies, where it is engaged by the resident piRNA biogenesis machinery.
In the above studies, we demonstrated that recruitment of Armi or Yb to a transcript identifies it as a substrate for primary piRNA processing in the fly ovarian somatic follicle cells. To further dissect the requirements from the tethered protein and the reporter RNA for efficient piRNA processing, we made use of the OSC culture system [19], which is a model for the ovarian somatic environment.
OSC cultures were co-transfected with plasmids expressing the BoxB reporter and different NHA- or HA-fusion proteins. Cells were harvested 48-hours post-transfection and libraries were prepared with small RNAs isolated from Piwi immunoprecipitations (Fig 4A and S3A Fig). These experiments largely confirm the findings with the transgenic flies: robust piRNA production when Armi or Yb is tethered to the reporter, but not when tethered with Shu (Fig 4B and 4C and S3B and S3C Fig). Structural integrity of Armi is essential for this functionality, as tethering of the helicase domain alone is unable to trigger piRNA production (Fig 4B and S3C Fig). As shown in fly ovaries, tethering of the ArmiGNT mutant resulted in reduced levels of piRNAs, while surprisingly, the catalytic-dead mutant ArmiDQAG induced piRNA levels comparable to that seen with the wildtype Armi protein. Interestingly, tethering of Piwi itself did not result in any piRNA production, and behaved similar to tethering of LacZ, a protein unrelated to the piRNA pathway (Fig 4B and S3C Fig). Introduction of a point mutation (D537A) in Yb that is shown to abolish its RNA binding property [26], did not affect its ability to induce piRNA generation (Fig 4B and S3C Fig). This is expected, as artificial tethering to the transcript via N/BoxB system likely negates the requirement for this RNA-binding activity. Finally, we report that we do not see the phasing pattern of piRNA generation in the OSC system (S3D Fig). We have no reason to believe that processing in the OSC proceeds differently than in the fly ovarian somatic follicle cells, so it is likely that technical aspects like transfections and small RNA library quality might have influenced our ability to detect it.
Next, we probed the requirement of uridines in the reporter for tethering-driven piRNA biogenesis, as the generated primary piRNAs display a strong U1-bias (~75%). We modified part of the reporter sequence by creating two patches lacking any Us; no-U#1 and no-U#2 to prepare a U-less reporter and also prepared a U-interval reporter [11] having Us distributed at specific intervals (Fig 4D and S1 Protocols). Lack of uridines in the no-U patches resulted in reduced levels of piRNAs (Fig 4E and 4F and S4 Fig), indicating that Us are preferred, but in the absence of Us any available nucleotide is used for creating 5′ ends of piRNAs. Dramatically, while the U-interval reporter had an overall uridine composition of only ~ 4%, majority of the piRNAs generated displayed a prominent U1-bias (~52%) (Fig 4G and S4E Fig). These results align with the proposed uridine specificity (S4D Fig) of the nuclease Zucchini that generates the piRNA 5′ ends [10, 11]. It is also possible that an additional enrichment of U1-containing piRNAs could be achieved by the nucleotide preference of the MID domain of PIWI proteins [3, 30].
The above studies demonstrate that recruitment of Armi or Yb to a transcript triggers primary piRNA biogenesis that loads Piwi, which is the only PIWI clade member in the fly ovarian somatic follicle cells and the OSC culture system. In contrast, all the three PIWI proteins (Piwi, Aubergine and Ago3) are expressed in the fly germline where there is a dominant dependence on PIWI slicing to initiate piRNA biogenesis. Slicing by Aubergine (Aub) and Ago3 reciprocally loads each other with secondary piRNAs whose 5′ ends are generated by direct slicer action [4, 31]. Additionally, slicer cleavage of a target transcript by Ago3/Aub is also required to load Piwi with a series of phased primary piRNAs [9–11]. So we wished to examine whether our tethering-driven primary piRNA biogenesis might work in the fly germline.
We created transgenic flies co-expressing the reporter and different fusion proteins in the fly ovarian germline using the NGT-GAL4 driver (Fig 5A). Deep sequencing libraries were prepared with small RNAs present in isolated PIWI complexes (Piwi, Aub and Ago3). When tethered to the reporter, Armi is able to induce piRNA biogenesis that loads all three PIWI proteins, with much more sequences being loaded into Piwi than Aub or Ago3 (Fig 5B and S5 Fig). However, since different polyclonal antibodies are used for PIWI immunoprecipitations, it would be difficult to definitively conclude preferential loading into any protein. The piRNAs associating with the three PIWI proteins display the phasing pattern (Fig 5C) and strong U1-bias (Fig 5D), confirming their generation by primary processing. Interestingly, the ArmiDQAG mutant and Shu are also able to trigger piRNA generation. In contrast, the ArmiGNT mutant and the soma-specific piRNA factor Yb were inactive in the germline tethering assay (Fig 5B and S5 Fig). Shu tethering induced piRNAs only to a low level compared to that initiated by Armi, but generated piRNAs from the upstream luciferase region also (S5 Fig). Finally, we find a broad correlation between sub-cellular localization of the tethered proteins in the perinuclear nuage (labelled with endogenous Ago3) of the germline nurse cells and their ability to initiate piRNA biogenesis on reporter RNA, with the exception of ectopic Yb, which was also localized in the nuage (Fig 5E). Armi is shown to associate with Ago3, and both proteins accumulate in the nuage along with other piRNA pathway factors [32], allowing entry of the tethered transcripts into piRNA processing machinery. In conclusion, we demonstrate that nuage-localizing factors are able to channel a transcript into primary processing pathway in the fly ovarian germline.
Primary processing is the default pathway that generates piRNAs in all animal germlines. Since precursors are not unlike other cellular mRNAs or non-coding transcripts, there should be mechanisms in place to specify their entry into the processing machinery. Much is known about the secondary processing pathway operating in the fly ovarian germline, where PIWI slicing of a target transcript results in its entry into piRNA processing [9–11]. However, this depends on pre-existing piRNAs, which are suggested to be provided by maternal deposition in the egg. In contrast, primary processing has to kick-start piRNA production in the absence of pre-existing piRNAs (as in fly ovarian soma), and without the use of PIWI slicing [12]. How this is achieved is poorly understood.
Previous work implicated a role for sequences at the 5′ end of precursors termed piRNA-trigger sequences (PTSs) in recruiting the primary processing machinery in the OSC culture model [9, 13]. PTS elements are poorly defined and lack conservation, preventing their detailed study, but our work provides strong support to the hypothesis that they might provide landing sites for specific piRNA biogenesis factors.
In this study, we demonstrate that presence of a perfectly complementary site for abundant piRNAs within a reporter did not trigger piRNA biogenesis in the OSC system (S1B Fig). Instead, we show that artificial recruitment of primary biogenesis factors, Armi and Yb, to a reporter transcript is sufficient to identify it as a primary piRNA precursor (Figs 1 and 2). Among these, Armi is highly conserved and works in all the systems tested: fly ovarian soma and germline, and in the OSC cultures. Armi [14, 15, 21] and its mouse orthologue MOV10L1 [22–25] are absolutely essential for biogenesis of all piRNAs in flies and mice. In contrast, Yb is restricted to Drosophila, pointing to a non-conserved role for the protein in the fly somatic follicle cells [16]. The known interaction between Yb and Armi [16, 17] might ensure that Yb-tethered transcripts enter primary processing in the fly soma and in the OSC system (Figs 2 and 4), while lack of functionality of ectopically expressed Yb in the germline (Fig 5) could be due to competition from its germ cell specific homologues BoYb and SoYb [16].
Armi- or Yb-mediated primary processing of the tethered transcript strongly resembles that initiated by PIWI slicing in the fly germline [9–11] or in the mouse male germline [33, 34]. In both situations the transcript undergoes phased processing to generate piRNAs with a strong U1-bias, and predominantly proceeds in a 5′→3′ direction. This points to different modes of precursor identification that eventually channels the transcript into a common piRNA biogenesis machinery. We propose that tethering by nuage- or Yb body-localizing factors results in a fast-track access for the transcript to the resident piRNA biogenesis machinery in these cytoplasmic processing sites. RNA helicases are shown to recognize target RNAs in a sequence-independent manner [35], and this raises the possibility that any spurious association of piRNA biogenesis factors with other cellular RNAs can lead to their entry into the piRNA pathway, a situation that germ cells must actively prevent from happening. We believe that our tethering-mediated piRNA biogenesis strategy provides a valuable tool for further exploring the molecular mechanisms of primary piRNA processing and may even be harnessed for creation of designer small RNAs that can target any germline gene for epigenetic silencing.
Antibodies to all three Drosophila PIWI proteins used in this study were previously described [9]. These include rabbit polyclonal antibodies (two rabbits: GJKO and GJLD) to Drosophila Piwi that were generated (EMBL Protein expression and purification core facility) against an insoluble antigen (Piwi antigen: 42–178 aa) produced in E.coli as an inclusion body. Single rabbits were used to generate the antibodies to Drosophila Aub and Ago3 (Aub antigen: 1–200 aa; Ago3 antigen:1–200 aa). Immunized rabbit sera were directly used for immunoprecipitation.
For expression in the Drosophila ovarian somatic cell (OSC) cultures [19], we used the pAC5.1 vector (Life Technologies) driving expression from the fly actin promoter [9]. For expression of either HA-tag (pAC-HA) or N-HA-tag fusions (pAC-NHA), the pAC5.1 vector was further modified to add the necessary coding sequences. The HA tag is for detection of the expressed protein and the λN-peptide is for artificially tethering the fusion protein to a transcript containing BoxB sequences [20].
For creating transgenic fly lines, the coding sequences for NHA- or HA-tagged fusions of Armi, Yb or Shu, and the point mutant versions were inserted into the pUASp_attB_delK10 plasmid containing the white+ gene marker. These were used for site-specific integration (BestGene, Inc) in the Drosophila genome using the PhiC31 (ΦC31) integrase-mediated transgenesis system. Details of crosses are given in S1 Protocols.
Drosophila ovarian somatic cell (OSC; gift of Dr. M. Siomi, University of Tokyo) culture system is representative of the fly ovarian somatic follicle cells [19]. OSCs were cultured in 75 cm flasks and grown to 80% confluence. Approximately 3.5x106 cells were used for each electroporation reaction using Cell Line Nucleofector Kit V (Lonza, Cat No. VCA-1003) and were plated in 6-well plate. Further details in S1 Protocols.
For production of recombinant proteins in the insect cells the following ovary-derived cells were used: Sf21 or Sf9 from Fall Army worm Spodoptera frugiperda or High Five (Hi5) from the cabbage looper, Trichoplusia ni. Expression of desired coding sequences was carried out with the use of recombinant Baculoviruses. Either single or multiple coding sequences were integrated into the Baculovirus genome using the MultiBac protein expression system [36]. The coding sequence for Drosophila Armitage (Armi) was isolated by RT-PCR from fly ovarian total RNA, while the codon-optimized DNA sequence was commercially synthesized (Shanghai ShineGene Molecular Biotech,Inc.). Detailed purification steps in S1 Protocols.
RNA unwinding reaction was performed as described [37, 38] with some modifications. Single stranded RNA oligos were chemically synthesized (Microsynth, CH) and sequences are given in S1 Protocols. Substrates for RNA unwinding assay were prepared by annealing a 5′-endlabelled strand that was annealed with its unlabelled complementary partner. For details see S1 Protocols.
Reads were sorted into individual libraries based on the barcodes and the 3′ adapter sequences were clipped using cutadapt (DOI:http://dx.doi.org/10.14806/ej.17.1.200). Reads which are at least 15 nucleotides in length were used for subsequent analysis and the independent replicated libraries were merged together. Reads were then aligned to the desired reporter sequence using bowtie [39] allowing no mismatches. Analyses were performed as previously described [9]. See S1 Protocols for details.
|
10.1371/journal.ppat.0030131 | Phylogenetic Analysis Reveals the Global Migration of Seasonal Influenza A Viruses | The winter seasonality of influenza A virus in temperate climates is one of the most widely recognized, yet least understood, epidemiological patterns in infectious disease. Central to understanding what drives the seasonal emergence of this important human pathogen is determining what becomes of the virus during the non-epidemic summer months. Herein, we take a step towards elucidating the seasonal emergence of influenza virus by determining the evolutionary relationship between populations of influenza A virus sampled from opposite hemispheres. We conducted a phylogenetic analysis of 487 complete genomes of human influenza A/H3N2 viruses collected between 1999 and 2005 from Australia and New Zealand in the southern hemisphere, and a representative sub-sample of viral genome sequences from 413 isolates collected in New York state, United States, representing the northern hemisphere. We show that even in areas as relatively geographically isolated as New Zealand's South Island and Western Australia, global viral migration contributes significantly to the seasonal emergence of influenza A epidemics, and that this migration has no clear directional pattern. These observations run counter to suggestions that local epidemics are triggered by the climate-driven reactivation of influenza viruses that remain latent within hosts between seasons or transmit at low efficiency between seasons. However, a complete understanding of the seasonal movements of influenza A virus will require greatly expanded global surveillance, particularly of tropical regions where the virus circulates year-round, and during non-epidemic periods in temperate climate areas.
| The winter seasonality of influenza A virus in temperate climates is one of the most puzzling epidemiological patterns in infectious disease. To help resolve the issue of influenza seasonality, we studied, using viral genome sequence data, the patterns of global migration of influenza A virus, particularly between the northern and southern hemispheres. A phylogenetic analysis of approximately 900 complete genomes of the H3N2 subtype of human influenza A virus sampled from New Zealand and Australia (southern hemisphere), and New York state, United States (northern hemisphere), revealed that cross-hemisphere migration frequently occurs in both directions and involves multiple viral strains. Such global viral traffic therefore contributes significantly to the introduction of new influenza epidemics in both northern and southern hemispheres. These results also show that influenza A virus migrates afar during non-epidemic periods, rather than persisting locally at low levels during the influenza “off-season”. However, although this represents the largest and first bihemisphere study of its kind to our knowledge, the results highlight the need for sampling from tropical regions and during non-epidemic periods in temperate areas. Studies of this kind are critical to fully understand the geographical dispersal of influenza A virus and the role of climate in triggering seasonal epidemics.
| Influenza A virus is able to persistently re-infect human populations by continually evading host immunity through the rapid evolution of surface antigens (“antigenic drift”) [1]. Influenza virus epidemics strike temperate latitudes of the world each winter, from November to March in the northern hemisphere and from May to September in the southern hemisphere [2]. In the United States alone, these influenza epidemics are associated with an annual average of 36,000 human deaths [3] and 226,000 hospitalizations [4]; globally, the virus is associated with as many as half a million annual deaths [5]. While rapid antigenic change is a hallmark of influenza evolution, recent studies have failed to detect antigenic drift over an epidemic season, suggesting that important evolutionary processes may occur during non-epidemic periods, either locally or perhaps elsewhere [6–8]. However, surveillance during non-epidemic periods is not conducted routinely by the network of World Health Organization influenza reference centers [9] and, consequently, little is known about how and where the virus persists in the human population in between winter epidemics at low levels. A key question is therefore whether the virus remains locally within its host population in between epidemics, perhaps persisting within hosts in a latent state [10], or whether the virus migrates afar to other reservoirs, such as the tropics, and is later reintroduced.
Although influenza virus has long been regarded a “cold-weather” pathogen due to its marked winter epidemics in temperate zones, recent studies show that tropical regions experience significant year-round influenza virus activity [11]. In theory, such a “tropical belt” could serve as a year-round reservoir that harbors endemic populations of influenza virus that seasonally reintroduce viral isolates into temperate zones to trigger new epidemics [12,13]. Whereas population crashes at the end of seasonal epidemics create severe evolutionary bottlenecks that limit genetic diversity, tropical zones may function as permanent mixing pools for viruses from around the world. Historically, Southeast Asia has been considered a potential epicenter for emergence of pandemic viruses due to the proximity with which humans live with their domestic animals [14]. However, abundant data from these regions is currently unavailable, so the origins of influenza pandemics and epidemics remain unclear.
Given the ease and speed with which the influenza virus is thought to spread between humans, it is generally accepted that global chains of direct person-to-person transmission are sufficient to maintain the influenza virus in the human population [15]. However, a complete understanding of how the influenza virus transmits between humans is lacking [16], and whether human-to-human spread alone accounts for the seasonal emergence of epidemics has been questioned [17]. The simultaneous appearance of influenza outbreaks separated by large longitudinal distances, as well as sporadic influenza cases during summer months, suggests that the virus may instead already be “seeded” and somehow reactivated by environmental stimuli. Thus, the alternating pattern of northern and southern hemisphere bi-epidemics could, in principle, also result from opposite climatic forces independently reactivating viral activity in these two hemispheres at alternating six-month intervals. Hence, instead of continually migrating across the equator, separate viral populations could persist locally in an asymptomatic latent state over the summer months until climatic stimuli sufficiently increase host susceptibility and/or viral transmissibility to induce another epidemic. However, hypotheses of how climatic change may directly or indirectly influence viral activity and/or host susceptibility remain largely untested [18,19].
Crucially, some theories for influenza seasonality produce testable phylogenetic hypotheses. On one hand, if influenza A virus persists locally over the summer in a latent state, then isolates sampled over multiple seasons from a single locality would cluster together on a phylogenetic tree, separate from isolates from other geographic regions (Figure 1). Alternatively, if the virus did not evolve in situ between epidemic seasons, but rather traveled globally between epidemics, then the resulting phylogeny would show extensive intermixing of isolates from different localities. To determine whether influenza virus migrates away between the northern and southern hemispheres during non-epidemic summer months, or remains there latently, we conducted an extensive phylogenetic analysis of 399 whole-genome A/H3N2 influenza A viruses sampled from New Zealand (most commonly Canterbury, South Island) from 2000 to 2005 (six seasons), 88 viral genome sequences from Australia (most commonly Western Australia) from 1999 to 2005 (seven seasons), along with a carefully selected sub-sample of 52 isolates that are representative of the clades present in a larger sample of 413 viruses from New York state, United States, collected between 1998 and 2005 and analyzed previously [7]. Given the relative geographic isolation and low population densities of New Zealand's South Island and Western Australia, as well as sampling limitations, this analysis provides a conservative estimate of the extent of cross-hemisphere migration occurring during this time period.
Populations of A/H3N2 influenza virus in Australia and New Zealand from 1999 to 2005 exhibit extensive genetic diversity across the entire genome (Figures 2–4), comparable to the diversity observed previously in New York state [7] (in all phylogenies, clades from Australia are shaded blue, those from New Zealand in green, those from New York state in orange, and global isolates in pink). In particular, multiple viral clades co-circulate during each influenza season in New Zealand and Australia, defined as clusters of isolates with high bootstrap support (>70%), or which are separated by exceptionally long branches (Table 1). As with the New York state data, these viruses were collected in the context of seasonal surveillance efforts in Australia and New Zealand and therefore likely provide a representative sample of the overall genetic composition of the viral populations.
On the neuraminidase (NA) tree (Figure 2), viruses from New Zealand and Australia fall into at least 15 distinct clades, some of which appear in multiple seasons: three clades circulated in the 1999 season (clades v, vi, and ix; when data was only available from Australia), five in 2000 (clades i, ii, iii, iv, viii), four in 2001 (clades i, iv, v, g), one in 2002 (clade a), one in 2003 (clade b; when a new reassortant virus predominated), two in 2004 (clades A and c), and three in 2005 (clades A, B, E). Clade lettering reflects the three main sections of the phylogeny, based on topology and time: clades A–E contain viral isolates from 2003 to 2005 that fall within section I of the tree (large pink rectangle in upper portion of tree); clades a–e contain isolates from 2001–2003 that fall within section II (large light green rectangle in middle portion of tree); clades i–xiii contain isolates from 1998 to 2000 that fall outside of sections I and II. The hemagglutinin (HA) tree (Figure 3) contains at least 15 clades from New Zealand and Australia: two circulated in 1999 (clades v and x), four in 2000 (clades i, ii, iii, iv), four in 2001 (clades g, i, iv, xiii), two in 2002 (clades h and j), one in 2003 (clade b), two in 2004 (clades A and c), and three in 2005 (clades A, B, E) (Figure 3). Finally, the phylogeny of the concatenated six non-surface glycoprotein segments (PB2, PB1, PA, NP, M1, NS1) contains the largest number of clades, presumably because this larger data set (9,636 bp) provides the greatest resolution. Southern hemisphere isolates form at least 18 clades on this phylogeny: four were present in 1999 (clades v, vi, x, xii), three in 2000 (clades i/ii, iii, iv), four in 2001 (clades g, i/ii, iv, xiii), three in 2002 (clades a, k, l), one in 2003 (clade b), three in 2004 (clades A, F, c), and three in 2005 (clades A, B, E) (Figure 4).
Differences in the number of clades among segments may also be indicative of reassortment, especially involving the HA gene, as previously demonstrated in New York state [7]. Indeed, several major reassortment events are immediately evident from topological incongruities among these three phylogenies. On the HA tree, clades b, c, and e fall in section I along with isolates from 2004 and 2005, while these clades fall in section II amidst 2002 and 2003 isolates in the NA and concatenated six non-surface glycoprotein phylogenies (Figures 2–4). Clade c also falls in section I on the PB2 phylogeny, showing a similar reassortment pattern as HA. However, aside from this lone reassortment event, the phylogenies of the six non-surface glycoprotein segments are very similar, enabling us to study them as a single concatenated entity (phylogenetic trees for individual segments are provided as Figures S1–S6).
Isolates from the 52 representative New York state genomes are clearly interspersed with southern hemisphere isolates throughout our phylogenetic trees (Figures 2–4; Table 1), indicating that these populations regularly intermix as a result of cross-hemisphere migration. However, patterns of viral intermixing are both variable and complex, as clades from all three phylogenies contain an array of different combinations of viral populations from New Zealand, Australia, and/or New York state. For example, the phylogeny of the concatenated six non-surface glycoproteins contains a relatively even mix of mono-hemisphere and bi-hemisphere clades (Table 1), with five New York state–only clades, seven southern hemisphere–only clades, six clades that contain New York state isolates and isolates from only one of the southern hemisphere countries, and three clades that contain isolates from all three countries. Thus, on an annual basis some, but not all, viral populations mix with other populations from the same and/or opposite hemisphere, and this number is likely to increase with additional sampling. Indeed, a majority (nine) of the 16 clades containing southern hemisphere isolates also contained northern hemisphere isolates, suggesting widespread viral traffic across the equator (Table 1). Migration between Australia and New Zealand is also extensive, as almost all clades containing isolates from New Zealand also contained Australian isolates, and vice versa, except for the 1999 season, for which no New Zealand isolates were available. Further, these figures are likely to be underestimates, as mono-national and mono-hemisphere clades may be at frequencies too low to be detected in the genome collections currently available. Alternatively, clades could also have originated in areas not sampled in our study, such as tropical regions, where influenza viruses typically circulate year-round [12].
Strikingly, even for those viral clades that do not exhibit cross-hemisphere migration, there is very little evidence for in situ evolution within specific localities. For example, in no case on any phylogeny are clades of A/H3N2 from New York state directly linked over multiple seasons. Rather, New Zealand and Australia viruses are always interspersed among New York state clades from different seasons, indicating that they are not evolving in geographic isolation across seasons. Most clades from New Zealand and Australia show equivalent patterns of discontinuous evolution, with very few clusters of southern hemisphere clades that are not separated by New York state viruses (although in situ evolution cannot be ruled out between a few 2004 and 2005 clades without additional sampling). Thus, even in relatively isolated areas of New Zealand and Australia, viruses do not regularly evolve in geographic isolation. Rather, evolution appears to be shaped by frequent cross-hemisphere migration and recurrent reintroduction.
Importantly, our phylogenetic analysis suggests that seasonal migration occurs from the northern hemisphere to the southern hemisphere, as well as south-to-north. Although inferring the direction of viral migration is in part dependent on sample composition, definitive evidence of a migration event are clades containing a single population of northern hemisphere viruses and a single population of southern hemisphere viruses supported by a high level of bootstrap support (>70%).
Because winter influenza seasons alternate by six-month intervals between the northern and southern hemispheres, it was also possible to determine, within the confines of sampling, the timescale and hence direction of cross-hemisphere migration. The inferred directionality of 11 cases of definitive cross-hemisphere migration evident on the HA, NA, and non-surface glycoprotein phylogenies are summarized in Table 2. Of these, all but two involve the concatenated non-surface glycoproteins, which provide a more reliable phylogeny as previously described. Eight of these migration events occur in a north-to-south direction, versus three in a south-to-north direction, suggesting that viruses may migrate more frequently from New York state to the southern hemisphere than in the opposite direction, although this will need to be confirmed with larger sample sizes. For example, on the concatenated gene tree, viral isolates from the 2003–2004 season in New York state form a single well-supported phylogenetic clade (clade b, 100% bootstrap support) with viruses from 2003 from New Zealand and Australia (Figure 4). Since the 2003 New Zealand and Australia viruses predate the 2003–2004 New York state viruses (i.e., the northern hemisphere winter), we infer that the lineage that gave rise to these southern hemisphere viruses migrated northward to infect New York state between the 2003 southern hemisphere winter (May to October) and the 2003–2004 winter in New York state.
It is also notable that cross-hemisphere migration does not follow any clear pattern. In addition to occurring in both north-to-south and south-to-north directions, migration events also appear to involve minor clades as frequently as major clades, assuming that our study sample is generally representative of the viral population structure (Table 2). Furthermore, the populations of southern hemisphere viruses that migrate northward are a mix of compositions, including some isolates only from New Zealand, some only from Australia, and populations with a mix of isolates from both countries (Tables 1 and 2). Finally, two clades contain global strains, but because the dates of these global isolates are not recorded, it is impossible to accurately determine the direction of migration events.
The 11 cases presented in Table 2 represent only the strongest examples of cross-hemisphere migration, using the strictest criteria to infer migration events from the phylogenetic data. Relaxing this stringency allows for the possibility of greater bi-directional cross-hemisphere migration, especially involving clades containing more than one viral population from the southern hemisphere. Although the direction of migration is less certain when populations of viruses from three geographical regions are present, the relative frequency of migration observed under the more stringent criteria suggests that cross-hemisphere migration likely operates in these cases as well. Finally, while it is likely that a more intensive sampling regime will increase clade diversity, and in doing so affect the inference of the direction of migration, the complexity of the patterns observed strongly argues for frequent bi-directional migration.
Our large-scale phylogenetic analysis of A/H3N2 influenza virus populations from opposite geographic hemispheres provides evidence for regular bi-directional cross-hemisphere viral migration between seasons, even among localities as distantly separated as New York state and Australia and as relatively geographically isolated as New Zealand's South Island. Multiple genetic variants of influenza virus co-circulate each season, even in geographically remote areas, and many of these viral clades are more closely related to isolates from the opposite hemisphere than to isolates from either the previous or following season in the same location. Thus, viral populations do not appear to “over-summer” locally, where they would evolve in situ and give rise to the next season's epidemic. Rather, cycles of viral migration and recurrent introduction have clearly played a significant role in generating the genetic diversity that characterizes influenza A virus in both hemispheres. Importantly, given the sample composition of our sequence data set, the extent of cross-hemisphere migration observed here undoubtedly represents a conservative estimate. Hence, including data from more populated areas could only reveal more instances of cross-hemisphere migration.
In addition, our finding that the virus migrates globally between epidemics and is reintroduced is clearly compatible with tropical regions, including Southeast Asia, playing a key role in the genesis of new clades and the global spread of these novel influenza virus variants. Thus, while limitations in global genome sampling necessarily means that the current study is directed toward testing hypotheses of viral migration versus latency, equivalent data from tropical regions would undoubtedly enable us to conduct a more refined analysis of global migration patterns and their determinants. Specifically, if tropical regions serve as year-long influenza reservoirs, we would expect to observe phylogenies in which tropical isolates display the greatest genetic diversity and are positioned basal to viruses sampled from temperate regions. Consequently, complete genome sampling from tropical regions where influenza viruses circulate year-round, including a record of the precise date of collection, is of key importance for understanding the global epidemiology of the influenza virus.
Notably, the viral migration we observe does not appear to follow any clear pattern, but rather occurs in all directions, involves all genes, and involves clades of all sizes and geographic compositions. This argues against a role of immune selection in determining which viral clades are able to migrate among localities, although it does not preclude a role for natural selection as the sieve that determines which clades are able to survive in specific host populations. Similarly, the observation that migration patterns vary to some extent among the HA, NA, and concatenated non-surface glycoproteins must reflect the effect of widespread genomic reassortment [7,20]. Frequent reassortment complicates the analysis of migration patterns, as individual viruses can carry genomic segments with differing phylogenetic, and hence geographical, histories. Consequently, the analysis of migration patterns based on single gene segments may paint a misleading picture.
Although the transmission of the influenza virus through population movements has been studied extensively, particularly for the spread of pandemic isolates across the globe by air travel [21,22], neither the routes nor the mechanisms of the virus's geographical spread have been fully resolved. Several recent studies have used empirical data to investigate the role of population movements on the spatial diffusion of seasonal epidemics, including an intricate analysis of the regional spread of influenza epidemics across the United States, which was strongly correlated with adult workflow movements [23]. A previous epidemiological study comparing the synchronicity with respect to timing of influenza epidemics between the United States, France, and Australia suggested that the inter-hemispheric circulation of epidemics follows an irregular pathway, with recurrent changes in the leading hemisphere [24], in accordance with the phylogenetic analysis presented here. More fine-scaled analyses of discrete viral populations have shown that frequent introduction of “foreign” viruses significantly impacts the viral population structure and geographic spread at local levels. For example, the rapid timescale of global mixing of influenza drowns out any impact of local heterogeneities on the spread of the epidemics through France [25]. Similarly, the seasonal importation of multiple global isolates appears to be a greater contributor to the genetic diversity of the influenza virus population in New York state from 1997 to 2005 than local in situ evolution [7]. While our findings confirm that human population movements play a role in introducing new viral variants at the start of an epidemic, some aspect of climate is clearly of importance in triggering epidemics. Additional research is required to define how human susceptibility to infection and viral transmissibility fluctuate under varying climate conditions and why influenza virus is absent in summer in temperate climates but exists year-round in tropical zones.
Although the underlying cause of the seasonality of the influenza virus remains uncertain, even in reservoir avian species [26], our findings illustrate the critical importance of expanding surveillance to elucidate the geographical movements and evolution of this virus throughout its entire annual cycle. The traditional focus on epidemic influenza may detract from the equally important epidemiological question of why influenza A virus does not circulate in humans for so many months of the year in temperate areas, especially given its apparent ability to infect humans in tropical areas year-round. Attempts to predict, model, or contain the spread of the influenza virus require a unified understanding of how the virus's spatial-temporal dynamics, antigenic evolution, and seasonal emergence interrelate [27]. Although this study is limited to only the three countries for which we have extensive data, our analysis exemplifies the capacity of phylogenetic analysis to elucidate challenging epidemiological questions by providing a level of finer resolution.
All influenza A (H3N2) virus complete genome sequence data were collected from the National Institute of Allergy and Infectious Disease's Influenza Genome Sequencing Project (http://www.niaid.nih.gov/dmid/genomes/mscs/influenza.htm) for the period 1998–2005 [28]. Influenza A/H3N2 viruses were sampled by a network of participating general practitioners. Viruses from all 11 regions in New York state were collected by the Virus Reference and Surveillance Laboratory at the Wadsworth Center, New York State Department of Health. Influenza viruses from both the North and South Islands of New Zealand were collected by Canterbury Health Laboratories in Christchurch, New Zealand. In Australia, viruses from Western Australia were collected by PathWest Laboratory Medicine, Western Australia; viruses from New South Wales were collected by the Prince of Wales Hospital, New South Wales; viruses from South Australia were collected by the Institute of Medical and Veterinary Sciences, South Australia; and viruses from Queensland were collected by the Queensland Health Science Services, Queensland. All sequence data were downloaded from the National Center for Biotechnology Information (NCBI) Influenza Virus Resource (http://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html). For Australia, 88 genome sequences from the 1999–2005 seasons were compiled, while for New Zealand, 399 genome sequences A/H3N2 sequences from the 2000–2005 seasons were collected. For New York state, United States, 52 phylogenetically representative genome sequences from the 1998–1999 to 2004–2005 seasons were carefully selected from a larger data set of 413 sequences from 1997–2005 analyzed previously [7] (excluding 2000–2001, for which few H3N2 sequences were available in an H1N1-dominant season). GenBank accession numbers for all sequences used in this study are listed in Table S1.
Sequence alignments were manually constructed for the major coding regions of each of the eight genomic segments. In addition to alignments for the HA (1,698 bp) and NA (1,407 bp), an alignment of the concatenated six non-surface glycoproteins segments (PB2, PB1, PA, NP, M1, NS1) was also compiled (9,636 bp), as these are expected to evolve differently from the HA and NA surface glycoproteins. Because the minor M2 and NS2 proteins are involved in overlapping reading frames, they were excluded from the analysis.
Initial phylogenetic trees were inferred for sequences of the HA, NA, and concatenated non-surface glycoproteins from New York state, New Zealand, and Australia under the HKY85 (Hasegawa-Kishino-Yano) model of nucleotide substitution using the Neighbor-Joining (NJ) method available in PAUP* [29]. Due to the very large size of all data sets, and the provisional nature of the analysis, the nearest-neighbor-interchange branch-swapping method was employed in this case. To assess the robustness of individual nodes on these phylogenetic trees, we performed a bootstrap resampling analysis (1,000 replications) using the NJ method. From these three starting phylogenetic trees, “major” clades (which contained the majority of isolates from a season) and “minor” clades of genetically related viruses were identified by exceptionally long branch lengths and/or high bootstrap values (>70%). A subset of sequences for the concatenated non-surface glycoproteins was constructed with 51 sequences from New Zealand, 45 sequences from Australia, and 52 from New York state (see above) for a total data set of 148 sequences. For the HA gene, these 148 isolates were placed in a more global context with the addition of 13 genetically unique HA sequences sampled from this time period available on GenBank, to produce a total of 161 HA sequences. Likewise, 22 global NA sequences were combined with the original 148 from New York state, New Zealand, and Australia for a total of 170 NA sequences. Maximum likelihood (ML) phylogenetic trees were then inferred using the PAUP* package [29] for these three new data sets: 161 HA sequences, 170 NA sequences, 148 concatenated sequences. ML trees were also inferred for each of the six non-surface glycoprotein segments to ensure that all exhibit similar tree topologies (Figures S1–S6). In each case, the best-fit model of nucleotide substitution was identified by MODELTEST [30] as the general reversible GTR+I+Γ4 model, with the frequency of each substitution type, proportion of invariant sites (I), and the gamma distribution (Γ) of among-site rate variation with four rate categories (Γ4) estimated from the empirical data. In all cases tree bisection-reconnection branch-swapping was utilized to determine the optimal tree. Finally, a bootstrap resampling process (1,000 replications) using the NJ method was used to assess the robustness of individual nodes on the phylogeny, incorporating the ML substitution model.
The analysis of the frequency and directionality of migration was undertaken through a visual inspection of the topological position of individual clades on each tree and in consideration of their time of sampling. Although more quantitative methods for determining migration patterns from gene sequence data have been established, particularly those based on parsimony reconstructions of changes in character state (i.e., geographical locality) [31], these were considered inappropriate for the current study because they ignore the temporal structure of the influenza virus genome sequence data. Specifically, we reasonably assume that older sampled clades give rise to younger sampled clades if they fall basal to them on phylogenetic trees. |
10.1371/journal.pntd.0006358 | Report of a series of 82 cases of Buruli ulcer from Nigeria treated in Benin, from 2006 to 2016 | Nigeria is one of the countries endemic for Buruli ulcer (BU) in West Africa but did not have a control programme until recently. As a result, BU patients often access treatment services in neighbouring Benin where dedicated health facilities have been established to provide treatment free of charge for BU patients. This study aimed to describe the epidemiological, clinical, biological and therapeutic characteristics of cases from Nigeria treated in three of the four treatment centers in Benin.
A series of 82 BU cases from Nigeria were treated in three centres in Benin during 2006–2016 and are retrospectively described. The majority of these patients came from Ogun and Lagos States which border Benin. Most of the cases were diagnosed with ulcerative lesions (80.5%) and WHO category III lesions (82.9%); 97.5% were healed after a median hospital stay of 46 days (interquartile range [IQR]: 32–176 days).
This report adds to the epidemiological understanding of BU in Nigeria in the hope that the programme will intensify efforts aimed at early case detection and treatment.
| Buruli ulcer (BU) is a neglected tropical disease that mainly affects the skin. The disease results from infection with Mycobacterium ulcerans, an environmental bacterium. In Benin, the BU treatment centres usually receive patients from Nigeria. In 2014, a study from one of the treatment centres (CDTUB, Pobe) which borders south-western Nigeria reported on a cohort of 127 PCR-confirmed cases between 2005 and 2013. We describe the epidemiological, clinical, biological and therapeutic characteristics of BU cases from Nigeria treated in the three other CDTUBs.
| Buruli ulcer (BU) is a neglected tropical disease that mainly affects the skin. The disease results from infection with Mycobacterium ulcerans, an environmental bacterium. BU is found in often swampy and humid areas. The mode of transmission remains obscure to this day, although several hypotheses have been proposed. Many authors have discussed potential reservoirs as well as vectors and transmission mechanisms that vary from region to region depending on the epidemiological, social and local environmental context. Direct human to human transmission of M. ulcerans is a rare possibility [1]. The main hypothesis is that the surface of the patient’s skin was contaminated by bacteria from an environmental source (e.g. swamps) and introduced into the skin by trauma. It is assumed that insects (aquatic bugs and mosquitoes) are the host and vector of M. ulcerans. Several experimental and environmental studies have demonstrated the implication of aquatic bugs in transmission of the disease [2–4]. DNA of M. ulcerans was detected in mosquitoes collected in Australia but a field study conducted in Benin suggested that mosquitoes do not play a central role in the ecology and transmission of M. ulcerans [5]. Fish has also been identified as a passive reservoir of M. ulcerans but generally not responsible for direct transmission of the disease [6]. Acanthamoeba species have also been identified as natural hosts of M. ulcerans and have been suggested as responsible for transmission of the disease [7]. No definitive conclusion has yet been drawn about how the disease is transmitted. The World Health Organization (WHO) has classified BU as a neglected tropical disease [8–11]. BU is the third most common mycobacterial infection in the world among immunocompetent individuals after tuberculosis and leprosy [12]. BU is characterized by a chronic necrosis of subcutaneous tissues, ranging from a simple nodule to a large cutaneous ulceration. Sometimes the bone is affected and the resulting damage can impair the functional mobility of the affected limb. Without early and effective treatment, the disease can progress and cause cosmetic complications and sequelae or functional limitations [13] with attendant stigma and social problems [14,15]. It can even lead to limb amputation.
WHO classifies BU lesions into three categories according to severity [16,17]. Category I lesions are single small lesions (e.g. nodules, papules, plaques and ulcers < 5 cm in diameter). Category II lesions consist of non-ulcerative or ulcerative plaques, oedematous forms and single large ulcerative lesions of 5–15 cm in cross-sectional diameter, while lesions in the head and neck regions and the face, disseminated and mixed forms including osteomyelitis, and extensive lesions of more than 15 cm are considered as Category III.
BU mostly affects poor people in rural areas with limited access to health care [11,18–20]. Children aged < 15 years are most commonly affected by the disease [21]. Worldwide, BU has been reported in > 33 countries, mostly within the tropical areas [22]. The majority of BU cases occur in Africa; however, cases have been reported in Australia, French Guiana, Peru and Papua New Guinea [23].
Recognizing that BU constitutes an emerging public health threat, in 1998 WHO established the Global Buruli Ulcer Initiative (GBUI) to coordinate control and research activities worldwide [9,18,22–24].
Up until 2004, the only curative treatment for BU was surgery, which consisted of wide excision to remove all infected tissues including some of the adjacent healthy tissues. Large lesions require skin grafting [18,21,23]. Scientific studies have shown the effectiveness of using different combinations of antibiotics to treat BU [25–27]. The standard treatment since 2004 has been the combination of rifampicin and streptomycin [16]. WHO has issued a provisional recommendation to use the new combination full oral therapy [28]. The introduction of antibiotic therapy has reduced the number of surgical procedures and recurrence rates. Almost all Category I and some Category II lesions can be cured without surgery [29].
In Benin, the BU control programme is known as the Programme National de Lutte contre la Lèpre et l’UB (PNLLUB). It coordinates BU control activities through four care facilities known as Centres de Dépistage et de Traitement de l’Ulcère de Buruli (CDTUB), located in the southern departments where BU is endemic [30].
During the course of their activities, the CDTUBs also receive patients from Nigeria. BU cases were officially reported in Nigeria in 1967 [31] and in 1976 [32] in different Nigerian states. Between 1998 and 2000, BU cases at the Leprosy and Tuberculosis Hospital in Moniaya-Ogoja were confirmed by the Institute of Tropical Medicine, Antwerp, Belgium. In 2006, the Nigerian authorities, in collaboration with a team from Benin and WHO, conducted an assessment of the BU situation in order to identify the endemic areas in Nigeria. The assessment covered only 5 states and therefore could not identify all the endemic regions as originally planned [33]. That was the first time in 2006 when Nigeria notified 9 BU cases to WHO. During 2009–2016, Nigeria reported 511 cases to WHO with increasing numbers of cases each year [34]. BU is known to be endemic in the south of Nigeria mainly in states such as Akwa Ibom, Anambra, Benue, Cross River, Ebonyi, Enugu, Ogun and Oyo [33,35].
Ogun State is divided by two drainage basins, the Yewa and the Ogun rivers, which discharge in separate lagoons. South-west Nigeria is characterized by tropical rainforest, similar to the environments where BU occurs in endemic areas of West Africa [35,36].
In order to establish an effective BU control system in Nigeria, it is imperative that all endemic areas are identified; hence the necessity of providing data on the states where Nigerian BU patients treated in Benin have come from. In 2014, a study from CDTUB in Pobè, Benin reported 127 PCR-confirmed cases of Nigerian BU patients treated in the facility [35]. Pobè is a town on the border with south-western Nigeria, making it the first facility of contact by patients from Nigeria. The objective of our study therefore is to describe the epidemiological, clinical, biological and therapeutic characteristics of BU cases from Nigeria treated in the other three CDTUBs and contribute data to the health authorities in that country.
This is a retrospective descriptive study of a series of 82 BU cases from Nigeria who were treated in three of the four CDTUBs, namely the CDTUB of Allada, the CDTUB of Lalo and the Centre Nutritionel et Sanitaire Gbemonten (CNSG) of Zagnanado from 1 January 2006 to 31 December 2016.
This was a comprehensive sampling. All the cases from Nigeria, who were clinically suspect for BU and treated in the CDTUBs of Allada, Lalo and Zagnanado during the study period and for whom data were available, were included in this study.
For each patient the following information was collected from their medical records and analysed:
The data on the BU cases from Nigeria treated in the three CDTUBs were taken from the PNLLUB’s Register BU 01, which is used to collect standard information for each BU patient. The data were supplemented with additional information from the patients’ medical records kept in each CDTUB. The complete data were recorded with the software Microsoft Excel 2010 and then analysed with the statistical software IBM SPSS Statistics version 20. We only performed a descriptive analysis of the epidemiological, clinical, biological and therapeutic variables of the cases. The maps were drawn from free-access shapefiles obtained from DIVA-GIS (http://www.diva-gis.org/)with QGIS 1.8.0 and ArcView 3.2 software.
This retrospective study was conducted after the approval and with the authorization of the Ministry of Health of Benin. The case data in the PNLLUB database used by the authors for this study were anonymized.
A total of 82 new patients from Nigeria suspected of having BU were treated in the three Beninese CDTUB from 1 January 2006 to 31 December 2016, with an annual average of 7 patients. Table 1 shows details of the socio-demographic, epidemiological characteristics and states of origin of the patients. Of the 82 patients, 45 (54.9%) were male. The median age of the patients was 20 years (IQR: 13.5–42.5 years). The State of residence in Nigeria was available for 66 patients, of whom 39 (59.1%) were from Ogun State; 25 (37.9%) from Lagos State and 2 (3.0%) from Oyo State. Some 55 patients (67.1%) were treated in the CDTUB of Zagnanado; 15 (18.3%) in the CDTUB of Lalo and 12 (14.6%) in the CDTUB of Allada. Patients from Lagos State were mostly treated in the CDTUB of Zagnanado while those from Ogun State were equally treated in the three centres (Fig 1). The person who referred the patient to the CDTUB was specified for 79 patients; 62 (78.5%) were referred by former BU patients from Nigeria who had been treated in Benin. The other patients were either referred by health agents or by family members. The median delay before seeking medical assistance was 203 days (IQR: 87.5–1638).
Clinically, 66 (80.5%) patients had ulcerative lesions; 11 (13.4%) had nonulcerative lesions (plaque, nodule, oedema) and 5 (6.1%) had osteomyelitis. Of the 82 patients, 53 (64.6%) had lesions on their lower limbs; 22 (26.8%) had lesions on their upper limbs; 3 (3.6%) had lesions on other parts of their body (abdomen, back, head/neck); and 4 (4.9%) had lesions on multiple parts of their body. Some 68 patients (82.9%) had Category III lesions (multiple lesions or lesions > 15 cm in diameter) (Fig 2), 13 (15.9%) had Category II lesions (lesions 5–15 cm in diameter or on their faces/breast/genitalia); only one patient (1.2%) had a Category I lesion (lesion < 5 cm in diameter). For 78 patients, the medical records of 78 patients mentioned whether or not the movements of the affected part were limited at the time of diagnosis; 24 (30.8%) had restricted joint movements.
Samples were taken from 12 patients (100.0%) in Allada, 12 patients (80.0%) in Lalo and 23 patients (41.8%) in Zagnanado giving a total of 47 out of the 82 patients (57.3%) for PCR confirmation; 36 out of 47 tested positive.
All the 82 patients received treatment free of charge including specific antibiotic therapy of combined rifampicin with streptomycin for 8 weeks as recommended by WHO as well as surgery and physiotherapy as required. However, 80 patients (97.6%) received surgery. The median length of hospital stay was 46 days (IQR: 32–176 days). The length of hospital stay was relatively longer for the patients treated in the CDTUB of Allada with a median of 197 days (IQR: 184-318) and the CDTUB Lalo with a median of 256 days (IQR: 170–340), while the median length of hospital stay was 35 days (IQR: 27–48) in Zagnanado. Some 78 patients (97.5%) were healed among whom two patients (2.4%) healed without surgery; one patient (1.2%) died during his hospital stay and one patient (1.2%) was lost to follow up. Treatment outcome data were missed for two patients.
The epidemiological, clinical, biological and therapeutic characteristics of the patients are summarized in Table 1.
The objective of our study was to describe the BU cases from Nigeria treated in CDTUBs of Allada, Lalo and Zagnanado, to supplement the previous data published from CDTUB, Pobè in 2015 [35]. Unlike other BU endemic countries where control programmes were started after the International Conference on Buruli ulcer Control and Research (Yamoussoukro, Côte d’Ivoire, 1998) [24], Nigeria’s BU control programme is still in its infancy and does not yet cover all the endemic areas of the country [37]. As a result, BU patients in certain areas of Nigeria use the BU care facilities in neighbouring countries such as Benin and Cameroon [35].
From 2006 to 2016, 82 patients from Nigeria who were clinically suspected for BU were treated free of charge in the CDTUBs of Allada, Lalo and Zagnanado in southern Benin. This number is smaller than the 127 PCR-confirmed BU cases from Nigeria who were treated in the CDTUB Pobè from 2005 to 2013. [35]. Pobè is a town in Benin that borders Nigeria, making it geographically accessible to patients from that country. This explains why the majority of BU cases from Nigeria are treated in Pobè. CDTUB Zagnanado is the oldest BU facility and, after Pobè, the closest CDTUB to Nigeria in Benin. This could explain why most of the cases described in our study were treated in Zagnanado.
Although known as a BU endemic country along with other African countries, it was only in 2009 that Nigeria started regularly reporting BU cases to WHO. From 2009 to 2016, 511 BU cases were reported by Nigeria [34]. However, this figure did not represent the actual incidence of the disease in Nigeria during that period. Together with the publication from Pobè, a total of 209 cases have been documented in Benin.
More than half of our BU patients are aged > 15 years, contrary to the literature which describes children aged < 15 years as being the most affected by the disease and as usually accounting for more than half of the cases reported in Africa [18,19,38]. In this study about Nigeria cases, less than half of the patients (41.7%) were aged < 15 years [39]. This result is similar to that observed by other authors in Nigeria in 2016 [40–42]. It cannot, however, be interpreted as BU mainly affecting older people in Nigeria since the opposite result was observed from the previous study in Pobè among the PCR-confirmed BU cases from Nigeria [35].
The majority of the patients in our study came from Ogun and Lagos, two states in the south-west of Nigeria that border the south-east of Benin. Several authors have also described BU cases in these states, thus confirming that BU may be highly endemic in these regions [33,40–42]. However, Ogun and Lagos states do not have any specialized BU care facilities [33,35,37]. It is therefore essential to develop BU control mechanisms at the local level in these two states to respond to this public health threat. Moreover, an assessment should be conducted in Oyo State in order to understand their epidemiological BU situation.
Contrary to what was observed in the Pobè study [35], there were more male patients in our sample, although gender difference among BU patients has not been described in past literature [18,19]. Here, this difference would be attributable to a random effect or to chance.
The median delay before seeking medical assistance was 203 days or approximately 8 months. This long delay prior to seeking medical attention was also described in the previous study in Pobè in 2015: in their study 24% of patients sought medical help after one year [35]. Another publication by authors from Nigeria in 2016 also described a relatively long delay (median of 16 weeks) prior to seeking medical treatment in a prospective study conducted between May 2014 and September 2015 in 4 BU endemic Nigerian states [39]. Several factors could be the root of cause of the patients’ long delay in seeking medical help. Some studies have shown that geographical inaccessibility to healthcare services is responsible for late recourse to medical care [43,44].
From a clinical perspective, our patients were mostly those with ulcerative lesions and Category III lesions (> 15 cm in diameter). The same results were obtained in the Pobè study [35] as well as by other authors who studied BU cases in Nigeria [38,40]. This could be explained by the patients’ long delay before seeking medical attention, by which time the lesions have had sufficient time to extend. This falls short of the objectives set by WHO, which called upon countries endemic for BU to conduct community actions in order to reduce the proportion of ulcerative cases to < 60% and of Category III cases to < 30% [45]. As the biological confirmation was not systematic at the CDTUB in Zagnanado, the majority of patients were not sampled. Because this facility has lengthy experience in managing the disease by the same team, patients are sampled only in case of doubt about their clinical diagnosis [46], However, PCR confirmation of all suspected BU cases is recommended in order to limit the inappropriate use of specific antibiotic and to ensure standardization of surveillance data across all treatment centres.
The majority of our patients were referred by former BU patients from Nigeria who had been treated in Benin. This confirms the finding that former patients are as effective as community relays, as described previously [47,48], and could therefore be useful in health education and active searches for BU cases in their respective communities.
Unlike the CDTUB of Zagnanado where the median length of hospital stay is 35 days, the lengths of hospital stay were relatively longer in the CDTUBs of Allada with a median of 197 days and Lalo with a median of 256 days. This difference is explained by the fact that, in the CDTUB of Allada and Lalo, patients benefitted from two full months of antibiotic treatment before any surgical procedure. This is the opposite in Zagnanado. The lengthy hospital stay observed in our study, particularly in Allada and Lalo, is linked to the size of the lesions, which are for the most part Category III lesions. Sarfo et al. observed the same connection in 2010 [49]. The late recourse to medical care is another factor that can explain the lengthy hospital stay. In a study in Ghana, a significant association between the length of hospital stay and the size of the lesion was also found [50].
In addition to local BU cases, the CDTUB of Benin receive BU cases from Nigeria, most of whom are advanced Category III lesions whose care requires more time as well as more material and financial resources and have a socioeconomic impact on both the patients and their caregivers. However, all the treatment was provided for free. These patients mainly come from Nigerian states that border Benin and are referred by former patients who received care from one of the CDTUBs. It is therefore important that BU control activities be intensified in these different states in order to detect cases early and reduce the severity of the disease.
|
10.1371/journal.pgen.1003667 | ATM Release at Resected Double-Strand Breaks Provides Heterochromatin Reconstitution to Facilitate Homologous Recombination | Non-homologous end-joining (NHEJ) and homologous recombination (HR) represent the two main pathways for repairing DNA double-strand breaks (DSBs). During the G2 phase of the mammalian cell cycle, both processes can operate and chromatin structure is one important factor which determines DSB repair pathway choice. ATM facilitates the repair of heterochromatic DSBs by phosphorylating and inactivating the heterochromatin building factor KAP-1, leading to local chromatin relaxation. Here, we show that ATM accumulation and activity is strongly diminished at DSBs undergoing end-resection during HR. Such DSBs remain unrepaired in cells devoid of the HR factors BRCA2, XRCC3 or RAD51. Strikingly, depletion of KAP-1 or expression of phospho-mimic KAP-1 allows repair of resected DSBs in the absence of BRCA2, XRCC3 or RAD51 by an erroneous PARP-dependent alt-NHEJ process. We suggest that DSBs in heterochromatin elicit initial local heterochromatin relaxation which is reversed during HR due to the release of ATM from resection break ends. The restored heterochromatic structure facilitates HR and prevents usage of error-prone alternative processes.
| Double-strand breaks (DSBs) are critical DNA lesions because they can lead to cell death or, which is even more devastating, the formation of genomic rearrangements. Cells are equipped with two main pathways to repair such lesions, homologous recombination (HR) and non-homologous end-joining (NHEJ). HR is an error-free process and completely restores the genetic information, whereas NHEJ has the potential to form genomic rearrangements. We have previously shown that the structure of the chromatin is one important factor which determines the choice between these two pathways, such that DSBs localizing to highly condensed heterochromatic regions are mainly repaired by HR and breaks in more open euchromatic DNA undergo repair by NHEJ. Here, we investigate this aspect of DSB repair pathway choice. We show that DSB end-resection, which channels DSB repair into the process of HR, counteracts the profound local relaxation which initially takes place at the break site and reconstitutes the heterochromatic structure. Cells which are genetically modified, such that they cannot reconstitute the heterochromatic structure at resected DSBs, fail to employ HR and instead repair heterochromatic DSBs by alternative NHEJ mechanisms. Thus, chromatin modifications which occur during the process of end-resection prevent error-prone repair pathways from generating genomic rearrangements.
| DNA double-strand breaks (DSBs) are among the most deleterious cellular lesions since they threaten genomic integrity and cell viability. To counteract cell degeneration and to preserve genomic integrity, a complex network of DSB repair and signaling processes has evolved [1]–[4].
Two main DSB repair pathways exist, canonical non-homologous end-joining (c-NHEJ) and homologous recombination (HR) [5], [6]. In mammalian cells, c-NHEJ represents the major repair pathway for ionizing radiation (IR)-induced DSBs [7]. C-NHEJ repairs unresected break ends without the need for sequence homologies and can function throughout the cell cycle [8]. The key factors in c-NHEJ involve the KU70/80 heterodimer, which binds to the DSB end, and the DNA-dependent protein kinase catalytic subunit (DNA-PKcs), which, together with KU70/80, constitutes the DNA-PK holoenzyme. The repair process is completed by a complex of DNA ligase IV, XRCC4, and XLF/Cernunnos [5]. In contrast to c-NHEJ, HR is restricted to the S and G2 phases of the cell cycle where break ends undergo extensive resection and homologous DNA sequences on the sister chromatid serve as a template for repair. In addition to the repair of DSBs, HR functions during the S phase to restart stalled or collapsed replication forks [9]. HR is initiated by CtIP-dependent resection to create 3′-overhangs at the DSB ends [10], [11]. Following extended resection by EXO1 or BLM/DNA2, loading of RAD51 onto single-stranded DNA (ssDNA) is facilitated by BRCA2, XRCC2, and XRCC3. RAD54-mediated homology search then promotes strand exchange and Holliday junction formation [6]. HR is completed after repair synthesis by Holliday junction resolution and DNA end ligation. In the absence of c-NHEJ factors, DSB repair can also occur by an alternative end-joining mechanism, termed alt-NHEJ [12], [13]. In contrast to c-NHEJ but similar to HR, alt-NHEJ involves CtIP-dependent resection. The resected break ends are subsequently rejoined by a process involving micro-homologies and various repair factors such as poly (ADP-ribose) polymerase (PARP), DNA ligase I or III, and XRCC1 [14]–[17]. Although alt-NHEJ can efficiently operate in cells devoid of c-NHEJ factors, little is known about its ability to compensate for HR defects.
It has become clear over the last years that higher order chromatin structure impacts on the response to DSBs [18]. Thus, IR-induced DSBs in densely compacted heterochromatin (HC) are more difficult to repair than euchromatic (EC) DSBs and they require additional structural changes in the surrounding chromatin [19], [20]. One example are ATM-mediated chromatin changes due to KAP-1 phosphorylation [21]. In undamaged cells, KAP-1 forms HC by recruiting HP1, CHD3 and other remodeling factors [22], [23]. DSB-induced KAP-1 phosphorylation leads to release of CHD3 which locally relaxes HC and facilitates repair [23]. Other studies involving HP-1 mobilization have observed either a release from [24] or a recruitment to damaged chromatin [25]–[27]. These apparently conflicting findings have led to the suggestion that a transient release might be followed by an accumulation of HP1 at sites of DNA damage [19], [28]. However, it is often unclear how the various processes of chromatin modification impact on DSB repair and if different repair pathways are differentially affected.
Repair kinetics for IR-induced DSBs are biphasic, exhibiting a fast and a slow component [29]. The slow component accounts for the repair of a subset (15–20%) of IR-induced DSBs that are localized to HC DNA regions, whereas DSBs induced in EC regions are typically repaired with fast kinetics. In G1 phase, the fast and the slow component of DSB repair comprise a c-NHEJ mechanism [29]. ATM-dependent phosphorylation of KAP-1 on serine 824 (S824) is specifically required for the slow component [30], [31]. In G2 phase, in contrast, c-NHEJ accounts only for the fast DSB repair process, while the slow ATM-dependent HC component represents HR [32]. Thus, in G2, defined DSB populations, EC vs. HC breaks, are repaired by either c-NHEJ or HR, respectively. Despite the existence of two repair pathways in G2, a mutation in one of them leads to elevated unrepaired DSBs. Thus, c-NHEJ and HR cannot compensate for each other which might be attributed to the fact that c-NHEJ is unable to repair DSBs which have undergone extensive resection. Consistent with this notion, c-NHEJ can compensate for HR if resection is prevented by CtIP depletion [33]. What remains unclear is why alt-NHEJ, which in principal is able to rejoin resected break ends, cannot compensate for a loss of down-stream HR factors such as BRCA2 or RAD51.
In the present study, we analyzed the process of HR at HC DSBs in G2 phase. We show that the intensity of phosphorylated ATM at DSBs decreases during the process of resection, suggesting that ATM initially binds to but is then released from DSBs which undergo repair by HR. Consistent with this notion, chemical inhibition of ATM prior to but not after resection causes a repair defect. Thus, ATM has an early role during HR but is dispensable for later stages. This contradicts the situation in G1 where continuous ATM activity is required for HC DSB repair by c-NHEJ [34]. In G1, ATM functions to phosphorylate KAP-1, leading to its inactivation and local relaxation of the HC structure [30]. Moreover, depletion of KAP-1 by siRNA overcomes the requirement for ATM in G1 but leads to reduced HR usage in G2. Finally, following KAP-1 siRNA or expression of a phospho-mimic form of KAP-1, both of which cause HC relaxation, resected DSBs can be repaired by a PARP-dependent alt-NHEJ process. Together, these data show that the HC structure represents a barrier for repair by c-NHEJ and alt-NHEJ but facilitates usage of HR. ATM, which initially binds to DSBs, is released from break ends during the process of resection. This prevents usage of c-NHEJ and alt-NHEJ and commits resected DSBs to repair by HR.
We have previously demonstrated that BRCA2-deficient cells exhibit elevated γH2AX foci levels at 8 h post irradiation in G2 [1], [32]. These unrepaired DSBs have undergone efficient end-resection as evidenced by RPA loading (Figure 1A) which might explain why they cannot be repaired by NHEJ. We sought to further characterize these breaks and observed that the pATM focal intensity in G2- but not in G1-phase cells is greatly diminished at 8 h compared with 30 min time points (Figure 1A and Figure S1A). In contrast, the γH2AX focal signal is equally intensive at 30 min and 8 h in G1 and G2 (Figure S1B). We also measured the pATM focal intensity at 2 h post IR, a time point when resected and unresected DSBs are present in G2-phase cells. Of note, the pATM focal intensity of RAD51-foci-positive resected breaks is reduced compared with RAD51-foci-negative unresected breaks. In contrast, the γH2AX focal intensity is similar or even slightly increased at resected versus unresected DSBs (Figure 1B). These findings suggest that the pATM focal intensity decreases during resection in G2. pATM contributes, together with DNA-PKcs and ATR, to the phosphorylation of H2AX [35], [36]. To test if the loss of pATM intensity at the break site leads to reduced ATM activity, we measured the γH2AX focal intensity in cells with strongly diminished levels of ATR, a kinase which is activated by ssDNA regions [37]. Significantly, although ATR-deficient cells show γH2AX focal intensities at unresected DSBs similar to wildtype (wt) cells, they exhibit greatly diminished intensities at resected breaks (Figure 1C). Consistent with the notion that ATM is active at unresected but not at resected DSBs, chemical inhibition of ATM only affects γH2AX foci intensities at unresected but not at resected DSBs (Figure 1D and Figure S1C).
We next sought to confirm the immunofluorescence (IF) measurements by Western blotting. We used A549 tumor cells which can be efficiently synchronized in G1 by serum starvation and moderately enriched in G2 by double thymidine blocking (Figure S2A). The level of chromatin-bound pATM decreases with time after IR due to ongoing repair in G1 and in G2 but, importantly, at later times the pATM level per γH2AX level is smaller in G2-enriched than in G1-synchronized cells (Figure 2A). We also measured pKAP-1 (S824) levels as a specific read-out for ATM activity [21] and obtained similar results (Figure 2A). We next wished to measure pATM bound to DSBs and employed immunoprecipitation (IP) experiments. For this, we used HeLa tumor cells which can be efficiently synchronized in G2 (Figure S2B). Strikingly, pATM bound to γH2AX is readily detected at 30 min but nearly absent at 8 h post IR in G2 (Figure 2B). To directly show that the diminished pATM activity in G2 is a result of resection, we inhibited resection by depleting CtIP or BLM [38] and measured pKAP-1 levels. G2-synchronized HeLa tumor cells show a strongly reduced pKAP-1 level at 4 h post IR compared with unsynchronized cells which is fully or partly restored after CtIP or BLM depletion (Figure 2C and Figure S2C). To provide evidence for the restoration of chromatin condensation at resected DSBs, we performed IP experiments as in Figure 2B. We observed that the level of KAP-1 bound to γH2AX continuously increases with repair time (Figure 2D), possibly due to an enrichment of HC DSBs at longer times and the recruitment of KAP-1 to damaged sites as previously reported [25]. Importantly, γH2AX-bound KAP-1 is substantially phosphorylated at early times post IR but largely unphosphorylated at later times (Figure 2D). Together, these biochemical approaches confirm the IF data above and provide strong evidence that ATM accumulation and activity is strongly reduced at DSBs which undergo resection. This leads to KAP-1 dephosphorylation and possibly the restoration of HC. The observed diminished ATM activity at resected DSBs is consistent with studies using a human cell extract-based assay in which ATM is activated by blunt DSB ends and ends with short ss overhangs but not by extended ssDNA regions which arise during the process of resection [39].
ATM has been implicated in early steps of HR [33], [40], [41]. A prediction of our findings above is that ATM is no longer required for HR after resection has occurred. To test this, we inactivated ATM either before or at 2 h post IR, a time point when resection has occurred (Figure S1C), and investigated the efficiency of DSB repair. γH2AX foci numbers at 8 h post IR were substantially elevated both in G1- and G2-phase cells treated with ATM inhibitor (ATMi) before IR but only in G1-phase and not in G2-phase cells if ATMi was added 2 h post IR (Figure 3A). We also analyzed mitotic chromatid breakage in G2-irradiated cells and observed substantially elevated break levels if ATMi is administered before irradiation but not if it is added 2 h post IR (Figure 3B). HR in G2 leads to sister chromatid exchanges (SCEs) [42] which are diminished if ATM is inhibited before but not at 2 h after IR (Figure 3C). Together, these data show that ATM is dispensable for HR stages that occur after resection has taken place.
It was previously shown that ATM operates in G1 by continuously phosphorylating KAP-1 at heterochromatic DSBs and that KAP-1 depletion overcomes the requirement for this ATM function [34]. Since ATM accumulation and activity is reduced at resected DSBs, we next asked if KAP-1 depletion might affect DSB repair in G2. KAP-1 siRNA did not alter γH2AX foci numbers in wt cells but strikingly rescued the repair defect in BRCA2 mutants and cells treated with BRCA2 siRNA (Figure 4A and Figure S3A). The same effect was observed in CHO cells deficient for the HR factor XRCC3 as well as in RAD51-depleted CHO cells (Figures S3B and S3C). Moreover, KAP-1 siRNA reduced the elevated level of chromatid breaks in BRCA2-deficient cells to that of wt cells (Figure 4A). We also measured the formation of SCEs and did not observe any IR-induced SCE formation in BRCA2/KAP-1-depleted cells (Figure S3D). Finally, we investigated cells containing an integrated HR reporter with two differentially mutated GFP genes [43]. Expression of the endonuclease I-SceI generates a DSB in one of the two genes which can be repaired by HR (gene conversion) with the second gene copy as a template, resulting in a cell with functional GFP. HR frequencies assessed by the fraction of GFP-positive cells are significantly decreased after BRCA2 depletion and dual depletion of BRCA2 and KAP-1, confirming that the repair events occurring in BRCA2/KAP-1-depleted cells do not represent HR (Figure S3E). A pathway switch from HR to c-NHEJ has recently been demonstrated for heterochromatic DSBs after the inhibition of resection by CtIP siRNA, consistent with the idea that resection determines DSB repair pathway choice [33]. Therefore, we asked if RPA foci formation, as a read-out for resection, is affected by KAP-1 depletion. Significantly, wt and BRCA2-depleted cells show the same initial level of RPA foci at 2 h post IR which is unaffected by KAP-1 siRNA. These RPA foci persist in BRCA2-depleted cells up to 8 h post IR consistent with their elevated γH2AX foci level. In contrast, RPA foci numbers decrease with time due to ongoing repair in wt and BRCA2-depleted cells treated with KAP-1 siRNA (Figure 4B and Figure S3F). We also investigated RAD51 loading at resected DSBs and observed normal RAD51 foci numbers after KAP-1 siRNA in wt but not in BRCA2-depleted cells (Figure 4B).
The finding that a BRCA2-independent process repairs resected DSBs after combined BRCA2 and KAP-1 siRNA suggests that the commitment for HR results from the loss of pATM at resected DSBs which is overcome by KAP-1 depletion. To consolidate this finding, we investigated DSB repair in cells treated with KAP-1 siRNA and complemented with siRNA-resistant KAP-1 constructs which were mutated at the ATM-dependent phosphorylation site on S824 [30]. The BRCA2 repair defect, which is rescued after KAP-1 siRNA, is restored after complementation with wt KAP-1 or with KAP-1 rendered unphosphorylatable by mutating serine at position 824 to alanine (S824A). Significantly, however, KAP-1 mutated to a phospho-mimic aspartate at position 824 (S824D) fails to restore the BRCA2 repair defect (Figure 4C). Thus, KAP-1 phosphorylation at the established ATM site 824 overcomes the commitment for HR and DSB repair in the absence of BRCA2 can proceed by an HR-independent process.
Next, we wanted to investigate the process which is employed in BRCA2-deficient cells for the repair of resected DSBs. For this, we depleted BRCA2 and/or KAP-1 in cells deficient in the c-NHEJ factor XLF. XLF-defective cells show greatly elevated γH2AX foci and chromatid breaks consistent with the notion that c-NHEJ represents the predominant repair pathway in G2 [32]. Interestingly, depletion of BRCA2 leads to a similar increase in γH2AX foci/chromatid break numbers in wt cells and XLF mutants, demonstrating additivity of the two major repair pathways in G2, c-NHEJ and HR (Figure 5A). But most importantly in the present context, dual depletion of BRCA2 and KAP-1 did not affect γH2AX foci/chromatid break numbers in XLF mutants, demonstrating that the HR defect is rescued by KAP-1 depletion even in the absence of the c-NHEJ factor XLF (Figure 5A). The same effect was observed in CHO cells deficient in the c-NHEJ factor KU80 (Figure S4A).
We then tested if an alt-NHEJ pathway repairs DSBs in BRCA2/KAP-1-depleted cells and employed chemical inhibition of PARP (PARPi), a factor which has been implicated in alt-NHEJ [14], [17]. γH2AX foci and chromatid breaks were not significantly affected in wt cells treated with PARPi, demonstrating that alt-NHEJ processes do not contribute substantially to IR-induced DSB repair in normal cells. However, the elevated level of γH2AX foci/chromatid breaks in BRCA2-deficient cells, which is rescued after KAP-1 siRNA, is restored by PARPi (Figures 5B and 5C). Thus, PARPi precluded the repair events which arose in BRCA2-deficient cells after KAP-1 siRNA, demonstrating that a PARP-dependent process can function as a back-up pathway for HR. We also investigated other factors which have been described to function in alt-NHEJ. In CHO mutants deficient in XRCC1 as well as in cells deficient for DNA ligase I and III, KAP-1 failed to rescue the elevated γH2AX foci level which is conferred by a deficiency in BRCA2 or RAD51 (Figure 5D and Figure S4B). Consistent with the notion that alt-NHEJ can function as a back-up pathway for HR, we observed greatly increased levels of chromatid fusions in BRCA2/KAP-1-depleted cells (Figure 5E). To characterize the nature of these chromatid fusion events, we employed fluorescence-in-situ-hybridization (FISH) analysis with chromosome-specific probes. In one set of experiments, we used probes for chromosomes 1, 2 and 4 and observed that all fusion events (∼40 fusions from the analysis of ∼800 cells) occurred between heterologous chromosomes, that is, between a stained and an unstained chromosome or between two differently stained chromosomes (Figure 5F). Further, we employed probes for chromosome 19 which is exceptionally rich in KAP-1 binding sites and for the similar-sized chromosome 18 which is largely devoid of these sites [44]. Following BRCA2 depletion, we observed significantly higher breakage levels in chromosome 19 compared with chromosome 18, confirming that HR in G2 occurs mainly in KAP-1-dependent HC (Figure 5G). Importantly, following dual depletion of BRCA2 and KAP-1, chromosome fusions occur more often in chromosome 19 than in chromosome 18 confirming the notion that they arise from the misrejoining of chromatid breaks in KAP-1-dependent HC (Figure 5G).
The data above show that KAP-1 depletion allows heterochromatic DSBs to be repaired by an alt-NHEJ pathway in the absence of BRCA2, XRCC3 or RAD51. It is, however, unclear how the efficiency of HR in wt cells is affected by KAP-1-mediated chromatin changes. As shown above, γH2AX foci and chromatid breaks are repaired with similar kinetics with and without KAP-1 siRNA (see Figure 4A) but it is not known if repair after KAP-1 siRNA involves HR or, as in the case of HR mutants, an alt-NHEJ pathway. To address this question, we investigated the formation of SCEs in mitotic cells and observed greatly diminished SCE levels after KAP-1 siRNA in wt cells (Figure 6A). We also employed the HR reporter assay described above (Figure S3E) and observed strongly reduced HR levels following KAP-1 depletion (Figure 6B). Thus, KAP-1-depleted cells do not employ HR although repair occurs efficiently. We also analyzed chromatid fusion events as a read-out for incorrect end-joining. Strikingly, KAP-1-depleted cells show elevated chromosomal fusions, suggesting that the DSBs are repaired by an error-prone alt-NHEJ pathway (Figure 6C). This notion is consolidated by the observation that PARPi increases γH2AX foci and chromatid break numbers in cells depleted for KAP-1 or complemented with phospho-mimic KAP-1 (S824D) (Figures 6D and 6E). Further, cells deficient in DNA ligase I and III or in XRCC1 show elevated γH2AX foci levels following KAP-1 depletion (Figures 6F and 6G). Taken together, this data shows that HR is efficiently used in cells with unphosphorylatable KAP-1 and cannot occur if KAP-1 is depleted.
HR involves resection of DSB ends. Here, we investigated the process of HR at HC DSBs in G2 and showed that pATM, which initially binds to DSB ends, is released from the break sites during the process of resection. This leads to diminished KAP-1 phosphorylation at HC breaks and a commitment to repair such resected DSBs by HR. If the loss of KAP-1 phosphorylation is overcome by KAP-1 depletion or expression of phospho-mimic KAP-1, both of which are known to cause local HC relaxation, this commitment to HR is abolished and resected DSBs are repaired by an alt-NHEJ process. Thus, KAP-1-dependent HC facilitates later stages of HR whereas c-NHEJ and alt-NHEJ both require continuous HC relaxation due to ATM-dependent KAP-1 phosphorylation (see Figure 6H).
ATM binding and activation at DSB ends occurs within minutes after damage induction and is important for the initiation of various signaling processes [45]. Concomitant with the induction of signaling pathways, a variety of chromatin remodeling processes are initiated. This involves modifications which either relax or condense the chromatin structure in the surrounding of DSBs. However, it is currently unclear how these changes are chronologically orchestrated and how they differentially affect different DSB repair pathways in different chromatin compartments. Therefore, we focused our investigation on chromatin modifications which occur in HC regions due to the process of resection in order to specifically investigate how such chromatin changes impact on later stages of HR. We did not examine chromatin remodeling processes at early times which affect the decision to initiate resection.
We have previously shown that ATM is dispensable for the majority of DSB repair in G1 but that HC breaks strictly require ATM [30]. ATM's function during HC DSB repair in G1 involves continuous KAP-1 phosphorylation which leads to local HC relaxation [23]. Our finding that ATM is released from resected DSBs in G2 was therefore unexpected. However, there is precedence in the literature that ATM changes binding properties upon resection of DSBs. First, ATM's binding affinity to break ends has been reported to be attenuated with the progressive presence of ssDNA at resected DSBs [39]. This ATM attenuation is accompanied by increasing ATR activity [39], consistent with our result that H2AX phosphorylation at RAD51-foci-positive DSBs requires ATR. Second, 53BP1, a damage response factor which localizes to and facilitates pATM accumulation at DSB sites [34], has been reported to show reduced occupancy at resected DSBs in G2 [46]. Although the reported reduction of ATM accumulation and activity at resected breaks is consistent with published data, the functional consequence of this finding was hitherto unclear.
In G2 phase, DSB repair can be performed by NHEJ and HR. It is therefore remarkable that cells with mutations in BRCA2, XRCC3 or RAD51 exhibit unrejoined DSBs, which obviously are refractory to repair by NHEJ. Thus, it has been suggested that the process of resection commits DSB repair to HR and prevents usage of NHEJ [33]. Here, we provide mechanistic insight into the processes determining pathway usage upon resection. Since ATM is released from resected DSBs we reasoned that the concomitant reduction in KAP-1 phosphorylation prevents repair of resected breaks by NHEJ. Indeed, if loss of ATM-dependent KAP-1 phosphorylation is overcome by KAP-1 depletion or expression of phospho-mimic KAP-1, BRCA2-, XRCC3- or RAD51-deficient cells exhibit normal repair kinetics. Thus, it is not the resection per se but the loss of ATM activity at resected breaks which commits repair to HR.
HC DSBs which remain unrepaired in BRCA2-, XRCC3- or RAD51-deficient cells can be repaired if HC relaxation is provided by KAP-1 depletion or expression of phospho-mimic KAP-1. Interestingly, these DSBs undergo resection as evidenced by normal RPA foci formation. Thus, HC repair occurring in the absence of BRCA2, XRCC3 or RAD51 must involve a pathway which is capable of dealing with resected breaks. Consistent with the notion that alt-NHEJ can repair resected DSBs, we showed that the HC repair events occurring in the absence of BRCA2, XRCC3 or RAD51 require PARP, XRCC1 and DNA ligase I/III. We also observed that HC repair in the absence of BRCA2 has a significant propensity to lead to chromatid exchanges in G2-irradiated cells. Because alt-NHEJ has been implicated in the formation of genomic exchanges [47]–[50], this finding supports our contention that HC repair in the absence of BRCA2, XRCC3 or RAD51 involves alt-NHEJ.
Perhaps surprisingly, we observed that the process of HR is nearly abolished in cells with depleted KAP-1, even in the presence of functional HR factors. This suggests that DSB repair pathway usage is significantly affected by chromatin modifications, favoring HR in condensed genomic regions. This notion is further supported by the observation that PARP inhibition or the loss of XRCC1 or DNA ligase I and III leads to elevated unrepaired breaks in KAP-1-depleted cells, which not only demonstrates that cells use alt-NHEJ but also, that they cannot employ HR in the absence of KAP-1-dependent HC. In summary, these findings establish that KAP-1-dependent HC is not only a barrier to repair by c-NHEJ or alt-NHEJ but, unexpectedly, also facilitates the process of HR.
Consistent with our results, depletion of HP1α or KAP-1 strongly reduces gene conversion frequencies in a I-SceI-based HR assay [25]. Furthermore, HP1α and KAP-1 is recruited to chromatin damaged by laser- or X-irradiation [26], [27] and depletion of HP1α diminishes SCE formation after treatment with camptothecin [51]. One explanation of how HC might promote HR is that a reduced spatial distance between sister chromatids in HC regions facilitates homology search [52]. In support of this idea, we have recently obtained preliminary evidence that the average distance between sister chromatids, measured by FISH analysis with locus-specific probes, is substantially larger in EC versus HC regions (Geuting et al., unpublished data). A similar mechanism has been suggested for cohesin proteins which might promote HR by providing the required proximity of sister chromatids in G2 phase [53]. Another explanation of how HC might facilitate HR is by suppressing alt-NHEJ processes. Although it is well established that the presence of KU70/80 at DSB ends prevents repair by alt-NHEJ, KU70/80 is likely released from resected DSB ends. Chromatin condensation occurring due to ATM release at resected DSBs might represent an alternative mechanism to keep error-prone alt-NHEJ processes in check.
In conclusion, our study provides mechanistic insight into sequential events determining DSB repair pathway usage. First, we demonstrate that ATM activity is diminished at DSBs which undergo resection during the process of HR. Second, the concomitant loss of pKAP-1 at resected DSBs leads to local reconstitution of the HC superstructure and prevents repair of resected DSBs by alt-NHEJ. Thus, our study links two seemingly unrelated findings by showing how modifications at DSBs undergoing resection affect chromatin remodeling processes and DSB repair pathway usage.
Immortalized and transformed cell lines were 82-6 hTert (wt), HSC62 hTert (BRCA2-deficient, kindly provided by Dr. M. Digweed), 2BN hTert (XLF-deficient, kindly provided by Dr. P. Jeggo) and F02-98 hTert (ATR-deficient, kindly provided by Dr. P. Jeggo) human fibroblasts, HeLa-S3, HeLa pGC (kindly provided by Dr. J. Dahm-Daphi) and A549 human tumor cells, and CHO-AA8 (wt), IRS1SF (XRCC3-deficient; kindly provided by Dr. L. Thompson), CHO-K1 (wt), XRS6 (KU80-deficient, kindly provided by Dr. P. Jeggo), CHO-9 (wt) and EMC11 (XRCC1-deficient, kindly provided by Dr. B. Kaina) hamster cells. HeLa-S3 and A549 tumor cells were cultured in DMEM with 10% FCS and 1% NEAA; HeLa pGC cells additionally in 0.3 µg/ml puromycin. Human fibroblasts and CHO cells were cultured in MEM with 20% FCS, 1% NEAA. All cells were maintained at 37°C in a 5% CO2 incubator.
SiRNA transfection was carried out with HiPerFect Transfection Reagent (Qiagen) following the manufacturer's instructions. siRNAs used in the experiments were: BLM (50 nM), Control (10 nM), CtIP (20 nM), KAP-1 (25 nM), RAD51 (20 nM), Lig I (20 nM), Lig III (20 nM) (Qiagen), and BRCA2 (25 nM) (SmartPool, Dharmacon). SiRNA sequences were: BLM (AAG CUA GGA GUC UGC GUG CGA), BRCA2 (GAA ACG GAC UUG CUA UUU A; GUA AAG AAA UGC AGA AUU C; GGU AUC AGA UGC UUC AUU A; GAA GAA UGC AGG UUU AAU A), Control (AAU UCU CCG AAC GUG UCA CGU), CtIP (UCC ACA ACA UAA UCC UAA UUU), KAP-1_A (CAG UGC UGC ACU AGC UGU GAG), KAP-1_B (CAU GAA CCC CUU GUG CUG UUU), RAD51 (AAG GGA AUU AGU GAA GCC AAA), Lig I (AAG GCA UGA UCC UGA AGC AGA), Lig III (AAC CAC AAA AAA AAU CGA GGA). Experiments were performed 48 h following siRNA transfection. For GFP-tagged siRNA-resistant KAP-1 plasmid transfection, HeLa tumor cells were incubated with KAP-1_B or KAP-1_B and BRCA2 siRNA and, 8 h later, transfected with 1 µg plasmid DNA using Lipofectamine LTX Transfection Reagent (Life Technologies). Cells were irradiated with 2 Gy, fixed and stained for γH2AX, EdU and GFP. Only GFP-positive cells were analyzed.
A549 tumor cells were used for G1 synchronization and G2 enrichment. HeLa tumor cells were only used for G2 enrichment. G1 synchronization was carried out by 48 h serum starvation in DMEM without FCS and NEAA. 0.5 h before irradiation, medium was replaced by DMEM with FCS and NEAA. For G2 enrichment, a double thymidine blocking was used. Cells were blocked 16 h with 2 mM thymidine (Sigma), released in fresh medium for 9 h, blocked again with 2 mM thymidine for 16 h and released in fresh medium for 7–8 h. Synchronization was controlled by FACs analysis as described previously [54]. X-irradiation was performed at 90 kV and 19 mA with an aluminum filter (dose rate: 2 Gy/min). Chemical inhibitors were added 0.5 h prior to IR and maintained during repair incubation. The ATM inhibitor (Tocris KU 60019), the DNA-PK inhibitor Nu7441 (Tocris NU7026) and the PARP inhibitor PJ34 (Calbiochem PARP inhibitor VIII PJ34) were used at concentrations of 5 µM, 10 µM and 20 µM, respectively. Repair incubation was limited to time periods which provided that the majority of G2-irradiated cells remained in G2 (controlled by FACs analysis).
Cells were grown on glass coverslips. EdU (10 µM) was added 0.5 h prior to IR to discriminate between S- and G2-phase cells. In experiments analyzing G1-phase cells, nocodazol (100 ng/ml) was added 0.5 h prior to IR to prevent G2-phase cells progressing into G1 during repair incubation [55]. Cells were fixed and stained as described [56] and additionally stained with Click-it EdU (Life technologies). Antibodies used were: mouse-α-γH2AX at 1∶2000 (Millipore); rabbit-α-γH2AX at 1∶2000 (Abcam), mouse-α-pATM at 1∶1000 (Biomol), rabbit-α-RAD51 at 1∶15000 (Abcam), mouse-α-RPA at 1∶2000 (Neomarkers) and mouse-α-GFP at 1∶200 (Roche). Cells were analyzed with a Zeiss microscope and Metafer software (Metasystems). Samples were evaluated in a blinded manner. Foci intensities were analyzed using ImageJ software (see Figure S1A).
HeLa pGC cells were incubated with siRNA and, 24 h later, transfected with 3 µg pBL464-pCBASce plasmid DNA using MaTra transfection (IBA). After 24 h, cells were again siRNA treated and, 48 h later, fixed and stained. 10000 cells per sample were analyzed with a Zeiss microscope and Metafer software (Metasystems).
Whole cell extracts were prepared as described [56]. For chromatin fractionation, cells were resuspended two times in NP-40 buffer (10 mM Tris/HCl pH 7.5, 10 mM NaCl, 3 mM MgCl2, 30 mM sucrose, 0.5% NP-40, 0.2 mM sodiumvanadate, 0.5 mM PMSF) and centrifuged for 10 min at 1500× g. Cell pellet was resuspended in Glycerol buffer (20 mM Tris/HCl pH 7.9, 100 mM KCl, 0.2 mM EDTA, 20% glycerol, 0.2 mM sodiumvanadate, 0.5 mM PMSF) and incubated 10 min on ice. After centrifugation (10 min, 1500× g) chromatin fraction was lysed and sonicated in RIPA buffer (50 mM Tris/HCl pH 8, 150 mM NaCl, 0.5 Na-deoxycholate, 1% Triton, 0.1% SDS). For immunoprecipitation, cells were fixed with 3% paraformaldehyd containing 2% sucrose for 5 min at 4°C, immediately washed with PBS, scraped in medium and centrifuged for 10 min at 400× g. Cells were resuspended two times in NP-40 buffer containing 15 mM caffeine and centrifuged for 10 min at 1500× g. Cell pellet was resuspended in equal volume Nuclease buffer (10 mM HEPES pH 7.5, 10 mM KCl, 1 mM CaCl2, 1.5 mM MgCl2, 0.34 M sucrose, 10% glycerol, 0.1% Triton-X-100, 0.2 mM sodiumvanadate, 0.5 mM PMSF, 15 mM caffeine), micrococcal nuclease (500 U/ml) was added and suspension was incubated for 45 min at 37°C. Equal volume of Solubilization buffer (2% NP-40, 2% Triton-X-100, 600 mM NaCl in Nuclease buffer) was added before mixing, brief sonication and clearifing for 10 min at 8000× g. Dynabead Protein G (Invitrogen) were blocked 1 h with 100 µg/ml salmon sperm DNA in 0.1% BSA/PBS and antibodies (4 µg) were linked to the beads, washed two times in 0.1% BSA/PBS and then incubated with the cell extract at 4°C over night. Beads were washed three times in Wash buffer (equal volume of Nuclease buffer and Solubilization buffer) and boiled in 2× Laemmli buffer for 5 min at 95°C.
Western blotting was carried out at 300 mA for 1 h or at 80 mA over night. Nitrocellulose membrane (Roth) was blocked for 1 h in 5% low fat milk or 5% BSA in TBS/0.1% Tween20. Antibody incubation was carried out in TBS/0.1% Tween20/1% low fat milk or 5% BSA over night at 4°C, followed by HRP-conjugated secondary antibody incubation in PBS/0.1% Tween20/1% low fat milk or 5% BSA for 1 h. Immunoblots were developed using ECL (Roche). Signal detection was carried out with a chemi-smart-system (Vilber Lourmat). Primary antibodies used were: rabbit-α-pATM at 1∶1000 (Epitomics); rabbit-α-pKAP-1 (S824) at 1∶10000 (Epitomics); rabbit-α-KAP-1 at 1∶1000 (abcam); mouse-α-BRCA2 at 1∶1000 (Cell signaling); rabbit-α-GAPDH at 1∶1000 (Santa Cruz); mouse-α-γH2AX at 1∶1000 (Millipore); mouse-α-H3 at 1∶1000 (abcam); mouse-α-RPA2 at 1∶1000 (Calbiochem); rabbit-α-pRPA2 (S4/8) at 1∶10000 (Bethyl).
EdU (10 µM) was added 0.5 h prior to IR and maintained to discriminate between S- and G2-phase cells. PCCs were harvested at 8 h, mitotic cells for SCE or FISH analysis between 5–8 h after IR as described [32]. Microscope slides were stained with DAPI (0.2 µg/ml) and Click-it EdU. For FISH analysis, whole chromosome probes for chromosomes 1, 2, and 4 or for chromosomes 18 and 19 were used (Metasystems). Chromosome spreads were recorded by Metafer software (Metasystems). Only EdU-negative chromosome spreads were analyzed.
|
10.1371/journal.pntd.0003683 | Down Regulation of NO Signaling in Trypanosoma cruzi upon Parasite-Extracellular Matrix Interaction: Changes in Protein Modification by Nitrosylation and Nitration | Adhesion of the Trypanosoma cruzi trypomastigotes, the causative agent of Chagas' disease in humans, to components of the extracellular matrix (ECM) is an important step in host cell invasion. The signaling events triggered in the parasite upon binding to ECM are less explored and, to our knowledge, there is no data available regarding •NO signaling.
Trypomastigotes were incubated with ECM for different periods of time. Nitrated and S-nitrosylated proteins were analyzed by Western blotting using anti-nitrotyrosine and S-nitrosyl cysteine antibodies. At 2 h incubation time, a decrease in NO synthase activity, •NO, citrulline, arginine and cGMP concentrations, as well as the protein modifications levels have been observed in the parasite. The modified proteins were enriched by immunoprecipitation with anti-nitrotyrosine antibodies (nitrated proteins) or by the biotin switch method (S-nitrosylated proteins) and identified by MS/MS. The presence of both modifications was confirmed in proteins of interest by immunoblotting or immunoprecipitation.
For the first time it was shown that T. cruzi proteins are amenable to modifications by S-nitrosylation and nitration. When T. cruzi trypomastigotes are incubated with the extracellular matrix there is a general down regulation of these reactions, including a decrease in both NOS activity and cGMP concentration. Notwithstanding, some specific proteins, such as enolase or histones had, at least, their nitration levels increased. This suggests that post-translational modifications of T. cruzi proteins are not only a reflex of NOS activity, implying other mechanisms that circumvent a relatively low synthesis of •NO. In conclusion, the extracellular matrix, a cell surrounding layer of macromolecules that have to be trespassed by the parasite in order to be internalized into host cells, contributes to the modification of •NO signaling in the parasite, probably an essential move for the ensuing invasion step.
| Interaction of Trypanosoma cruzi with the extracellular matrix (ECM) is an essential step in the invasion of mammalian cells. However, the nature of the signaling triggered in the parasite is poorly understood. Herein the key role of nitric oxide in T. cruzi signaling is described, using an ECM preparation, in the absence of host cells. Inhibition of NOS activity, with the expected decrease in •NO production, as well as decrease in cGMP concentration were observed by the incubation of T. cruzi trypomastigotes with ECM. Additionally, lower levels of protein S-nitrosylation and nitration were detected. These post-translational modifications have been analyzed by biotin-switch and protein immunoprecipitation approaches coupled to mass spectrometry. The presence of both modifications was confirmed for specific proteins, as mucin II (S-nitrosylation), histones, enolase and tubulins. To our knowledge, decrease in the •NO signaling pathway upon T. cruzi trypomastigotes adhesion to ECM, affecting both the canonical pathway (•NO-soluble guanylyl cyclase-cGMP) and protein S-nitrosylation and nitration is described for the first time in this parasite.
| Trypanosoma cruzi is the etiological agent of Chagas disease, an infectious disease affecting areas of poor socioeconomic development. The parasite infects a wide range of mammalian hosts, including humans, from which 7–8 million are infected and other 25 million are at risk of contamination [1]. T. cruzi trypomastigotes, the classical parasite infective form, invade almost all mammalian cells, including macrophages [2,3,4], being exposed to nitrosative and oxidative stress during the life cycle [5,6,7]. The cytotoxic effect of •NO and its derivatives on pathogens such as T. cruzi is well known.
In mammals and other organisms, the free radical •NO is endogenously synthesized by nitric oxide synthase catalyzing the conversion of L-arginine to L-citrulline [8], a reaction that depends on heme, FAD, FMN and tetrahydro-L-biopterin (BH4) as co-factors. •NO is highly reactive towards O2, but reactions with biological molecules preferentially occur with •NO- derived species (N2O3, NO2• or ONOO-) [9]. Biologically, •NO plays essential role in cell signaling, acting by two main mechanisms: (i) activation of guanylyl cyclase, yielding cGMP—the classical pathway; or (ii) acting in post-translational modifications such as S-nitrosylation and tyrosine nitration- the non-classical pathway [10,11]. Protein S-nitrosylation and tyrosine nitration affect the activity of many relevant targets of several biological processes [12,13].
Proteins are S-nitrosylated (SNO) by the addition of a nitroso group into a cysteine residue in a non-enzymatic process, dependent on the local nitric oxide concentration or by transnitrosylation, a key mechanism in •NO signaling (acquisition of a •NO from another S-nitrosothiol) [14,15,16]. Denitrosylation may occurs by nonenzymatic mechanisms or by the action of denitrosylases [17,18,19]. New targets of S-nitrosylation are being extensively described in different organisms due to the development of tools such as the traditional biotin-switch technique associated with proteomic analysis [20,21]. As an example, 319 putative S-nitrosylation targets, as well as enzymatic denitrosylating and transnitrosylating activities in Plasmodium falciparum were recently described [22]. Of note, P. falciparum lacks a NOS ortholog and probably produces •NO from a nitrate/nitrite chemical reduction pathway [23].
In contrast to S-nitrosylation, tyrosine nitration of proteins was classically regarded as an undesired byproduct of radical species with greater reactivity capable of oxidizing tyrosine to 3-nitro-tyrosine. However, tyrosine nitration of proteins occurs under physiological conditions, with an increment of 3-nitro-tyrosine in many physiopathological and aging processes. Protein nitration is mediated by free radical reactions, with the intermediate •Tyr reacting with •NO or •NO2 [24]. Not all proteins are nitrated, pointing out to the specificity of the modification. Only one or two specific protein residues are preferentially modified and a close relationship between protein tyrosine nitration and the presence of a transition metal has been made [rev. 25]. Although not well established as S-nitrosylation, evidence gathered in the past few years suggests that tyrosine-nitrated proteins regulate several biological processes, such as stress response in plants [26], cytochrome c regulation [27], protein degradation [28], control of the redox environment [29] and PKC signaling [30].
There is a limited knowledge of •NO signaling in T. cruzi, as happens with other parasites [31]. S-nitrosylation or tyrosine nitration of T. cruzi proteins remains largely unexplored, despite the relevance of •NO and •NO-derived species produced by mammals in response to T. cruzi infection. In vitro treatment of cruzipain, the major T. cruzi papain-like cysteine proteinase [32], with •NO donors led to inhibition of the enzyme activity [33], but this modification has not been reported in vivo. Additionally, the putative signaling in response to the endogenous •NO formation is mostly unknown in T. cruzi.
Biochemical evidence of NOS activity, with •NO donors leading to an increase in the cGMP concentration was described in T. cruzi extracts [34]. Probing with an anti-neuronal NOS antibody the enzyme was localized in the inner surface of cell membranes, cytosol, flagellum and apical extremity [35]. However, NOS and guanylyl cyclase orthologs seem to be lacking in the parasite genome. An adenylyl cyclase containing a putative guanylyl cyclase domain was suggested to be responsible for cGMP production [36,37] and a soluble dual-specificity phosphodiesterase (TcrPDEC), capable of cleaving both cAMP and cGMP with similar Km (20–31.6 μM and 78.2 μM, respectively) would be responsible for the degradation of cGMP [38,39,40,41]. The downstream effector of cGMP is assumed to be the cGMP-dependent protein kinase (PKG). While the involvement of cAMP is relatively well known in biological processes, such as T. cruzi metacyclogenesis [40,42,43] and downstream proteins that interact with PKA were characterized [44] the knowledge of cGMP signaling is far from being understood. Although the role of cGMP signaling pathway is currently unresolved in T. cruzi and other kinetoplastids, the presence of cGMP-specific kinase in T. brucei [45] and Leishmania [46] points out to the presence of the pathway in these parasites.
In addition to a structural role in tissues, the extracellular matrix (ECM), a complex tridimensional structure composed of more than 300 proteins and glycoproteins [47], is relevant for many cellular signaling pathways, including •NO signaling in mammalians. T. cruzi trypomastigotes bind to components of the extracellular matrix (ECM), such as laminin, fibronectin, collagen, heparan sulfate, thrombospondin or galectin-3, as an early event of the infection process of mammalian cells [2,3,48,49]. Despite this, the signaling pathways triggered in T. cruzi after adhesion to ECM or its components are less characterized. Adhesion to laminin or fibronectin leads to changes in the phosphorylation level of T. cruzi proteins, including paraflagellar rod proteins and tubulins, probably involving the ERK1/2 pathway [50].
Herein, the role of nitric oxide in post-translational modification of proteins as a consequence of trypomastigotes adhesion to ECM is focused. A decrease of the •NO signaling pathway, including S-nitrosylation and tyrosine nitration of proteins was observed in T. cruzi trypomastigotes upon adhesion to host cell-derived ECM, an essential event for mammalian host cell invasion. To our knowledge this phenomenon is described for the first time.
S-methyl-methanethiosulfonate (MMTS), imidazole, L-arginine and L-citrulline, 3-isobutyl-1-methylxanthine (IBMX), GTP, sulfanilamide, N-(1-naphthylethylenediamine dichloride, HgCl2, S-nitrosoglutathione, neocupreine and also the antibodies: anti-alfa tubulin, anti-nitrosocysteine, anti-mouse FITC conjugated, anti-rabbit HRP conjugated were purchased from Sigma-Aldrich (St. Louis, USA). Phosphoric acid and sodium hydroxide were acquired from Synth (São Paulo, Brazil). Sodium dihydrogenphosphate was purchased from Merck (Darmstadt, Germany). The antibody anti-nitro-tyrosine was obtained from Millipore (Billerica, USA). Sepharose beads, anti-rabbit Alexafluor 555 conjugated and DAPI were acquired from Invitrogen (Carlsbad, USA). EZ-link HPDP Biotin was from Thermo Scientific (Waltham, USA). The L-[3H]-Arginine was from PerkinElmer (Waltham, USA).
T. cruzi epimastigotes, Y strain, were cultivated at 28°C in Liver infusion Tryptose (LIT) medium supplemented with 10% fetal bovine serum (FBS), up to 107 epimastigotes per mL. [51]. T. cruzi trypomastigotes, Y strain, were maintained by infection in LLC-MK2 cells in DMEM supplemented with 2% FBS at 37°C and 5% CO2. Five days after infection, trypomastigotes released into the medium were collected, washed in DMEM 2% FBS (10,000 x G for 12 minutes) and resuspended to adequate cell density accordingly to the experiment [52].
Trypomastigotes (1x109/mL) were incubated with 10 mg/mL ECM (Gibco) for 2 h, unless otherwise stated, at 37°C and 5% CO2. After incubation, parasites were washed twice in PBS containing 5 mM NaF, 2 mM Na3VO4, 50 μM Na β-glicerophosphate, 1 mM PMSF and protease inhibitor cocktail (Sigma-Aldrich), and kept at -80°C until used.
After incubation with ECM, parasites were centrifuged (10,000 x G, 5 minutes) and the supernatant separated for nitric oxide quantification, as described by the manufacturer (Measure-iT High-Sensitivity Nitrite Assay Kit, Invitrogen)
Nitric oxide synthase activity was determined by following the conversion of L[3H]-arginine to L[3H]-citrulline in T. cruzi cell lysates (5x108), accordingly to the manufacturer (Cayman), and as previously described [34]. The presence of [3H]-citrulline was confirmed by thin layer chromatography and radioactivity measurement of the spots, as described [34].
cGMP was measured in T. cruzi extracts (5x108 cells) accordingly to the commercial EIA test Biotrak instructions (GE Healthcare). Due to the intrinsic presence of extracellular matrix proteins in some of the experimental assays specific activity could not be calculated and, thus, results are expressed in total femtomoles produced.
Total S-NO was quantified by the Saville-Griess method, as described elsewhere [53]. Briefly, the parasite pellet (109) was lysed in 20 mM Tris-HCl buffer, pH 7.4, containing 0.1% Triton X-100, 5 mM NaF, 2 mM Na3VO4, 50 μM Na β-Glycerophosphate, 1 mM PMSF and protease inhibitor cocktail (Sigma-Aldrich) and centrifuged (10 minutes at 14,000 x G, 4°C). Twenty μL of supernatant were then added to 180 μL reaction buffer (57 mM Sulfanilamide, 1.2 mM N-(1-Naphthyl) ethylenediamine dihydrochloride in PBS, pH 7.4) and the reaction started by the addition of 100 μM HgCl2. After 30 minutes at room temperature and in the dark, the absorbance at 496 nm was measured. Controls without HgCl2 were included to account for NO already present in the sample. The amount of total S-NO was estimated against a standard curve with S-nitrosoglutathione.
The parasite pellet was lysed in H2O: methanol (1:1, v:v), centrifuged for 10 minutes at 14,000 x G, 4°C and the supernatant collected was dried in SpeedVac. The resulting dried pellet was resuspended with 200 μL MilliQ water and centrifuged (10 minutes at 14,000 x G, 4°C) to remove impurities. The supernatant was analyzed by a capillary electrophoresis system (model PA 800, Beckman Coulter Instruments, Fullerton, USA), equipped with DAD detector and a temperature control device. Data acquisition and treatment were carried out by the vendor software (32 Karat Software version 8.0, Beckman Coulter). A fused silica capillary (Polymicro Technologies, Phoenix, USA) of 50.2 cm total length, 40.0 cm effective length and 50 μm i.d. was used. The capillary was preconditioned as follows: 1 mol.L-1 NaOH (5 minutes/20 psi), MilliQ water (5 min/20 psi) and background electrolyte (BGE) (5 minutes/20 psi). The BGE was comprised of 50 mmol.L-1 of sodium dihydrogenophosphate at pH 2.5, adjusted with phosphoric acid. Samples were injected hydrodynamically by applying a 0.5 psi pressure during 2 s. The conditions applied during separation were voltage of 25 kV and detection at 200 nm. To quantify arginine and citrulline in the samples, analytical curves were constructed with background electrolyte the linear range of 50–200 mg.mL-1 and 1–50 mg.mL-1, respectively. Imidazole was used as internal standard (50 mg.mL-1).
After incubation with ECM and subsequent washes, parasites were fixed in 2% paraformaldehyde for 15 minutes at room temperature, pelleted by centrifugation (5,000 x G for 5 minutes), resuspended in PBS and added to a cover glass and left to dry for 16 hours at room temperature. After permeabilization of the parasites (PBS containing 1% BSA and 0.1% Triton X-100 for one hour at 37°C), anti-nitrosocysteine (rabbit, 1:200), anti-nitro-tyrosine (rabbit, 1:500) or anti-alpha-tubulin (mouse, 1:500) were added and incubated for 2 h at 37°C. After exhaustive washes with PBS containing 1% BSA, the correspondent secondary antibodies were added (anti-rabbit Alexa 555 conjugated, 1:500; anti-mouse FITC conjugated, 1:100), followed by one hour incubation at 37°C. After successive washes in PBS-1% BSA, the slides were mounted in a solution containing 50% glycerol, 50% milliQ H2O and 10 μg DAPI. The images were taken on an ExiBlue camera (Qimaging) coupled to a Nikon Eclipse E 600 optical microscope and deconvoluted using the software Huygens Essential (Scientific Volume Imaging).
Proteins from the parasite were extracted with Laemmli Buffer [54] without reducing agents. SDS polyacrylamide gel electrophoresis was performed in a 6–16% gradient polyacrylamide gel and transferred to a 0.45 μm nitrocellulose membrane for 16 hours at 15 V. The membrane was blocked in 5% BSA and incubated with the primary antibody (anti-nitrosocysteine or anti-nitro-tyrosine, produced in rabbit, 1:2000), washed thrice in PBS-0.1% Tween 20, incubated with secondary antibody (anti-rabbit conjugated with HRP, 1:8000 dilution), washed five times in PBS-0.1% Tween 20 and developed by electrochemiluminescence.
Conversion of protein SNO to biotinylated groups was performed as described [20]. Briefly, parasites were lysed by sonication in 25 mM HEPES, 50 mM NaCl, 0.1 mM EDTA, 1% NP-40, 5 mM NaF, 2 mM Na3VO4, 50 μM Na β-glicerophosphate, 1 mM PMSF and protease inhibitor cocktail (Sigma), then clarified by centrifugation at 14,000 x G for 10 min at 4°C. The supernatant was blocked in 250 mM Hepes-NaOH buffer, pH 7.7, containing 1 mM EDTA, 0.1 mM Neocuproine, 2.5% SDS and 20 mM MMTS for 20 minutes at 50°C, protected from light. Proteins were then precipitated by the addition of six volumes of ice cold acetone for one hour at -20°C. After successive washes in 70% acetone, the precipitate was resuspended in 250 mM Hepes-NaOH buffer, pH 7.7, containing 1 mM EDTA, 0.1 mM neocuproine, 1% SDS, 4 mM Biotin-HPDP and 1 mM sodium ascorbate. The mix was incubated for one hour at 25°C at dark. Proteins were then precipitated in ice cold acetone for one hour at -20°C and washed extensively in 70% acetone.
Biotin-containing proteins were prepared for sequencing as described [21] Proteins were digested by trypsin (sequencing grade 1:100, mass/mass) for 16 hours at 37°C and the reaction stopped by the addition of 0.5 mM PMSF. Then, biotin-containing proteins were pulled-down using streptavidin beads. After 2 hours incubation at room temperature under gentle shaking, the beads were washed thrice in 20mM Tris-HCl buffer pH 7.7, containing 1 mM EDTA and 0.4% Triton X-100. Proteins were eluted with 5 mM ammonium bicarbonate containing 150 μL of 50 mM 2-mercaptoethanol for 5 minutes. Sequencing was determined (Veritas Life Sciences, Brasil) using a LTQ-Orbitrap coupled to nLC-MS/MS. Acquired data were automatically processed by CPAS (Computational Proteomics Analysis System) [55] and only peptides with high quality were considered (expected score <0.2). The TryTripDB was used for protein search combining Esmeraldo-like, non-Esmeraldo-like and unassigned.
For Mucin II validation, proteins were biotinylated as described above, ressuspended in Tris-HCl 20 mM and incubated at 12°C for 16 hours with streptavidin beads. After three successive washes, the beads were ressuspended in Laemmli buffer without reducing agents, incubated at 100°C for 5 minutes, followed by SDS-PAGE using a 6–16% gradient gel. The proteins were transferred to a nitrocellulose membrane and the immunoblotting was performed using anti-rabbit Mucin II antibodies (1: 500 in PBS- 5% BSA).
Parasites were lysed by sonication using 100 μL RIPA buffer, containing 5 mM NaF, 2 mM Na3VO4, 50 μM Na β-Glicerophosphate, 1 mM PMSF and protease inhibitor cocktail (Sigma). Samples were diluted in 1.4 mL of 20 mM Tris-HCl buffer, pH 7.4, containing the same concentration of inhibitors. After centrifugation at 14,000 x G for 10 minutes at 4°C, the supernatant was added to protein A- Agarose with the desired antibody or normal serum (control) or to covalent-linked antibody-Agarose (anti-nitro-tyrosine-resin). After 16 h at 12°C, the beads were washed with 20 mM Tris-HCl buffer, pH 7.4 containing 0.1% Triton X-100 and the bound material eluted with Laemmli buffer without reducing agent. In the particular case of immunoprecipitation using covalent-linked anti-nitro-tyrosine antibodies, the elution was performed by incubation in 5 M acetic acid for 5 minutes. The eluted proteins were identified by a commercial facility (Veritas Life Sciences, Brasil), as described above.
It was previously shown that T. cruzi is able to produce •NO via a putative tcNOS [34] although no NOS ortholog can be found on the parasite genome. The synthesis of •NO was then first confirmed in T. cruzi extracts of the non infective (epimastigote) and infective (trypomastigote) stages by measuring the conversion of 3H-L-arginine to 3H-L-citrulline by thin layer chromatography, as described [34] and the product of the reaction confirmed by capillary electrophoresis. Importantly, approximately 260 fold-enrichment in NOS activity was obtained after partial purification of the enzyme from epimastigotes, as described [34].
Even though the identity of the enzyme remains elusive, the possible involvement of •NO signaling during T. cruzi binding to ECM was pursued. Trypomastigotes were incubated with purified ECM for to 2 h and the amount of extracellular •NO was quantified (Fig. 1A). The extracellular •NO concentration dropped 43% under this condition, in the same order of magnitude observed when trypomastigote extracts were incubated with 10 mM L-NAME, a NOS inhibitor (Fig. 1A). The simultaneous incubation with ECM and L-NAME leads to an even higher inhibition of •NO production (approximately 72%). The data strongly suggest that interaction of the parasites with ECM hinders •NO responses.
Accordingly, 37% decrease in NOS activity was observed in parasite extracts previously incubated with ECM, as compared to parasites incubated in the absence of ECM under the same experimental conditions (Fig. 1B). Partial or total inhibition of •NO production by 10 mM L-NAME or by boiling the cellular extracts for 10 minutes at 100°C, respectively, confirmed that the •NO measured is a product of an enzymatic activity (Fig. 1B). The enzymatic activity was reduced to 9% by the addition of 50 mM L-NAME. Furthermore, changes in L-arginine/L-citrulline ratio upon incubation with ECM strengthen the evidence of declining NOS activity upon parasite adhesion to ECM (Fig. 1C, D). Whereas Intracellular concentration of L-citrulline decreased 83% upon adhesion of trypomastigotes to ECM, no significant change was observed in the L-arginine levels. This could be attributed to the contribution of other metabolic routes, but it is important to note that T. cruzi lacks a pathway to convert citrulline to arginine (i.e. arginase, an enzyme of the urea cycle, is absent) [56], which strongly suggests that the decline in the L-citrulline levels might be, at least in part, a consequence of NOS activity inhibition.
Since biological signaling by •NO is primarily mediated by activation of guanylyl cyclase, the production of cGMP in trypomastigotes incubated or not with ECM was quantified (Fig. 2). The levels of cGMP production fell from 3.5 to 0.6 fmoles after adhesion of the parasite to ECM (Fig. 2). Taken together, the findings strongly suggest that parasite adhesion to ECM leads to inhibition in •NO production, consequently deactivating a classical •NO signaling pathway.
To check whether parasite adhesion to ECM would modulate protein S-nitrosylation (SNO) and tyrosine nitration, immunological assays were performed employing anti-S-nitroso-cysteine and anti-3-nitro-tyrosine antibodies. Immunoblotting experiments reveal a time-dependent decrease of SNO in specific bands, mainly in the 37 kDa region, but also noticeable in protein bands at the 47, 20, 18, 15 and 13 kDa regions (Fig. 3).
Differently from SNO, the number of nitrated-proteins detected by anti-3-nitro-tyrosine was considerably less and differences in tyrosine-nitrated proteins were not significant at the first hour of the experiment (Fig. 4). However, the levels of tyrosine-nitrated proteins were extensively reduced at 2 h incubation, affecting proteins in the range of 10 to 37 kDa (Fig. 4).
Likewise, the general decrease in SNO and nitrated-proteins can be observed by immunofluorescence microscopy. Paraformaldehyde-fixed T. cruzi trypomastigotes previously incubated with ECM for 2 h showed a significant decrease in the immune reaction for both S-nitrosylation (Fig. 5) and tyrosine nitration of proteins (Fig. 6). Indeed, image pixels/spots quantification showed a reduction higher than 50% and around 30% in the immunoreaction for S-nitrosylated and for tyrosine nitration proteins, respectively, when parasites were incubated with ECM.
In order to confirm the adhesion effect on protein S-nitrosylation, total S-nitrosylated proteins in T. cruzi extracts were quantified by the Saville-Griess method [53]. A pronounced decrease of 87% was observed in the total SNO trypomastigote proteins when parasites were incubated with ECM, as compared to the control (Fig. 7A). As additional controls, parasites were also incubated in the presence or absence of 100 μM CysNO (•NO donor) or 100 μM cPTio (•NO scavenger). As expected, increasing •NO availability led to an improvement in SNO, as well removing •NO from the system resulted in SNO decrease (Fig. 7A). Also, addition of CYsNO to parasites incubated with ECM did not restore the levels observed for trypomastigotes incubated with CYsNO only, showing the predominance of the ECM effect. On the other hand, cPTio added to ECM-treated parasites reduced even more the amount of S-nytrosylated proteins. The different treatments did not affect parasite viability (Fig. 7B).
Preliminary data using the biotin-switch technique and mass spectrometry further confirmed the presence of S-nitrosylated proteins in T. cruzi. Although a low number of SNO proteins were detected, an even lower amount was present in ECM-incubated trypomastigotes, as predicted by the experiments herein described.
Examples of putative modified proteins that have been identified under different conditions were: (1) calpain-like cysteine peptidase, retrotransposon hot spot protein, surface protease GP63, trans-sialidase and mucin TcMUCII, in both untreated and ECM-incubated parasites; (2) fucose kinase, glycerophosphate mutase and kinesin K39 only in ECM-incubated parasites; (3) DGF-1, fatty acid elongase and helicase only in untreated parasites. Additionally, 27 hypothetical S-nitrosylated proteins were detected in ECM-treated (7) or untreated (20) trypomastigotes. However, it must be emphasized that the presence or absence of a modification in a particular protein due to the incubation of the parasite with ECM needs validation in each case, due to the possibility of a nonspecific binding during the enrichment of the SNO proteins.
The existence of S-nitrosylation in T. cruzi proteins was validated for mucin TcMUCII. SNO proteins were converted to biotin-containing proteins, pulled-down by streptavidin beads as described in Materials and Methods, and the biotinylated-proteins were subjected to immunoblotting using anti-rabbit Mucin II antibodies (Fig. 8A). As a negative control, proteins prepared by the biotin switch method in the absence of ascorbate gave no reactivity with anti-mucin II antibodies (Fig. 8A). A significant increase in the level of S-nitrosylation was detected in ECM-treated trypomastigotes in relation to the untreated parasites (Fig. 8B) and taking into account the protein loaded in each case.
In relation to the tyrosine-nitrated modifications, immunoprecipitated proteins with anti-nitro-tyrosine antibodies were identified by nLC-MS/MS. A number of putative nitrated targets identified decreased in ECM-incubated trypomastigotes (Table 1). Hypothetical and ribosomal proteins comprise the majority of the sequences obtained in untreated trypomastigotes. Also, the majority of the identified proteins were detected in ECM- treated or untreated parasites.
To confirm this post-translational modification, histone 2A, histone 4B, enolase, alpha-tubulin, beta-tubulin and paraflagellar rod proteins (PAR) were selected. Modified histones and tubulins were detected herein (Table 1), enolase was included since nitrated-enolase was already described in the literature [57] and PAR was chosen as a negative control of the method. The mentioned proteins were immunoprecipitated with commercial specific antibodies (except for anti-PAR monoclonal antibody prepared in the laboratory [50]) followed by Western blot developed with anti-nitro-tyrosine antibodies. An increase in the nitrosylation levels of enolase and histones 2A and 4 were observed after the incubation with ECM (Fig. 9A, B). No changes in the nitration levels were observed when a similar experiment was performed using anti-alpha and beta-tubulin antibodies, while no reactivity was detected with paraflagellar proteins immunoprecipitated with anti-PAR monoclonal antibody (Fig. 9C, D). The results have shown that in spite of the general down regulation of protein S-nitrosylation and nitration upon incubation of the parasites with ECM, there is a specific response for each protein, including the stimulation of the nitration levels of enolase and histones 2A and 4B.
Taken together, the results herein presented strongly suggest that T. cruzi responds to the interaction with ECM by the involvement of •NO regulated pathways.
Nitric oxide is a key signaling molecule affecting many biological activities. Human parasites such as T. cruzi are exposed to the anti-parasitic •NO produced by the host but also to its own •NO. Since the interaction of T. cruzi trypomastigotes with ECM is an essential step during the infective process [2,48,49], the in vitro model in the absence of host cells has been used to study the role of •NO in the parasite response to the interaction. ECM is a very dynamic structure and its relevance in •NO signaling was shown, for example, by thrombospondin-1 inhibition [58] of the •NO pathway in vascular cells. Additionally, the relevance of ECM to T. cruzi signaling was previously shown by changes in parasite protein phosphorylation levels [50].
Endogenous nitric oxide is predominantly produced in T.cruzi by enzyme catalysis, probably by NOS, as described [34] since the addition of the inhibitor L-NAME drastically reduces •NO production (Fig. 1A, B). In addition to •NO, the reaction produces L-citrulline from the substrate L-arginine. Of note, a strong reduction of citrulline, but not of arginine, was measurable in ECM-incubated trypomastigotes, suggestive of an inhibited NOS activity, although other possibilities such as its utilization in another activated metabolic route could not be ruled out. Arginine, on the other hand, is a substrate for protein synthesis and a precursor of •NO and other important metabolites as phosphoarginine, an energy buffer synthesized in T. cruzi by arginine kinase. Due to its relevance, it was previously suggested that arginine concentration under different external conditions may be buffered by TcAAP3, a specific permease [59].
Adhesion of trypomastigotes to ECM resulted in a remarkable inhibition not only on NOS activity but also in •NO and cGMP concentrations (Fig. 1, 2). However, a direct correlation between •NO levels and cGMP concentrations is difficult to make since the amount of •NO that may activate cGMP synthesis in trypomastigotes is unknown, due to the fact that no typical guanylyl cyclase is present in the T. cruzi genome. Moreover, the putative cGMP synthetic activity of an ubiquitous adenylyl cyclase remains non characterized [37]. Additionally, cGMP concentrations would depend on its degradation by a soluble dual-specificity phosphodiesterase (TcrPDEC) [38,39,40,41]. Presumably, the downstream signal transmission may be dependent on a protein kinase A activated by cGMP, as described for T. brucei [45] and Leishmania [46].
The decrease in total nitration and S-nitrosylation levels of proteins as described here for ECM-incubated trypomastigotes probably reflects the lower level of nitrosative stress. Changes in S-nitrosylated proteins were easily noticed by Western blot experiments after 30 minutes incubation of the parasites with ECM, with a marked decrease after 2 h period time (Fig. 5) and confirmed by immunofluorescence and decrease in the total SNO measured (Figs. 5 and 7). S-nitrosylation of proteins is a key player in diverse biological functions of •NO and is associated with processes such as apoptosis [60] and regulation of numerous signaling pathways, for example, PKC [61] and MAPK [62]. Of interest, S-nitrosylation has been associated with activation and desensitization of the human soluble guanylyl cyclase that possesses 37 cysteine residues (review in [58]). To our knowledge, the only study describing S-nitrosylation in T. cruzi proteins used •NO-donors to investigate a possible role of the host derived •NO in the inhibition of cruzipain, a cysteine protease important for the parasite infection [33]. However, the relevance of S-nitrosylation in T. cruzi signaling was not further explored.
Here proteins putatively S-nitrosylated in normal conditions and after interaction of the T. cruzi with the extracellular matrix were analyzed for the first time, using mucin II to validate the modification (Fig. 8). A number of other interesting targets were identified including some proteins already described as S-nitrosylated with relevant modification in function, such as Dual Specificity Phosphatase (DUSP) [62], Serine-Threonine Protein Kinase [61] and HSP 90 [63]. Interestingly, phosphorylation levels in DUSP and Serine/Threonine Protein Kinase were also modified in ECM-incubated parasites [50], but the possibility that both modifications are somehow related, as happens in other cases, has not been addressed. The results described point out to a possible role of the S-nitrosylation in T.cruzi signaling pathways and will be further explored.
Tyrosine nitration is not as well understood as S-nitrosylation, although relevant processes seem to be modulated by this covalent modification, such as PKC signaling [30] and protein degradation [28]. The present study was able to identify some of the proteins previously described as potential nitration targets (Table 1), such as enoyl-CoA hydratase [64], glyceraldehyde-3-phosphate dehydrogenase [65], heat shock protein 70 [66] and histone 2A [67]. Furthermore, validation of the data for histone 2A, histone 4, enolase and tubulins was achieved (Fig. 8). In the literature, for example, nitration of histones was associated with the induction of autoimmunity in systemic lupus erythematosus and rheumatoid arthritis [67]. Remarkably, a large number of 40 and 60 S-ribosomal proteins were modified by nitration, but whether this modification affects protein expression in T. cruzi remains to be elucidated. Nitration of ribosomal proteins was also described in native and differentiated PC12 cells, but tryptophan was identified as the modified amino acid [11].
In summary, the present work was able to confirm previous claims on the existence of enzymatic NOS activity in T. cruzi and demonstrated that the •NO classic signaling pathway is greatly inhibited in the presence of ECM regarding the synthesis of •NO and cGMP. Furthermore, numerous possible S-nitrosylation and tyrosine nitration targets have been identified, with a total decrease in the level of modified proteins upon interaction of the parasite with ECM. However, in spite of general down regulation of protein S-nitrosylation and nitration, the increase in the S-nitrosylation level of mucin II or in the nitration of enolase or histones 2A and 4, in contrast to the constant nitration levels of alpha and beta-tubulins under any condition, point out to the specificity of the modification for each particular target. The biological relevance of each of these target modifications remains to be explored and may give clues to the function of each target in parasite internalization into host cells. In this regard it must be stressed that •NO availability appears to be essential for parasite motility [68]. Thus, it is tempting to speculate that adhesion of T. cruzi trypomastigotes to ECM, an obligatory path to reach the host cell, triggering decrease in •NO levels, may also decrease parasite motility, somehow facilitating its binding to the host cell plasma membrane prior to invasion.
|
10.1371/journal.pmed.1002095 | Cardiometabolic Risk Factor Changes Observed in Diabetes Prevention Programs in US Settings: A Systematic Review and Meta-analysis | The Diabetes Prevention Program (DPP) study showed that weight loss in high-risk adults lowered diabetes incidence and cardiovascular disease risk. No prior analyses have aggregated weight and cardiometabolic risk factor changes observed in studies implementing DPP interventions in nonresearch settings in the United States.
In this systematic review and meta-analysis, we pooled data from studies in the United States implementing DPP lifestyle modification programs (focused on modest [5%–7%] weight loss through ≥150 min of moderate physical activity per week and restriction of fat intake) in clinical, community, and online settings. We reported aggregated pre- and post-intervention weight and cardiometabolic risk factor changes (fasting blood glucose [FBG], glycosylated hemoglobin [HbA1c], systolic or diastolic blood pressure [SBP/DBP], total [TC] or HDL-cholesterol). We searched the MEDLINE, EMBASE, Cochrane Library, and Clinicaltrials.gov databases from January 1, 2003, to May 1, 2016. Two reviewers independently evaluated article eligibility and extracted data on study designs, populations enrolled, intervention program characteristics (duration, number of core and maintenance sessions), and outcomes. We used a random effects model to calculate summary estimates for each outcome and associated 95% confidence intervals (CI). To examine sources of heterogeneity, results were stratified according to the presence of maintenance sessions, risk level of participants (prediabetes or other), and intervention delivery personnel (lay or professional).
Forty-four studies that enrolled 8,995 participants met eligibility criteria. Participants had an average age of 50.8 years and body mass index (BMI) of 34.8 kg/m2, and 25.2% were male. On average, study follow-up was 9.3 mo (median 12.0) with a range of 1.5 to 36 months; programs offered a mean of 12.6 sessions, with mean participant attendance of 11.0 core sessions. Sixty percent of programs offered some form of post-core maintenance (either email or in person). Mean absolute changes observed were: weight -3.77 kg (95% CI: -4.55; -2.99), HbA1c -0.21% (-0.29; -0.13), FBG -2.40 mg/dL (-3.59; -1.21), SBP -4.29 mmHg (-5.73, -2.84), DBP -2.56 mmHg (-3.40, 1.71), HDL +0.85 mg/dL (-0.10, 1.60), and TC -5.34 mg/dL (-9.72, -0.97). Programs with a maintenance component achieved greater reductions in weight (additional -1.66kg) and FBG (additional -3.14 mg/dl).
Findings are subject to incomplete reporting and heterogeneity of studies included, and confounding because most included studies used pre-post study designs.
DPP lifestyle modification programs achieved clinically meaningful weight and cardiometabolic health improvements. Together, these data suggest that additional value is gained from these programs, reinforcing that they are likely very cost-effective.
| In the United States and in many other countries, rates of both obesity and diabetes continue to increase every year.
The US Diabetes Prevention Program (DPP) trial studied people with a high risk of diabetes and placed them in a program encouraging adoption of both a healthy diet and exercise, consisting of 16 core sessions with ongoing support afterwards.
The DPP trial showed that such a program could decrease the incidence of diabetes and that it helped people lose weight and decrease their cardiovascular risk factors.
This study was done to collect all of the available data from programs based on implementation of the DPP in different settings and to combine them into a large analysis to understand whether such programs were associated with weight, blood sugar, blood pressure, and cholesterol.
Our study was designed specifically to find out if the programs worked, which health parameters changed and which program characteristics were the most useful.
We performed a systematic review and meta-analysis by looking at four different databases of articles and selecting all of those which used principles from the DPP.
We identified 44 studies with a total of 8,995 participants who were enrolled in different versions of the DPP.
These programs differed in many ways from the original DPP trial—they were delivered in different settings, had different providers, and involved group sessions rather than individual sessions.
We found that even with these modifications, the programs were still associated with favorable changes in weight, blood pressure, cholesterol and blood sugar.
Programs based on the DPP have the potential to help a large number of people in the community to lose weight, decrease their risk of diabetes, and help improve other markers of health.
These programs could also be a cost-effective way to help prevent diabetes and help people live healthier lives.
| Diabetes currently affects approximately 9.3% of the United States population [1], and by 2050, its prevalence is expected to reach 25% [2]. Adults with diabetes have two to four times higher rates of death from heart disease or stroke, and they have medical expenses that are more than two times higher than those for people without diabetes [3–6]. The total annual economic burden associated with diabetes was US$245 billion in 2012, with US$176 billion incurred as direct medical expenditures [3]. In addition, 86 million US adults (35% of the population) have prediabetes [1], which puts them at over four times the risk of progressing to diabetes compared to those who are normoglycemic [7,8].
Large randomized controlled studies [9–11], including the US Diabetes Prevention Program (DPP) trial, have shown that intensive and structured lifestyle modification interventions in people with impaired glucose tolerance can lower the incidence of diabetes by 30%–58% compared to basic lifestyle advice [10–15]. Although primary prevention of diabetes through lifestyle changes is deemed cost-effective, the first-year cost of delivering the original DPP lifestyle intervention was prohibitive (US$1,399 per participant) [16–18]. In addition, lifestyle and cultural patterns vary significantly, across and even within communities, necessitating tailoring of interventions according to regional and ethnic differences to achieve effectiveness, acceptability, and sustainability [19]. To find acceptable, lower-cost alternatives to the resource-intensive DPP lifestyle interventions, a number of studies tested adaptations of DPP delivery in typical US clinics and communities, but still retained the DPP’s core principles of modest weight loss, calorie-restricted diets, and 150 min of moderate-intensity exercise per week; on average, these DPP lifestyle interventions were associated with meaningful pre-post weight loss of approximately 4% [20]. However, it remains unknown whether these nonresearch lifestyle intervention programs were associated with meaningful changes in glycemic markers (fasting blood glucose [FBG] and glycosylated hemoglobin [HbA1c]), blood pressure (BP), and lipids (high density lipoprotein [HDL] and total cholesterol [TC]). Furthermore, there are no comparisons of how data from these translation or effectiveness studies compare to metabolic changes observed in the DPP efficacy trial itself [20,21]. It also remains unclear whether benefits of lifestyle modification programs for diabetes prevention are equally beneficial across persons with objectively defined prediabetes versus those identified only through risk factors for diabetes (e.g., being overweight and hypertensive), and whether program characteristics (e.g., presence of a maintenance component or type of provider) are associated with better outcomes.
We systematically searched four electronic databases (MEDLINE, EMBASE, Cochrane Library, and Clinicaltrials.gov) for translation or effectiveness studies that tested delivery adaptations of DPP-lifestyle principles in the US and were published between January 1, 2003, and May 1, 2016. Including studies with a similar exposure (DPP-lifestyle principles), albeit via different delivery modalities, and similar outcome measurements allows for the heterogeneity in these studies to answer questions regarding external validity that the original study could not. Indeed, the focus of this work was to inform wider implementation and scaling of interventions in the US.
The search terms used are listed in S1 Table. We supplemented our searches by hand searching reference lists of included articles and other reviews of this topic [22].
We included studies if they met three eligibility criteria. First, each study needed to evaluate implementation of lifestyle intervention programs based on tested DPP principles in US settings. Second, the studies had to have reported pre- and post-intervention estimates for at least one of the following measures: weight, HbA1c, FBG (venous or capillary), systolic or diastolic BP (SBP, DBP), HDL, or TC. Finally, each eligible study had to include adults (age ≥18) at high risk of developing diabetes. To qualify as “high risk,” the target population could have either of the following criteria:
OR
OR
This systematic review and meta-analysis was focused on studies in the US that implemented specific principles tested in the DPP. The study needed to specifically state that it was DPP inspired, and have both an exercise and dietary component in its intervention. We excluded studies with children, adolescents, or lifestyle interventions that did not involve combined diet and exercise principles tested in the DPP trial. If participants had polycystic ovarian syndrome, current or recent pregnancy, or tested the use of medications (such as metformin) to prevent diabetes, these studies were excluded. If studies had a majority of prediabetes participants (greater than 50%), we included these studies in order to be able to include all available data.
Abstracts were reviewed independently by two authors (UM and AZ) who used the criteria above to determine study eligibility. All discrepancies were resolved by consensus with a third study author (MKA.) When necessary information was not reported in the study, authors of the original article were contacted for further details.
Data regarding study population characteristics, study designs, characteristics of interventions tested (number of core and maintenance sessions, duration of interventions, and follow up time periods), and on baseline and follow-up values for each outcome (weight, FBG, HbA1c, SBP, DBP, HDL, and TC) were extracted by one author (UM) from each eligible article.
The DPP trial findings established that standard health advice alone had a low level of efficacy in reducing diabetes incidence among overweight participants with biochemically confirmed impaired glucose tolerance and elevated fasting glucose [13]. Structured behavioral modification interventions, adherence to rigorous lifestyle principles, and associated weight loss were shown to be effective [27]. Unlike large trials such as the DPP, translation studies tend to be small and often use quasi-experimental designs (e.g., single group pre- and post- intervention or pre-post evaluations with control groups) in which random allocation and blinding are impossible [13]. To assess study quality and facilitate interpretation of the available literature, we applied a scoring system with three criteria, each contributing one point in the criteria adapted from those proposed by Juni et al. [28]. The first criterion assessed whether studies used any steps to minimize attrition bias by using an intention to treat analysis, achieving low attrition rates (≤20%), or by comparing characteristics of completers and noncompleters. The second criterion was to assign higher quality to studies that included a control group (randomized, matched, or unmatched comparison). The third criterion focused on whether the study reported on four or more of the following aspects of translating evidence: describing the process of designing the program, describing the enrollment process, documenting session attendance, reporting costs and/or resource inputs, documenting the training process or qualifications of personnel, or describing the qualitative feedback from participants or providers. The latter was considered an important aspect of study quality given that studies were included based on their ability to provide information about how DPP lifestyle programs were implemented and how successful they were. These quality assessment criteria were chosen because they support study replication and comparison. A study was categorized as “high quality” if it had at least two out of three possible points. Further details are provided in the S3 Table.
Using pooled data, descriptive characteristics of the study populations were calculated as weighted means based on sample size. Since very few of the studies had comparison groups, the intervention arms of controlled studies were treated as pre-post groups and aggregated with the remaining single-group pre-post studies. When studies included multiple intervention arms, each distinct intervention arm was treated as a separate pre-post group. For example, in studies in which a lifestyle intervention was delivered using two different media (in person versus remote), these counted as two separate intervention groups within the same study.
Estimates of pre-post intervention change in each outcome and the corresponding standard errors were obtained or calculated based on the data reported. To standardize the length of follow-up, we focused on the 1 y interval between pre- and post- intervention measures. If a study conducted several follow-up assessments for a given outcome, the reported value at the time point closest to 1 y was used to calculate the change from the baseline. We used random effects meta-analysis techniques to calculate aggregate estimates and measures of dispersion to account for inter-study heterogeneity. All estimates were accompanied by a χ2 test for heterogeneity with a corresponding I2 value.
To explore what program characteristics may have been associated with risk factor changes, we performed stratified analyses according to delivery format (individual or group), intervention delivery personnel (by health professionals, lay community center staff, or electronic media), location of the intervention (clinical or community setting), and inclusion of a maintenance component in the study protocol (yes or no). When there was more than one type of delivery format, the delivery format of the core sessions took precedence. For example, if an article had core sessions in the community and a remote component for maintenance, this was coded as “community.” Stratified analyses were conducted for each outcome, except HbA1c, for which data were insufficient to allow stratification. We also stratified by the method used for risk classification (blood glucose criteria or risk factor criteria to classify as “high-risk”). We tested whether any program characteristics were associated with weight change using Spearman’s rank correlation analyses. Program characteristics of interest, in addition to those listed above, included the mean number and duration of core sessions offered, average participant age and BMI, proportions of males and non-Hispanic whites (NHWs) in the study sample, and attrition. Lastly, we evaluated aggregate outcomes in high versus average quality studies.
When studies included a control arm, we included them in separate analyses of the intervention arms against the control arms to determine what incremental changes in weight and cardiometabolic risks was observed.
All calculations were carried out using MIX 2.0 statistical software [29]. Statistically significant differences were noted based on a conservative definition of non-overlapping 95% confidence intervals. All findings were reported in reference to the 1 y findings reported by the original DPP trial [13].
A total of 44 studies, which included 48 intervention groups, met eligibility for inclusion. Of these, 22 studies were single group pre-post designs [30–51], 3 studies had two intervention arms (each contributing two separate groups for analysis) [52–54], 18 studies had separate control arms [55–72], and 1 study had three arms—two intervention groups and one control [73]. Data on weight change from baseline were available from all intervention groups. There were 21 groups that had pre- and post-intervention data on FBG, 8 had HbA1c measures, 23 had SBP measures, 22 had DBP measures, 14 had HDL measures, and 12 had TC measures. Further details are listed in Fig 1.
A total of 8,995 participants were enrolled across all intervention arms. Aggregate participant demographic and clinical characteristics were similar to participants enrolled in the original DPP study (Table 1). At enrollment, participants’ mean age was 50.8 y, mean weight was 99.3 kg, mean BMI was 34.8 kg/m2, 25.2% were male, and 32.9% were non-Hispanic white. Mean baseline levels of cardiovascular risk factors were: 5.9% (HbA1c), 104.6 mg/dl (FBG), 128.7/79.5 mmHg (SBP/DBP), 46.1 mg/dl (HDL), and 183.7 mg/dl (TC). Of the 48 intervention groups, 18 used blood glucose measures to classify risk for diabetes, 22 used the presence of risk factors, and 8 included participants defined as high risk by either of the above criteria.
Programs amended the original DPP lifestyle intervention (Table 1) by changing the number or duration of core sessions offered (the DPP offered 16 in-person sessions over 24 wk), conducting group (instead of individual) sessions, modifying the type of lifestyle coach (DPP coaches were qualified dietitians, exercise physiologists, behavioral psychologists, or health educators), and changing or removing the monthly maintenance component where participants met or were contacted for the remainder of the follow-up period to promote continued adherence to healthy lifestyle principles.
The number of core sessions offered in included DPP-lifestyle programs ranged from 1 to 24, with a mean of 12.6 core sessions (median 12.0) offered and 11.0 mean core sessions attended (median 10.1). Thirty interventions incorporated a maintenance component, which varied from emails to intermittent in-person group sessions. Programs with scheduled maintenance ranged from 3–8 monthly sessions as follow-up after the initial core component. Mean study duration was 9.3 mo with a standard deviation (SD) of 5.5 mo and a range from 3 to 15 mo (median 12.0) Across all studies, overall attrition was 23.5% (range: 0.0 to 43.2).
Across all studies, mean absolute pre-post weight change was -3.77 kg (95% CI: -4.55 to 2.99, I2 of 99.06% (95% CI: 98.96, 99.15) (Table 2). Aggregate data from 16 studies that randomly assigned participants to intervention or control groups showed 2.66 kg greater weight loss among intervention participants (-3.25 kg) compared to control participants (-0.59 kg). For further details, see forest plot in S13 Fig.
Among 8 studies that measured HbA1c, mean pre-post change was -0.21% (95% CI: -0.29; -0.13, I2 82.72% [95% CI: 67.29%, 90.87.]) Among 21 studies reporting FBG, mean change in FBG was -2.40 mg/dl (95% CI: -3.59; -1.21, I2 90.63% [95% CI: 87.08, 93.21]). Among the 23 studies evaluating BP, we observed a mean pre-post change in SBP of -4.29 mmHg (95% CI:-5.73, -2.84, I2 75.40% [95% CI: 61.15, 84.43]) and -2.56 mmHg in DBP (95% CI:-3.40, 1.71, I2 57.96% [95% CI: 29.09, 75.08]). There were 14 and 12 studies which reported on HDL and total cholesterol, respectively. There was an overall pre-post increase in HDL of +0.85 mg/dL (95% CI: -0.10, 1.60, I2 63.12% [95% CI: 34.41, 79.26]) and change in TC of -5.34 mg/dl (95% CI: -9.72, -0.97, I2 56.09% [95% CI: 16.19, 76.99]). There were insufficient data to perform between-group comparisons (intervention versus control) for outcomes other than weight.
Using our conservative definition of non overlapping CIs, studies with a maintenance component had a statistically significantly greater decrease in mean FBG (-3.14 mg/dl) and a greater decrease in weight (-1.66 kg) than intervention programs without a maintenance component (Table 3).
No statistically or clinically significant differences in risk factor changes were observed when comparing studies testing interventions delivered by community workers to studies that employed health professionals, or those that used electronic media (Table 4). Similarly, no outcome differences were noted in studies classifying high-risk for diabetes based on blood glucose testing versus other criteria (Table 5), nor by study quality, (high versus average quality) or setting (clinic, community or remote). (Data shown in S7, S11 and S10 Figs).
This is the first meta-analysis to aggregate both weight and cardiovascular risk factor changes from US community-based studies of DPP-based lifestyle interventions. Characteristics of participants in these studies of DPP lifestyle programs were very similar to those of the original trial participants, but translation study participants had a slightly higher mean starting weight and higher proportion of females [13]. The original DPP participants had a greater mean weight loss at 1 y than the participants in this meta-analysis (6.8 kg versus 3.8 kg), which was likely due to the more resource intensive intervention and individualized support in the trial [13]. However, this weight change was closer to the 4.2 kg weight loss reported in the Finnish Diabetes Prevention Study (DPS) [11]. Studies with adequate control groups showed an additional 1.9 kg weight lost across intervention arms when compared with their respective control arms (3.3 versus 0.6 kg). Indeed, the control groups in these effectiveness studies achieved some benefit from participation, even if only exposed to minimal intervention. Compared to the original DPP, HbA1c, and SBP reductions observed in translation studies were similar; FBG and DBP reductions were somewhat lower than the reductions achieved in the efficacy trial; and comparisons for HDL and TC were not possible.
We noted no difference in cardiometabolic risk factor changes in people with biochemically confirmed prediabetes versus those with diabetes risk factors. That said, progression to diabetes and its complications varies by type of prediabetes. The DPP enrolled patients with both IGT and IFG, who are at approximately three times higher risk of progression to diabetes compared to those with IFG alone [7]. The Finnish DPS, Malmo, and Da Qing studies also included participants with IGT or combined IGT and IFG, who are at higher risk than those with IFG alone [7,9,14]. Meanwhile, the US-based DPP-translation studies primarily used IFG criteria, and none used oral glucose tolerance tests to determine high-risk status. This suggests that participants in these studies had a lower risk profile, which was also reflected in their lower baseline FBG and HbA1c levels. Assuming the participants in this analysis were at a lower baseline risk, the changes in cardiometabolic risks observed were commensurate with the starting risk level, and are therefore still noteworthy. It remains unclear whether the DPP intervention is effective in preventing diabetes among participants with impaired fasting glucoses but normal post-load glucose levels. Also, given ease of testing, HbA1c is now commonly used to diagnose prediabetes, and it is unclear whether DPP results can be extended to this prediabetes population defined by HbA1c.
Our findings are also similar to other recent studies. A Community Guide Review, which evaluated interventions across diverse countries and settings, had similar decreases in FPG with a nonsignificant trend towards decreased blood pressure and cholesterol [22]. The MOVE! program evaluated a ten-module program among 238,000 veterans; high intensity intervention participants achieved 2.7% weight loss at 6 mo compared to a 0.6% weight loss in the low intensity group [13,74]. Our findings were also similar to a systematic review and meta-analysis that pooled 22 studies published before July 2012 that translated diabetes prevention for real-life settings in multiple countries (US, Australia, Europe, and Japan) and had ≥12 mo of follow-up [75]. Since multiple countries were involved, heterogeneous study interventions were benchmarked to Europe-wide diabetes prevention implementation guidelines and showed overall pre-post changes in weight (-2.32 kg), HbA1c (-0.13 mmol/mol), FPG (-0.10 mmol/L [-1.8 mg/dl]), SBP and DBP (-4.30/-4.28 mmHg), HDL (+0.01 mmol/L [+0.39 mg/dl]), and TC (-0.18 mmol/L [-6.96 mg/dl]); importantly, greater adherence to recommendations was associated with larger weight reduction. Our study expands on this work with a larger number of pooled studies and participants, all using a similar core set of intervention principles and comparison to the original DPP Study.
Effective translation of a program depends on multiple components, including referral, uptake, engagement, completion, and post-program sustainability of outcomes in the whole population. In our review, after eligibility criteria was applied, 25.5% of all eligible participants did not enroll; of those who enrolled, there was an additional 23.8% attrition. Rates of attrition also inherently select for those who are the most motivated participants, which biases the results towards effectiveness. This limits the generalizability of our findings, which more accurately apply to those who complete the program.
Implementation of DPP lifestyle programs have been limited by both cost and sustainability of ongoing program participation and risk factor reductions [76]. Most of the programs studied in this analysis provided free testing and intervention supplies but offered few additional incentives to encourage participation. The most common methods used to decrease cost were modifications to the intervention, such as offering the intervention in accessible locations, delivering the intervention through lay providers, and taking advantage of group classes and electronic delivery options. Over 80% of studies tested group interventions, most had fewer mean core sessions compared to DPP, and only 60% offered a maintenance component. Importantly, we noted similar risk factor benefits were achievable in interventions delivered by different providers in both group and individual formats. The similarities in weight loss and secondary outcomes compared to the DPP is encouraging for the ability to make the intervention cost-effective without sacrificing the effectiveness. With options that include group sessions, community-based programs with social support, cultural tailoring, and remote low-cost maintenance such as text messages or phone calls, the interventions allow for scaling to a wider audience.
Reach and sustainability of behavior change interventions remain, as do other challenges of implementing diabetes prevention. The advantages of community-based interventions that were pooled in this study include familiar context, peer support, and convenience to facilitate continued participation. The success of electronic and remote interventions is also encouraging, as these could be distributed nationally with ease. The option of pre-recorded workouts on in-home cable TV illustrates a low-cost method of delivery that does not necessitate travel and is available on demand, in contrast to on-site workout regimens for which participants pay to participate [77]. This preliminary analysis also suggests that programs that implemented a maintenance component after the completion of the core sessions had greater reductions in weight and fasting glucose. The duration and intensity of maintenance that is most effective and the utility after the 1 y mark is largely unknown. Further evaluation of types of maintenance programs following a year-long program would be helpful to understand long-term benefit and sustainability.
A key limitation of our analysis was the heterogeneity of the studies included, which is inherent in all meta-analyses. Differences in duration of follow-up (from 1.5 to 36 mo), location of delivery, and other delivery format adaptations of the original DPP program were the most likely sources of heterogeneity. However, as the intervention (DPP-lifestyle program principles) and outcomes were similar, this study adds to the literature by providing external validity and noteworthy pre-post and between group cardiometabolic risk factor changes.
The lack of statistical significance found in most of the stratified analyses is likely due to lack of statistical power, which resulted in large, overlapping confidence intervals, as well as our conservative definition of statistical significance based only on non-overlapping CI’s. As most meta-analyses, our study is confined to the use of previously reported results. We used a more conservative definition of statistical significance by comparing stratum specific results, though this did not allow a more consistent adjustment for confounders. However, a less conservative analytical approach may have found other program characteristics that had “statistically significant” associations with cardiometabolic changes.
Additionally, studies varied significantly in quality. We conducted a sensitivity analysis to evaluate change in weight stratified by study quality (high versus average quality), which showed no significant difference. However, this alone is not expected to account for variation that arises during recruitment, enrollment, and study conduct. The majority of studies in our review used pre-post single group study designs and may be subject to confounding. To address this, we separately examined studies that had control groups and demonstrated that intervention groups achieved larger benefits than control groups.
Delivery of lifestyle programs adhering to DPP principles tested in community and clinical settings achieved similar 1 y decreases in weight, FBG, and HbA1c as the original DPP study, despite the modifications made to lower cost and improve acceptability across various settings. Though unclear if these changes truly translate into reductions in diabetes incidence, prior studies have found decreased incidence to be most closely related to weight loss [13]. Methods to increase uptake and decrease attrition are both needed to enable long-lasting, sustainable lifestyle change in patients with the highest risk of progression to diabetes and its associated complications.
The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the US Centers for Disease Control and Prevention.
|
10.1371/journal.pntd.0003552 | Lipid-Free Antigen B Subunits from Echinococcus granulosus: Oligomerization, Ligand Binding, and Membrane Interaction Properties | The hydatid disease parasite Echinococcus granulosus has a restricted lipid metabolism, and needs to harvest essential lipids from the host. Antigen B (EgAgB), an abundant lipoprotein of the larval stage (hydatid cyst), is thought to be important in lipid storage and transport. It contains a wide variety of lipid classes, from highly hydrophobic compounds to phospholipids. Its protein component belongs to the cestode-specific Hydrophobic Ligand Binding Protein family, which includes five 8-kDa isoforms encoded by a multigene family (EgAgB1-EgAgB5). How lipid and protein components are assembled into EgAgB particles remains unknown. EgAgB apolipoproteins self-associate into large oligomers, but the functional contribution of lipids to oligomerization is uncertain. Furthermore, binding of fatty acids to some EgAgB subunits has been reported, but their ability to bind other lipids and transfer them to acceptor membranes has not been studied.
Lipid-free EgAgB subunits obtained by reverse-phase HPLC were used to analyse their oligomerization, ligand binding and membrane interaction properties. Size exclusion chromatography and cross-linking experiments showed that EgAgB8/2 and EgAgB8/3 can self-associate, suggesting that lipids are not required for oligomerization. Furthermore, using fluorescent probes, both subunits were found to bind fatty acids, but not cholesterol analogues. Analysis of fatty acid transfer to phospholipid vesicles demonstrated that EgAgB8/2 and EgAgB8/3 are potentially capable of transferring fatty acids to membranes, and that the efficiency of transfer is dependent on the surface charge of the vesicles.
We show that EgAgB apolipoproteins can oligomerize in the absence of lipids, and can bind and transfer fatty acids to phospholipid membranes. Since imported fatty acids are essential for Echinococcus granulosus, these findings provide a mechanism whereby EgAgB could engage in lipid acquisition and/or transport between parasite tissues. These results may therefore indicate vulnerabilities open to targeting by new types of drugs for hydatidosis therapy.
| Echinococcus granulosus is a causative agent of hydatidosis, a parasitic disease that affects humans and livestock with significant economic and public health impact worldwide. Antigen B (EgAgB), an abundant product of E. granulosus larvae, is a lipoprotein that carries a wide variety of lipids, including fatty acids and cholesterol. As E. granulosus is unable to synthesize these lipids, EgAgB likely plays an important role in parasite metabolism, participating in both the acquisition of host lipids and their distribution between parasite tissues. The protein component of EgAgB consists of 8 kDa subunits encoded by separate genes. However, the biochemical properties of EgAgB subunits, particularly their ability to bind and transfer lipids, are poorly known. Herein, using in vitro assays, we found that EgAgB subunits were capable of oligomerizing in the absence of lipids and to bind fatty acids, but not cholesterol. Moreover, EgAgB subunits showed the ability to transfer fatty acids to artificial phospholipid membranes. These results indicate new points of attack at which the parasite might be vulnerable to drugs.
| Cystic echinococcosis (CE), one of two major types of hydatid disease, is a worldwide zoonosis caused by the larval stage (metacestode) of Echinococcus granulosus sensu lato (E. granulosus s.l.), which includes a series of species traditionally considered to comprise different strains or genotypes of E. granulosus [1,2]. The larva forms unilocular bladder-like cysts (referred to as hydatid cysts) that establish and gradually grow within the viscera (mainly liver and lungs) of a wide range of mammalian species (mainly domestic ungulates) as well as humans [1]. CE is considered a chronic, complex and neglected disease, which is re-emerging as an important public health problem [3–6]. For many years surgery was considered the only effective therapy, although it is not recommended for patients with cysts disseminated into different organs, and had a relatively high morbidity, relapse and mortality rates [7,8]. Currently, the advent of antihelminthic drugs (benzimidazole carbamates, mainly mebendazole and albendazol) has led to an alternative therapy which comprises pre- and post-operative chemotherapy, combined with percutaneous drainage of hydatid cysts (a procedure known as PAIR for puncture, aspiration, injection, reaspiration) [7,8]. In comparison with surgery this approach showed greater clinical and anti-parasitic efficacy (lower rates of morbidity, mortality, and disease recurrence and shorter hospital stays [9]). In addition, antihelminthic drugs are chosen for the treatment of uncomplicated cysts, as well as for long-term post-surgical treatment [8]. Benzimidazoles bind to β-tubulin and interfere with microtubule formation, thus affecting motility, cell division, secretion processes, as well as perturbing the uptake of glucose by helminths [10,11]. Nevertheless, benzimidazole treatment has shown limited efficacy against large cysts and the occurrence of side-effects has also been reported [12]. Therefore, the development of novel drugs against E. granulosus s.l. therapy is required. Advancing knowledge of parasite biology would facilitate the identification of new drug targets for a more specific CE therapy.
Antigen B (EgAgB) is an abundant lipoprotein of hydatid cyst fluid that has been postulated as a carrier of essential lipids for E. granulosus s.l. [13–16]. This is based on the fact that cestodes have lost both degradative and biosynthetic pathways for common fatty acids and sterols [17,18], and EgAgB contains a heterogeneous mixture of lipids including free and esterified fatty acids and sterols [15]. Thus, EgAgB may be of foremost importance for parasite lipid metabolism, representing an interesting target for chemotherapy.
EgAgB is an alpha helix-rich 230 kDa lipoprotein [15], which has been considered to be the most specific Echinococcus-genus antigen for human serodiagnosis of hydatid disease [19–21] as well as being an immunomodulatory parasite component [22,23]. It belongs to a cestode-specific family of proteins that bind hydrophobic ligands, referred to as hydrophobic ligand binding proteins (HLBPs). The ligands present in the EgAgB complex account for approximately 40–50% of the total mass of the native antigen and consist of a variety of neutral (mainly triacylglycerides, sterols and sterol esters) and polar (mainly phosphatidylcholine) lipids [15]. At the protein level, native EgAgB contains around a dozen apolipoproteins or subunits [15], which are encoded by a polymorphic multigene family comprising five clades named EgAgB1 to EgAgB5 [24–29]. Interestingly, EgAgB isoforms are expressed differentially during the life-cycle stages of the parasite, as well as within distinct tissues of a given developmental stage; EgAgB1 to EgAgB4 are expressed in the metacestode stage whereas EgAgB5 seems to be expressed in the adult stage. Furthermore, in the metacestode, EgAgB1 to EgAgB4 are expressed in the germinal layer, but EgAgB3 seems to be the most abundantly expressed in protoscoleces [29]. Similar evidence of differential expression of antigen B subunits has also been obtained for the closely related species E. multilocularis [30]. The proteins encoded by EgAgB genes are each approximately 8 kDa in mass, and the different isoforms designated EgAgB8/1 to EgAgB8/5. Comparison of their amino acid sequences shows that EgAgB8/1, EgAgB8/3 and EgAgB8/5 are more similar to each other than to EgAgB8/2 and EgAgB8/4 (Fig. 1).
One of the features of EgAgB subunits is their ability to self-associate into large complexes. Analysis of native EgAgB from hydatid cyst fluid showed oligomers of 16, 24 and 32 kDa that are built from the 8 kDa subunits [31], as found in SDS-PAGE analysis under reducing conditions [32]. Recombinant subunits of EgAgB8/1, EgAgB8/2 and EgAgB8/3 are also capable of self-associating into oligomers of 16 and 24 kDa, and also into high-order oligomers of more than 100 kDa, estimated by size exclusion chromatography [33] or native polyacrylamide gel electrophoresis [34]. These results indicate that recombinant EgAgB subunits share structural features with native EgAgB, which agrees with the fact that EgAgB subunits present an electrostatic profile compatible with molecular aggregation [16,33] and suggest that lipids may not be indispensable for oligomerization. Nevertheless, the contribution of the lipid moiety to the oligomerization process has not yet been formally considered.
As already mentioned, EgAgB particles contain a mixture of lipid classes ranging from highly hydrophobic lipids to a variety of phospholipids [15]. The lipid binding properties of different members of EgAgB family have been partially characterised and constitute a piece of information that may contribute to elucidate EgAgB functions. EgAgB8/1 and EgAgB8/2 ligand binding properties were analysed by Chemale and collaborators, who used recombinant subunits (delipidated using the hydrophobic resin Lipidex 1000) and fluorescent probes to examine their interaction with fatty acids [35]. They observed that both subunits were capable of binding the fatty acid analogue 16-AP (16-(9-anthroyloxy) palmitate), but not the fluorescent probes DAUDA (11-(dansylamino) undecanoic acid), ANS (1-anilinonaphthalene-8-sulfonic acid) or DACA (dansyl-DL-alpha-aminocaprylic acid). In these studies, EgAgB8/1 and EgAgB8/2 were found to bind fatty acids with similar affinity. However, differences in the ability to bind lipids among other EgAgB members could not be ruled out. In fact, the 150 kDa HLBP entity of Taenia solium (TsM150) contains two classes of HLBP subunits (7 and 10 kDa members) which are capable of binding different lipid probes; members of the 7kDa-TsM150 subfamily bind DAUDA, ANS and DACA, but not 16-AP, whereas members of the 10kDa-TsM150 subfamily only bind 16-AP [36]. Furthermore, analysis of the ability of EgAgB subunits to bind lipid classes different from fatty acids has not been analysed, which, considering the biochemical composition of native EgAgB-associated lipids [15], is a potentially serious gap in our knowledge.
As previously stated, the fact that EgAgB carries lipids that are essential to E. granulosus s. l. suggests a role for EgAgB in the uptake and delivery of these lipids [37]. If so, then EgAgB would be expected to transfer ligands between donor and acceptor membranes and/or membrane-embedded receptor proteins. The ability of EgAgB to transfer ligands or to interact with membranes has not yet been examined in detail.
In the present work, we report a novel method of preparing lipid-free EgAgB8 subunits, which allowed the proper analysis of their oligomerization capacity and lipid binding properties. For this purpose, we purified the recombinant subunits EgAgB8/2 and EgAgB8/3 as representatives of the two distinct types of subfamilies within the multigenic EgAgB family (Fig. 1), and then removed their co-purifying ligands (mainly phospholipids) by reverse-phase high performance liquid chromatography (RP-HPLC). We found that lipid-free EgAgB8/2 and EgAgB8/3 subunits were able to self-associate into larger oligomers, suggesting that lipids are not indispensable for oligomerization. Regarding their lipid binding properties, both subunits bound a stearic acid anthroyloxy-derivative with similar affinity, but not a fluorescent cholesterol analogue. Furthermore, using small unilamellar vesicles we found that EgAgB8/2 and EgAgB8/3 subunits are potentially capable of transferring fatty acids analogues to phospholipid membranes employing different mechanisms, in which electrostatic interactions might play an important role. Overall, these results indicate previously unsuspected features of EgAgB particle assembly and suggest interaction with cell membranes (possibly of both parasite and host) that present potentially new classes of therapeutic target.
Inorganic salts were acquired from Sigma Chemicals (USA), Merck (Germany) or Carlo Erba (France). Organic solvents were purchased from JT Baker (USA), Merck (Germany) or Carlo Erba (France). For purification and delipidation of recombinant EgAgB8 subunits Gluthathione Sepharose 4B resin was obtained from GE Healthcare Life Sciences (Sweden), reduced glutathione and thrombin from human plasma were purchased from Sigma Chemicals (USA) and C8-bonded silica column was purchased from Vydac (USA). For lipid analysis silica TLC plates (20 x 20 cm) were obtained from Merck (Germany). For size exclusion chromatography (SEC) experiments, Superdex 200 HR 10/30 column was from GE Healthcare Life Sciences (Sweden), bovine serum albumin (BSA), carbonic anhydrase and cytochrome c were acquired from Sigma Chemicals (USA). For crosslinking experiments, N-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC) was purchased from Sigma Chemicals (USA). For ligand binding assays, fatty acid analogues 12-(9-anthroyloxy) stearic acid (12-AS) and DAUDA were obtained from Molecular Probes (USA), whereas cholesterol analogue dehydroergosterol (DHE) was obtained from Sigma Chemicals (USA). For ligand transfer assays, egg phosphatidylcholine (EPC), brain phosphatidylserine (PS), heart cardiolipin (CL) and N-(7-nitro-2,1,3-benzoxadiazol-4-yl)-phosphatidylcholine (NBD-PC),) were purchased from Avanti Polar Lipids (USA).
Genes encoding EgAgB8/2 and EgAgB8/3 subunits subcloned into pGEX plasmids were generously donated by Dr. Arnaldo Zaha (Federal University of Rio Grande do Sul, Brazil). EgAgB8/2 and EgAgB8/3 gene sequences were confirmed employing Macrogen sequencing facility (Macrogen Inc., Korea). EgAgB recombinant subunits EgAgB8/2 and EgAgB8/3 were expressed in Escherichia coli BL21 Codon Plus pRIL, as glutathione S-transferase (GST) fusion proteins, purified by affinity chromatography on immobilized glutathione and recovered by thrombin cleavage as previously described [38]. The removal of hydrophobic ligands derived from the bacterial expression system was achieved by RP-HPLC in a HPLC System (Merck-Hitachi, Japan) with a C8-bonded silica as stationary phase and water/acetonitrile/trifluoroacetic acid mobile phase, based on a procedure described by Meenan and collaborators to delipidate other lipid binding proteins (LBPs) [39]. After elution and freeze drying, proteins were refolded in a large volume of phosphate buffered saline, pH 7.4 (PBS) and then concentrated using centrifugal filter units (Millipore EMD). Proper delipidation of recombinant subunits was controlled by analysing the lipid content of protein fractions subjected and not-subjected to RP-HPLC method. Lipids of pre and post-HPLC fractions were extracted using the Folch method [40], analysed by thin layer chromatography (TLC) and compared with standards, as previously described for the analysis of the lipid moiety of native EgAgB [15]. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE, 15%) followed by Coomassie blue staining was used to assess subunits purity [41]. Protein concentration was estimated by measuring the absorbance at 280 nm, employing molar extinction coefficients of 7030 and 1340 M−1cm−1 for EgAgB8/2 and EgAgB8/3, respectively (calculated from their amino acid sequences employing Biology Workbench 3.2 free software [Computational Biology Group, Department of Bioengineering, University of California, San Diego]).
Lipid-free EgAgB8 subunits were analysed by circular dichroism (CD) spectroscopy to examine their structure. CD spectra of EgAgB8/2 and EgAgB8/3 (30 μM) at 25°C in PBS were recorded on a Jasco J-810 spectropolarimeter. Data in the Far-UV (195–250 nm) region were collected in 1-mm path cuvettes using a scan speed of 20nm/min with a time constant of 1 second. Molar ellipticity [θ] (deg cm2 dmol−1) was calculated as described elsewhere [42]. Secondary structure calculations of EgAgB subunits were undertaken employing k2d program (http://kal-el.ugr.es/k2d/k2d.html) [43,44].
SEC experiments were carried out in an Âkta FPLC System (GE Healthcare Life Sciences, Sweden). Briefly, 100 μL of lipid-free EgAgB8/2 (324 μM) or EgAgB8/3 (320 μM) were loaded separately on a Superdex 200 HR 10/30 column equilibrated in PBS. A flow rate of 0.5 mL/min was used, and elution profiles were recorded following the UV absorption at 215 and 280 nm. The column was calibrated using bovine serum albumin (66 kDa), carbonic anhydrase (29 kDa) and cytochrome c (12 kDa) as protein standards under the same chromatography conditions. Samples, as well as standard proteins, were analysed twice under the same conditions. Molecular weight estimation of the proteins was undertaken as described previously [45]. In order to investigate the stability of the oligomeric EgAgB complexes, serial dilutions of the proteins were also loaded under the same conditions.
Cross-linking with EDC was carried out by adding a fresh solution of EDC to 30 μM lipid-free EgAgB8 subunits in PBS up to a final concentration of 5, 10 and 20 mM EDC. Controls without adding EDC were carried out under the same conditions at the same time. The mixture was incubated for 30 minutes at 25°C under continuous stirring. The reaction was stopped by adding SDS-PAGE sample buffer, the reaction products were analysed on a 15% SDS-PAGE followed by silver staining. In parallel, non-delipidated recombinant EgAgB8 subunits, and native EgAgB (purified from hydatid cysts as previously described [15]) were used for comparison. In addition, cross-linking experiments were performed using a non-related lipid binding protein (ABA-1-A1 from Ascaris suum) as a control.
Fluorescence measurements were performed at 25°C in a Fluorolog-3 Spectrofluorometer (Horiba-Jobin Yvon) using 2 mL samples in a quartz cuvette. Fluorescent probes DHE, DAUDA and 12-AS were used to determine lipid binding properties of EgAgB8/2 and EgAgB8/3. Briefly, 0.5 μM of these probes were incubated at 25°C for 3 min in buffer 40 mM Tris, 100 mM NaCl, pH 7.4 (TBS) with increasing concentrations of EgAgB8 subunits. Emission spectra were recorded at 350–515, 365–665 or 400–500 nm, employing an excitation wavelength of 325, 345 and 383 nm for DHE, DAUDA and 12-AS, respectively. For 12-AS titration curves, fluorescence data were fitted using SigmaPlot software and different fittings were tested. The “one site saturation ligand binding” model showed the best r values (r ∼ 0.99), and was thus selected to estimate the binding constants (Kd). Average values for three independent experiments are reported. Changes in the intrinsic tryptophan (Trp-derived) fluorescence of EgAgB8/2 (5 μM or 2 μM) were monitored upon the addition of oleic acid (0.5 to 6.5 μM) or cholesterol (0.7 to 19.4 μM). The emission spectra were recorded at 310–400 nm, employing an excitation wavelength of 295 nm.
Small unilamellar vesicles (SUV) were prepared by sonication and ultracentrifugation as described previously [46]. The standard vesicles were prepared to contain 90 mol % of EPC and 10 mol % of NBD-PC, which served as the fluorescent quencher of the anthroyloxy-derivative. To increase the negative charge density of the acceptor vesicles, either 25 mol % of PS or CL was incorporated into the SUVs replacing an equimolar amount of EPC. Vesicles were prepared in TBS except for SUV containing CL which were prepared in TBS with 1 mM EDTA.
The relative partition coefficient (Kp) of 12-AS between EgAgB and NBD-containing SUVs was determined employing a method described by Massey and collaborators [47]. The Kp was defined as:
Kp=[12ASEgAgB][EgAgB]×[SUV][12ASSUV]
(1)
where [12ASEgAgB] and [12ASsuv] are the concentrations of 12-AS bound to EgAgB and vesicles, respectively, and [EgAgB] and [SUV] are the concentrations of protein and vesicles. The Kp for 12-AS partitioning was determined by measuring 12-AS fluorescence (440 nm) at different molar ratios of SUV:EgAgB after addition of SUVs into a solution containing 7.5 μM EgAgB8/2 or EgAgB8/3 and 0.5 μM 12-AS in buffer TBS at 25°C. The Kp was calculated by fitting the equation described by De Gerónimo and collaborators [48] to our data, as follows:
Frel=a×KpKp+[SUV][EgAgB]−b×[SUV][EgAgB]
(2)
where Frel is the difference between the fluorescence of 12-AS at a given SUV:EgAgB ratio and the fluorescence of 12-AS with an excess of vesicles, relative to the maximum difference in 12-AS fluorescence; [SUV] and [EgAgB] are the molar concentrations of SUV and EgAgB, respectively; and “a” and “b” are the fitting parameters. The partition coefficient was used to establish the conditions, for the 12-AS transfer assay, that ensure essentially unidirectional transfer, as detailed below.
A Förster Resonance Energy Transfer (FRET) assay was used to monitor the transfer of 12-AS from EgAgB to acceptor model membranes as described previously for other LBPs [49–56]. All the transfer experiments were conducted employing a Stopped-Flow RX2000 module (Applied Photophysics) attached to the spectrofluorometer. Transfer assay conditions were established according to Kd and Kp values previously obtained. Kd was used to ensure low levels of unbound 12-AS (< 5%), and Kp to determine SUV:EgAgB molar ratio in order to assure unidirectional transfer to SUVs. EgAgB8/2 or EgAgB8/3 subunits with bound 12-AS were mixed with NBD-PC SUVs. The NBD moiety is an energy transfer acceptor of the anthroyloxy donor group, therefore the fluorescence of the anthroyloxy fatty acid (AOFA) is quenched when the ligand is bound to SUVs which contain NBD-PC. Upon mixing, transfer of AOFA from protein to membrane is directly monitored by the time dependent decrease in anthroyloxy fluorescence. Final transfer assay conditions were 15:1 mol:mol EgAgB:12-AS ratio. SUVs were added ranging from 1:10 mol:mol to 1:40 mol:mol EgAgB:SUVs in TBS buffer at 25°C. Controls to ensure that photobleaching was eliminated were performed prior to each experiment, as previously described [50]. To analyse the influence of membrane surface charge on fatty acid transfer rate, SUVs with 25% negatively charged phospholipids (PS or CL) were employed. Data were analysed employing SigmaPlot software and all curves were well described by a single exponential function. For each experimental condition, at least five replicates were performed. Average values for three separate experiments are reported. Statistical analysis of the data was performed applying one-way analysis of variance (ANOVA) followed by Tukey's Post Hoc Test from GraphPad Prism software.
GenBank accession numbers for EgAgB8 subunits: EgAgB8/1: AAD38373, EgAgB8/2: AAC47169, EgAgB8/3: AAK64236, EgAgB8/4: AAQ74958, EgAgB8/5: BAE94835.
We purified lipid-free EgAgB8 subunits in order to analyse their capacity to oligomerize, their lipid binding properties, as well as their ability to transfer fatty acids to membrane vesicles. Among the five distinct EgAgB8 subfamilies, EgAgB8/2 and EgAgB8/3 were chosen because they represent the two distinct subfamilies within EgAgB family as mentioned above (Fig. 1). The recombinant subunits rEgAgB8/2 and rEgAgB8/3 were purified as GST fusion proteins from E. coli and recovered after thrombin treatment, as previously described [38]. We anticipated that rEgAgB8/2 and rEgAgB8/3 subunits would bind lipids during their synthesis in E. coli. We therefore analysed by TLC the lipids recovered by Folch extraction from the purified recombinant subunits and compared them with those from E.coli. Fig. 2A shows that fatty acids and phospholipids, mainly phosphatidylethanolamine (PE) and cardiolipin (CL), are the major lipids present in E. coli under our extraction conditions. Extractions from both rEgAgB8 subunits yielded mainly PE and CL (Fig. 2B). The absence of phosphatidylcholine, the main phospholipid found in native EgAgB [15], may be due to the lack of this phospholipid in E. coli (Fig. 2A).
Our next step was to identify a method to efficiently remove bacterial ligands from the proteins. Despite published precedent [35], we found that chromatography with Lipidex 1000 is ineffective at removing lipid from helminth LBPs [39,57,58]. We therefore exploited a method for removal of co-purifying ligands from other recombinant LBPs produced in E. coli [39], based on RP-HPLC of unfolded protein using a C8-bonded silica and water/acetonitrile/trifluoroacetic acid as stationary and mobile phases, respectively. Both rEgAgB8/2 and rEgAgB8/3 subunits were found to bind to C8 column and to elute from this phase at about 70:30 (v/v) acetonitrile/water. After careful refolding employing a large volume of aqueous buffer, the efficacy of this delipidation procedure was assessed by comparing the lipid moiety of treated (post-HPLC) recombinant subunits compared to non delipidated (pre-HPLC) subunits. We found that RP-HPLC successfully removed E. coli ligands from EgAgB8 subunits (Fig. 2B).
In order to check the structural integrity of the proteins after delipidation, we analysed the CD spectra of lipid-free rEgAgB8 subunits after refolding. The spectra of both recombinant subunits presented two minima at 208 and 222 nm, and were consistent with predominantly alpha-helical structures (66% and 30% for rEgAgB8/2 and rEgAgB8/3, respectively) as is shown in Fig. 3. These results are similar to CD data obtained for non-delipidated EgAgB8 subunits (35–40% of alpha-helix [33]) and native EgAgB (between 42 and 65% [14,59]).
Overall, these results showed that the delipidation method based on RP-HPLC provided lipid-free subunits without significant alteration, at least of their secondary structures, permitting us to analyse the interactions of the protein components in isolation.
As said, since previous oligomerization studies were performed with non-delipidated proteins, the involvement of lipids in the oligomerization of rEgAgB subunits cannot be ruled out [33,34]. In order to examine whether lipid-free forms are capable of self-associating, we analysed rEgAgB8 subunits by SEC in aqueous solutions. This showed that lipid-free rEgAgB8/2 and rEgAgB8/3 were eluted in defined peaks that were not commensurate with monomeric forms, according to the elution of the standard (Fig. 4). The apparent molecular weight of these subunits indicated the presence of oligomers of 62 and 39 kDa for rEgAgB8/2 and rEgAgB8/3, equivalent to the formation of oligomers of 7–8 subunits and of 4–5 subunits, for lipid-free rEgAgB8/2 and rEgAgB8/3, respectively. Previous reports using non-delipidated recombinant proteins revealed the presence of larger complexes of around 164 kDa for EgAgB8/2 and 113 kDa for EgAgB8/3, as well as secondary peaks of higher molecular mass for both proteins [33]. Nevertheless, our results suggest that oligomerization of EgAgB subunits is not absolutely dependent on the presence of lipids.
In order to further examine the self-assembly of EgAgB8 apolipoproteins, we performed covalent cross-linking experiments analysing the formation of oligomers by SDS-PAGE under reducing conditions (Fig. 5A). We found that in the absence of EDC both subunits were predominantly present as monomers, and EDC addition led to the formation of different cross-linked products. In the range of 5 mM to 20 mM EDC, lipid-free rEgAgB8/2 formed mainly a ∼45 kDa oligomer, whereas lipid-free rEgAgB8/3 formed a heterogeneous array of oligomers (a ladder-like pattern from around 16 to more than 97 kDa). Cross-linked products of higher molecular mass were found for both apolipoproteins. These results suggest that rEgAgB8 subunits have an intrinsic ability to self-assemble, but with differences in the structural organization acquired by rEgAgB8/2 and rEgAgB8/3 in aqueous solution, since the oligomers that they form have different apparent molecular weights according to SEC and EDC cross-linking analysis. For comparison, we also analysed the cross-linked products of non-delipidated rEgAgB8 subunits to assess the role of lipids in EgAgB oligomerization. This showed that non-delipidated proteins form even larger oligomers of apparent MW around 60–90 kDa and higher than 97 kDa (Fig. 5B). These results agree with those previously described by Monteiro and collaborators, obtained using glutaraldehyde as cross-linker [33], and indicate that, whilst lipids are not absolutely required, they may participate in the formation of larger complexes. In parallel, the cross-linking of natural, parasite-derived EgAgB led to the formation of oligomers of high molecular mass, showing a similar pattern to non-delipidated recombinant subunits (Fig. 5C), also in agreement with previous reports [34]. Both, non-delipidated EgAgB subunits and parasite-derived EgAgB, presented larger complexes even in the absence of cross-linker (Fig. 5B and 5C), in contrast to lipid-free EgAgB subunits (Fig. 5A). This suggests that lipids are involved in the oligomerization of EgAgB that is commonly observed in SDS-PAGE analysis under reducing conditions [32]. As a control for cross-linking experiments, we used lipid-free ABA-1-A1 (a helix-rich small LBP from the nematode parasite Ascaris suum), a protein which is not expected to oligomerize even at high concentrations [39]. Both lipid-free and non-delipidated ABA-1-A1 were treated under the same conditions and oligomers were not detected (Fig. 5D).
It therefore appears that lipids are not essential for EgAgB subunit self-association, although they may participate in the oligomerization process contributing to the organization of very large EgAgB particles. rEgAgB8/2 and rEgAgB8/3 exhibited different behaviors, although lipid-free rEgAgB8/2 forms an oligomer of greater size than rEgAgB8/3, and rEgAgB8/3 appears to form more heterogeneous oligomers. These results are in concordance with previous findings on non-delipidated rEgAgB8/3 subunits, which formed the more heterogeneous oligomeric states in comparison with rEgAgB8/1 and rEgAgB8/2 [34].
Since the lipid moiety of native EgAgB includes a wide range of lipids, we analysed the capacity of our particular recombinant subunits to bind different fluorescent lipid probes, such as the fatty acid analogues 12-AS and DAUDA, and the cholesterol analogue DHE. These probes have been widely used for characterising the lipid binding properties of LBPs because of their environment-sensitive spectral properties. They have very low fluorescence emission values when free in aqueous solution, but exhibit a significant increase in emission intensity, often accompanied by a blue shift in emission spectrum, when bound to a protein's apolar ligand binding site [60].
We found that 12-AS showed a significant increase in its fluorescence emission, accompanied by a blue shift from 456 nm to 446 nm when lipid-free rEgAgB8/2 or rEgAgB8/3 subunits were added to the probe solution (Fig. 6A and 6C). For both subunits, titration experiments described curves that reached saturation, in accordance with a ligand binding stoichiometry consistent with 1:1 binding per monomer (Fig. 6B and 6D), with a Kd of 0.16 ± 0.09 μM for rEgAgB8/2 (r = 0.9976) and 0.34 ± 0.02 μM for rEgAgB8/3 (r = 0.9927). In contrast, negligible enhancement of fluorescent emission was observed for the fatty acid analogue DAUDA upon adding lipid-free rEgAgB8/2 or rEgAgB8/3 (S1 Fig.). Lipidex-treated recombinant subunits EgAgB8/1 and EgAgB8/2 behaved similarly in previous studies since they bound the fatty acid anthroyloxy derivative, 16-AP, but not DAUDA [35]. The Kd values determined for the binding of these probes by our delipidated subunits and by Lipidex-treated subunits were similar, but the latter exhibited a lower value for binding sites for monomer (n value of approximately 0.3), further emphasising the distinction between methods employed to remove bacterial lipids bound to the recombinant subunits. These findings suggest that the hydrophobic Lipidex resin method is not appropriate to achieve adequate delipidation.
In addition, the fatty acid binding properties of EgAgB subunits could be compared with those of other HLBPs on the basis of the use of similar probes for this analysis, although in these previous studies HLBP delipidation was not reported [36,61,62]. In particular, 16-AP was employed for characterizing the binding capacity of HLBPs, from Taenia solium metacestode (TsM150) [36] and Moniezia expansa (MeHLBP) [61]. EgAgB subunits seem to be more similar by amino acid sequence to the 10 kDa than to the 7 kDa TsM HLBP subfamily. Moreover members of the former (referred to as recCyDA, recb1 and recm13h) but not of the latter (referred to as RS1) subfamily bind 16-AP [36]. Moreover, recb1 and recm13h were positioned into the same clade with EgAgB, whereas RS1 belonged to a different clade [62]. Furthermore, a recently described HLBP from T. solium, whose subunits grouped together with MeHLBP rather than with the 7 and 10 kDa TsM HLBP subfamilies, is also capable of binding 16-AP [62]. In the case of the MeHLBP monomers, the binding is consistent with a stoichiometry value of 1:1, and a Kd value of 2 μM, suggesting that lipid-free rEgAgB8 subunits exhibit a higher affinity for AOFA probes [61]. Taken together, these results indicate that various HLBPs are potentially capable of binding fatty acids, and that differences in their ligand binding properties are not easy to predict from their presumed evolutionary relationships.
On the other hand, the fluorescent cholesterol analogue, DHE, was found not to be bound by neither rEgAgB8/2 nor rEgAgB8/3 (S2 Fig.), suggesting that lipid-free EgAgB8/2 and EgAgB8/3 subunits do not directly interact with cholesterol. The presence of cholesterol in EgAgB purified from hydatid cyst fluid [15], suggests that it may be incorporated into the particle once EgAgB subunits have already bound their lipid ligands. On the other hand, the interaction of cholesterol with other EgAgB subunits cannot be ruled out.
In order to employ natural ligands instead of fluorescent ones we evaluated the effect of cholesterol and oleic acid on the intrinsic rEgAgB8/2 fluorescence using Trp16 as a reporter. This methodology has been successfully used for other HLBP members [61,63] and could not be applied to rEgAgB8/3 characterisation since this subunit does not contain Trp residues. Changes in Trp16 fluorescence were not observed when cholesterol or oleic acid were added to lipid-free rEgAgB8/2, indicating that the environment of Trp16 was not sensitive to the binding of these lipids (S3 Fig.). These results agree with previous observations obtained using Lipidex-treated rEgAgB8/2 and natural fatty acids as ligands [35], and suggest that Trp16 is not close to the putative ligand binding site for fatty acids present in this subunit.
Overall, our results, together with previous data, suggest that lipid-free rEgAgB8/2 and rEgAgB8/3 apolipoproteins have a selective capacity to bind lipids, showing affinity at least for 16- and 18-C fatty acids, but not for cholesterol, indicating that these components of the natural EgAgB lipoprotein particles would not interact directly with cholesterol.
How lipids carried by EgAgB complexes may be distributed within the parasite is unknown, so, we attempted to characterise the capacity of these subunits to transfer lipids to membranes. Since fatty acids, but not cholesterol, were bound by delipidated rEgAgB8 subunits, assays were designed to examine the transfer of 12-AS to phospholipid artificial membranes (SUVs). To establish the conditions for the transfer measurements we used the Kd values obtained for rEgAgB8/2 and rEgAgB8/3 to ensure that less than 5% of 12-AS remained free in solution. We also determined the Kp of 12-AS between rEgAgB8 subunits and vesicles to assure unidirectional transfer of 12-AS from rEgAgB8 to the vesicles. In order to do this, SUVs containing a FRET acceptor of the anthroyloxy group donor (NBD-PC) were added to a solution of 12-AS:EgAgB8/2 or 12-AS:EgAgB8/3 complex. The 12-AS fluorescence decay upon incremental increase in SUV concentration for both 12-AS:EgAgB8 complexes is shown in Fig. 7, indicating the transfer of 12-AS from EgAgB8 subunits to NBD-PC-containing vesicles. Using Eq. 2 (see Materials and Methods) a KP value of 0.62 ± 0.09 was obtained for rEgAgB8/2 and of 0.88 ± 0.15 for rEgAgB8/3. Both indicate that there is a slightly higher preference of 12-AS for the phospholipid membranes. Once these conditions were established, we analysed the transfer rates of 12-AS bound to rEgAgB8/2 or rEgAgB8/3 to the vesicles. Firstly we examined the transfer rates as a function of SUV concentration to determine whether the limiting step for ligand transfer is the effective protein-membrane interaction or the dissociation of the protein-ligand complex, as has been previously established for other LBPs [49–56]. A representative time trace of 12-AS fluorescence change upon SUV addition to EgAgB8/2:12-AS or EgAgB8/3:12-AS complexes is shown in Fig. 8A. We found that the 12-AS transfer rate from rEgAgB8/2 to EPC-SUVs increased significantly from 0.039 ± 0.003 s−1 to 0.084 ± 0.005 s−1 (p < 0.05) when SUV:protein ratio raised from 10:1 to 40:1. In the case of rEgAgB8/3, a trend towards an increase in 12-AS transfer rate (from 0.07 ± 0.02 s−1 to 0.08 ± 0.02 s−1) was observed, but this trend did not reach statistical significance (Fig. 8B). These results using zwitterionic SUVs suggest that the mechanism of 12-AS ligand transfer differed between EgAgB subunits; for rEgAgB8/2 the limiting step for transfer is the direct contact with the vesicle (collisional mechanism), whereas for rEgAgB8/3 the dissociation of 12-AS from the complex seems to be the limiting step (diffusional mechanism).
We then proceeded to examine whether electrostatic interactions between EgAgB subunits and SUVs could affect the ligand transfer mechanism. For this purpose we carried out similar transfer rate assays using negatively charged vesicles, which were obtained by incorporating PS or CL into the acceptor vesicles (one net negative charge per molecule of PS, two net negative charges per molecule of CL). Using PS-SUVs (Fig. 8C), 12-AS transfer rates of both rEgAgB8 subunits were similar to those observed for zwitterionic SUVs. Once again, the 12-AS transfer rate showed a statistically significant increase for rEgAgB8/2, but not for rEgAgB8/3 when the SUV:protein ratio exhibited a 4-fold increase. This suggests that the transfer of 12-AS to phospholipid membranes by EgAgB8 apolipoproteins was not significantly modified by the increase in the negative charge of the vesicles caused by PS incorporation. In contrast, when CL-SUVs were used, 12-AS transfer rates of both EgAgB8 proteins were remarkably increased. Moreover, for both EgAgB8 subunits, the transfer rates reached higher values at higher SUV:protein ratios (Fig. 8D), ranging from 0.43 ± 0.03 s−1 to 3.1 ± 0.6 s−1 for rEgAgB8/2 and from 0.9 ± 0.5 s−1 to 3.6 ± 0.6 s−1 for rEgAgB8/3. These results indicate that the greater increase in the negative charge of SUVs caused by CL incorporation enhanced the ability of rEgAgB8/2 and rEgAgB8/3 to transfer their ligands to vesicles, via a collision-mediated mechanism. Overall, these results suggest that electrostatic interactions between EgAgB8 subunits and phospholipid membranes are of foremost importance for determining the rate at which these proteins can transfer their ligands. Nevertheless, an increase in the affinity of EgAgB8 subunits to CL-SUVs could not be ruled out. Furthermore, the greater ability of rEgAgB8/2 to transfer 12-AS to CL-SUV vesicles than rEgAgB8/3 (p < 0.05, S4 Fig.) may be related to the negative and positive charge distribution in these proteins [16].
This report provides new approaches towards an understanding of cestodes HLBPs and how they may interact with phospholipid membranes to transfer their ligands. Since there are no reports for other HLBP members, EgAgB lipid transfer properties can only be compared to those of other LBP families [49–56]. As mentioned above, although the mechanisms employed by EgAgB8/2 and EgAgB8/3 to transfer 12-AS to EPC vesicles appear to be different, in both cases transfer rates are of the same order of magnitude. Previous studies have shown that LBPs that exchange ligands by collisional or diffusional processes also differ in the range of rates employed to transfer their ligands to acceptor membranes. For example, mammalian intestinal fatty acid binding protein (IFABP) and Schistosoma japonicum FABP (Sj-FABPc), both employing collisional mechanisms, transfer their ligands at a higher rate compared to diffusional proteins such as liver FABP and ABA-1-A1 protein from A. suum [49,51]). Furthermore, for proteins that transfer ligands through a collision-mediated mechanism, transfer rates increase proportionally to vesicle concentration (e.g. intestinal FABP, Sj-FABPc, EgFABP1 from E. granulosus or YLSCP2 from the yeast Yarrowia lipolytica [49,51,55,56]), and this is the case for EgAgB8/2. On the other hand, diffusional proteins have not shown variations in transfer rates as vesicle concentration or vesicle composition changes (e.g., liver-FABP, ABA-1-A1 or Ov-FAR-1 protein from Onchocerca volvulus [49,51]). Remarkably, the EgAgB8/3 transfer rate of 12-AS does not change with increasing concentrations of EPC or PS-SUVs, but a significant increase was observed with CL-SUVs, suggesting EgAgB8/3 employs a mechanism that is different to those described so far. Overall, a comparison of our results with those reported for other LBPs suggests that while EgAgB8/2 behaves as a typical collisional protein, EgAgB8/3 does not, even compared to other α-helical proteins such as ABA-1-A1, Ov-FAR-1 or YLSCP2. Whether this is an exclusive feature of EgAgB8/3 subunit or is a conserved behaviour of certain subunits of other HLBPs, needs further investigation.
Finally, our results showed that EgAgB8/2 and EgAgB8/3 subunits are able to deliver their cargo to phospholipid membranes, supporting the hypothesis that EgAgB is involved in lipid transport between parasite and host tissues [16]. Nevertheless, the capacity of EgAgB particles to transfer fatty acids to the parasite or to the host´s cells remains to be formally demonstrated.
The small proteins that collectively form the large Antigen B complexes present in the hydatid cysts of Echinococcus spp. and the similar entities in the cysticerci of other highly pathogenic cestodes, comprise the most abundant proteins present in the fluids of these parasites. Their native structures and role in lipid dynamics of the parasites remain to be elucidated, a difficulty being that our understanding of how the various components of the large Antigen B complexes form, what ligands they bind, and how, is at best rudimentary. In particular, the separate or synergistic roles of the small protein isoforms and lipid components remain mysterious. The advance that we report here is to provide for the first time a method for the complete removal of lipids from single recombinant isoforms of the protein components without detectably compromising their structures or biochemical activities, and the demonstration that self-assembly of complexes does not absolutely depend on lipids. It still appears, however, that lipids may enhance the process for the formation of the large complexes found in the natural product. In the natural particles, the complexes comprise heterogeneous mixtures of several different isoforms of the EgAgB proteins, with different types of lipids [15]. We have therefore provided the basis for the analysis of self-assembly that should eventually permit elucidation of potentially cooperative interactions of EgAgB isoforms and lipids. In early studies it has been suggested that EgAgB subunits are elongated and amphipathic molecules which could form multimeric structures thermodynamically more stable than individual monomers [33]. Recently, the prediction of the tertiary structure of EgAgB subunits suggests that the position of hydrophilic and hydrophobic amino acids defines pocket-like regions where hydrophobic ligands could interact with the proteins, and a partial charge distribution showing a plausible electrostatic profile for molecular oligomerization [16], in concordance with the experimental data we obtained in this work. Based on the information obtained from molecular modelling studies we can now begin to identify the motifs in EgAgB apolipoproteins that are involved in oligomerization or ligand binding.
The developments we report also allowed us to address the question of how EgAgB complexes may interact with cell membranes of the parasite in order to acquire, transfer and deliver lipids within the parasites. We found that they can deliver their cargo to phospholipid membranes through direct physical interaction with them, and that the charge composition of the membranes is crucial to this process. The way is now open to examine such processes using parasite cell lines or miniature hydatid cysts which are now available [64–67].
It has been more than 40 years since EgAgB was described as an abundant lipoprotein in the hydatid cyst fluid [13], and it has been successfully employed as an Echinococcus genus-specific target antigen for human serodiagnosis [19–22]. We know that lipids, such as cholesterol and fatty acids, are essential for E. granulosus s. l. [18] and that EgAgB is able to bind them in vivo [15]. Thus, EgAgB’s unusual construction and its participation in lipid uptake and delivery, together with the fact that it has no homologue in other animal phyla, places it in an excellent position to be seriously considered for therapeutic intervention. Our findings on EgAgB’s unusual lipid-binding, self-assembly, and membrane interaction properties therefore potentially present new avenues for drug developments that could disable a range of physiological processes essential to the parasite’s establishment and survival such as membrane construction and lipid-based signalling systems.
|
10.1371/journal.pgen.1004938 | Asymmetry of the Budding Yeast Tem1 GTPase at Spindle Poles Is Required for Spindle Positioning But Not for Mitotic Exit | The asymmetrically dividing yeast S. cerevisiae assembles a bipolar spindle well after establishing the future site of cell division (i.e., the bud neck) and the division axis (i.e., the mother-bud axis). A surveillance mechanism called spindle position checkpoint (SPOC) delays mitotic exit and cytokinesis until the spindle is properly positioned relative to the mother-bud axis, thereby ensuring the correct ploidy of the progeny. SPOC relies on the heterodimeric GTPase-activating protein Bub2/Bfa1 that inhibits the small GTPase Tem1, in turn essential for activating the mitotic exit network (MEN) kinase cascade and cytokinesis. The Bub2/Bfa1 GAP and the Tem1 GTPase form a complex at spindle poles that undergoes a remarkable asymmetry during mitosis when the spindle is properly positioned, with the complex accumulating on the bud-directed old spindle pole. In contrast, the complex remains symmetrically localized on both poles of misaligned spindles. The mechanism driving asymmetry of Bub2/Bfa1/Tem1 in mitosis is unclear. Furthermore, whether asymmetry is involved in timely mitotic exit is controversial. We investigated the mechanism by which the GAP Bub2/Bfa1 controls GTP hydrolysis on Tem1 and generated a series of mutants leading to constitutive Tem1 activation. These mutants are SPOC-defective and invariably lead to symmetrical localization of Bub2/Bfa1/Tem1 at spindle poles, indicating that GTP hydrolysis is essential for asymmetry. Constitutive tethering of Bub2 or Bfa1 to both spindle poles impairs SPOC response but does not impair mitotic exit. Rather, it facilitates mitotic exit of MEN mutants, likely by increasing the residence time of Tem1 at spindle poles where it gets active. Surprisingly, all mutant or chimeric proteins leading to symmetrical localization of Bub2/Bfa1/Tem1 lead to increased symmetry at spindle poles of the Kar9 protein that mediates spindle positioning and cause spindle misalignment. Thus, asymmetry of the Bub2/Bfa1/Tem1 complex is crucial to control Kar9 distribution and spindle positioning during mitosis.
| In asymmetrically dividing cells, proper positioning of the mitotic spindle relative to polarity determinants is crucial to ensure the unequal fate of daughter cells. In stem cells, derangement of the mechanisms controlling asymmetric cell division, including spindle positioning, affects the developmental fate of daughter cells and can promote tumourigenesis. The budding yeast Saccharomyces cerevisiae is an outstanding model system to study spindle positioning and its links with cell cycle progression. Indeed, budding yeast has redundant mechanisms driving spindle positioning and a “spindle position checkpoint” (SPOC) that delays cell division whenever the spindle is not properly aligned. The target of the SPOC is the small GTPase Tem1 that controls both spindle positioning and mitotic exit and whose activity can be inhibited by the GTPase-activating protein Bub2/Bfa1. Tem1, Bub2 and Bfa1 form a complex at spindle poles that becomes asymmetric and accumulates on one spindle pole when the spindle is properly aligned, while it remains symmetric in case of spindle mispositioning. Through expression of several mutant or chimeric proteins leading to symmetric distribution of the Bub2/Bfa1/Tem1 complex, we establish that asymmetry of these proteins does not drive mitotic exit but rather it contributes to spindle alignment.
| Asymmetric cell division generates two daughter cells genetically identical but that differ in fate and/or in size and cytoplasmic material. During asymmetric cell division, polarity factors are first concentrated to specific locations to define the poles of cell division. Afterwards the spindle orients according to these polarity cues to segregate one set of chromosomes towards a given polarity determinant and the other away from it, thereby generating two unequal daughter cells (reviewed in [1–3]). Correct spindle positioning is therefore critical to preserve the right lineage of asymmetrically dividing cells. Accordingly, spindle mispositioning in asymmetrically dividing stem cells, which normally generate one daughter stem cell with self-renewal potential and one cell destined to differentiation, steers tumourigenesis by increasing the pool of undifferentiated stem cells [4, 5]. Surveillance mechanisms, or checkpoints, must therefore respond to spindle positioning errors and delay cell cycle progression until the mitotic spindle is properly oriented with respect to the cell polarity axis [6, 7].
The budding yeast Saccharomyces cerevisiae is a widely recognized model system to study asymmetric cell division. Spindle positioning in budding yeast requires either one of two redundant pathways, one that depends on the APC (Adenomatous Polyposis Coli)-related protein Kar9, and the other on dynein (reviewed in [8]). Spindle positioning errors are monitored by a surveillance mechanism, referred to as spindle position checkpoint (SPOC), that delays mitotic exit and cytokinesis to provide the time for proper spindle realignment (reviewed in [6, 9]). The target of the SPOC is a small GTPase called Tem1, which acts as molecular switch for the activation of a kinase cascade related to the Hippo pathway and named Mitotic Exit Network (MEN). In the fission yeast S. pombe a kinase cascade similar to MEN and referred to as Septation Initiation Network (SIN) triggers cytokinesis [10]. The MEN effector of Tem1 is the kinase Cdc15, which in turn promotes the activation of the downstream Mob1/Dbf2 kinase complex that ultimately leads to activation of the Cdc14 phosphatase [11]. Cdc14 is the main phosphatase that in budding yeast counteracts the activity of cyclin-dependent kinases (CDKs), and it is essential for mitotic exit and cytokinesis by dephosphorylating CDK substrates, as well as by triggering inactivation of mitotic CDKs [12]. Cdc14 is sequestered in the nucleolus in an inactive form throughout most of the cell cycle, until it is released and activated. Although the MEN is necessary for the full release of Cdc14 into the cytoplasm to promote mitotic exit [13, 14], another pathway called FEAR (Cdc Fourteen Early Anaphase Release) causes a partial release of Cdc14 from the nucleolus into the nucleus at the metaphase-to-anaphase transition [15]. The FEAR pathway involves the polo kinase Cdc5, the redundant Spo12 and Bns1 proteins, and the separase Esp1, which by inhibiting the phosphatase PP2ACdc55 allows the dissociation of Cdc14 from its nucleolar inhibitor Net1 [15, 16]. The FEAR-mediated activation of Cdc14 in anaphase is thought to regulate spindle dynamics and to contribute to timely activation of the MEN (reviewed in [17]).
Recent data have shown that MEN is not only important for triggering mitotic exit in telophase, but also has an earlier function in metaphase to promote correct spindle positioning along the polarity axis [18]. In most MEN mutants, except for cdc14, spindles are indeed misoriented relative to the cell division plane. Notably, the Mob1/Dbf2 kinase was found to phosphorylate the spindle positioning Kar9 protein, thereby favouring its concentration on astral microtubules emanating from only one of the two spindle poles [18]. Asymmetric distribution of Kar9 at spindle poles in metaphase is in turn crucial for proper spindle positioning because it targets the Kar9-decorated aster to the bud, due to Kar9 interaction with the type V myosin Myo2 [19].
The yeast centrosomes, named spindle pole bodies (SPBs), play an important role in the regulation of mitotic exit, as they act as a scaffold for MEN components, such as Tem1 and its downstream kinases (reviewed in [20]). The constitutive SPB component Nud1 recruits MEN proteins to SPBs and is essential for mitotic exit [21], suggesting that binding of one or several MEN factors to SPBs is required for mitotic exit. Consistently, Tem1 association to SPBs is critical for MEN activation [22].
Like all GTPases, Tem1 is active when bound to GTP and inactive in its GDP-bound form. The common element of the GTPase superfamily is the 160–180 residue G domain involved in nucleotide binding [23]. Within the G domain two flexible “switch regions”, referred to as Switch I and II undergo the most dramatic structural rearrangement upon GTP hydrolysis and therefore define the major conformational changes conferred by GTP versus GDP binding [24]. On the basis of sequence alignment with human Ras, Switch I and II in Tem1 correspond to residues 50–55 and 77–84, respectively.
Upon spindle misalignment the two-component GTPase-activating protein (GAP) Bub2/Bfa1 inactivates Tem1 by stimulating GTP hydrolysis [25, 26]. GTPase-activating proteins accelerate GTP hydrolysis, promoting the GDP-bound inactive form of GTPases [27, 28]. The GAP activity of the Bub2/Bfa1 complex resides on Bub2, which carries a TBC domain (Tre-2, Bub2 and Cdc16; [29]), whereas Bfa1 mediates Bub2 interaction with Tem1 and prevents Tem1 dissociation from guanine nucleotides, thereby acting as guanine dissociation inhibitor (GDI) [25, 26, 30, 31].
Often, the release of GDP from GTPases is a slow and thermodynamically unfavourable reaction. This is why GTPase activation requires in most cases the intervention of nucleotide exchange factors (GEFs) that catalyse the release of GDP, promoting its replacement by GTP [27]. The identity of the GEF(s) for Tem1, if any, remains elusive. The early proposal based on genetic data that the Lte1 protein might be the GEF for Tem1 has not been confirmed by biochemical assays [30]. Therefore, if inactivation of a GEF for Tem1 could play any role in the SPOC, besides Tem1 inhibition by the GAP Bub2/Bfa1, remains to be established.
The Kin4 protein kinase is a key component of the SPOC (reviewed in [6]). During spindle misalignment it phosphorylates Bfa1, thereby preventing the inhibitory phosphorylation of the GAP Bub2/Bfa1 by the polo kinase [32]. During the unperturbed cell cycle Kin4 is strategically restricted to the mother cell compartment, where it is thought to sense the anomalous persistence of both SPBs in anaphase. In addition, it is present on both SPBs of misaligned spindles [32–35].
Localization of MEN and SPOC proteins to the SPBs changes during the cell cycle, in that some proteins (like Cdc15, Mob1 and Dbf2) are loaded on both SPBs at the anaphase onset, whereas other proteins (like Tem1, Bfa1 and Bub2) are present on both SPBs already in metaphase and their localization becomes much more asymmetric in anaphase, when they preferentially accumulate on the old, bud-directed SPB [36–44]. Moreover, the position of the spindle seems to play a role in controlling the asymmetric localization of Tem1, Bub2 and Bfa1 in anaphase. Indeed, these proteins localize less strongly but more symmetrically on the two SPBs when a misoriented spindle elongates within the mother cell and the SPOC turns on [36, 38, 40]. Whether the asymmetric localization of Tem1 or its GAP is important for triggering MEN signalling remains to be elucidated. Remarkably, the SIN counterparts of several MEN components also localize asymmetrically on SPBs during anaphase, with the homologs of the GAP components Bub2 (Cdc16) and Bfa1 (Byr4) occupying one SPB, while the GTP-bound form of the GTPase Spg1 and its effector kinase Cdc7 occupy the other [45, 46]. Thus, in spite of the different modes of cell division in the two yeasts (asymmetric in S. cerevisiae and symmetric in S. pombe) MEN and SIN signalling is asymmetric in both. SIN asymmetry has been proposed to be crucial for timely cytokinesis, as cdc16 and byr4 mutants where Spg1 and Cdc7 are symmetric in anaphase undergo multiple rounds of septation [46, 47].
In S. cerevisiae, Kin4 was found to increase the turnover of the Bub2/Bfa1 at SPBs. However, Kin4 does not promote asymmetric localization of Bub2/Bfa1 on properly oriented spindles [36]. Chimeric proteins obtained by fusing Bub2 or Bfa1 to the structural SPB component Cnm67 cause unscheduled mitotic exit in the presence of mispositioned spindles, which led to the proposal that high turnover of Bub2/Bfa1 might be important to inhibit Tem1 in the cytoplasm during SPOC activation [36]. Conversely, a modified version of Bub2 carrying 9 myc epitopes at the C terminus localizes the GAP and Tem1 rather symmetrically at SPBs and prevents mitotic exit in some sensitized MEN mutant backgrounds [25].
In order to shed light onto the relationship between Bub2/Bfa1 symmetry and SPOC response, we report the characterization of a series of mutants altering either the Bub2/Bfa1 subunits or Tem1 and causing symmetric localization of Tem1 and its GAP during properly oriented anaphase. Remarkably, these mutant proteins as a whole tend to activate, rather than inhibit, the MEN. In addition, they lead to more symmetric distribution of Kar9 on spindle poles and to spindle positioning defects, indicating that a delicate balance between MEN activation and inactivation is required for proper spindle alignment.
The catalytic mechanism of GTPase-activating proteins (GAPs) requires an ‘arginine-finger’, where the lateral chain of a conserved arginine (R85 for Bub2, [25, 29]), interacts with the nucleotide-binding site of a G protein, thus stimulating hydrolysis of the γ−phosphate. A new catalytic mechanism, called “dual finger”, was proposed for the family of GAPs with TBC (Tre-2, Bub2 and Cdc16) domain. According to the dual finger mechanism a conserved glutamine residue contributes to stimulate GTP hydrolysis together with the canonical catalytic arginine [48]. To investigate if Bub2 acts indeed via a dual finger mechanism, we generated a mutant Bub2 variant, Bub2-Q132L, where we replaced by leucine the conserved glutamine at position 132 that identifies the glutamine finger on the basis of sequence alignment [48]. Bacterially purified His-tagged Tem1, Maltose Binding Protein (MBP)-tagged Bfa1 and glutathione-S transferase (GST)-tagged Bub2 or Bub2-Q132L proteins were used in in vitro GTPase assays, as previously described [25, 26]. The rate of GTP hydrolysis and dissociation was measured using Tem1 bound to γ[32P]-GTP, whereas the rate of GTP dissociation alone was measured using Tem1 bound to the non-hydrolysable GTP analogue γ[35S]-GTP (Fig. 1A). As shown in Fig. 1A, the kinetics of radioactivity loss from wild type Tem1 loaded with either γ[32P]-GTP or γ[35S]-GTP were very similar, suggesting that Tem1 on its own mostly dissociates GTP without hydrolysing it. The presence of Bfa1 stabilized Tem1 in the GTP-bound form (Fig. 1B), whereas Bub2 stimulated Tem1 GTPase activity in the presence of Bfa1 (Fig. 1C), but not GTP dissociation (Fig. 1B). We then compared the GAP activity of purified GST-Bub2 and GST-Bub2-Q132L. Interestingly, Bub2-Q132L did not display any GAP activity towards GTP-bound Tem1 (Fig. 1C), behaving as the GAP–dead mutant Bub2-R85A previously characterized [25]. Furthermore, it did not stimulate GTP dissociation, exactly like wild type Bub2 (Fig. 1B).
In vivo, the Q132L substitution completely abolished the checkpoint function of Bub2. Indeed, similar to bub2Δ cells, bub2-Q132L cells escaped the mitotic arrest upon nocodazole treatment, as indicated by their ability to re-replicate their chromosomes (Fig. 1D). Checkpoint response to spindle misalignment was also impaired in bub2-Q132L cells. Indeed, when spindle mispositioning was induced by DYN1 or KAR9 deletion [49, 50] bub2-Q132L cells did not arrest in mitosis as large budded cells but re-budded, similar to bub2Δ cells (Fig. 1E). Thus, consistent with the proposed model [48], Bub2 GAP activity, and thereby its role in the SPOC, relies on a dual finger mechanism involving two catalytic residues, R85 and Q132.
The bub2-Q132L allele did not accelerate mitotic exit during the unperturbed cell cycle. Indeed, synchronized bub2-Q132L cells could divide and disassemble bipolar spindles with wild type kinetics (Fig. 1F–G). Furthermore, kinetics of degradation of the main mitotic cyclin Clb2 were very similar in wild type and bub2-Q132L cells (Fig. 1H). Interestingly, we found that cell cycle-dependent phosphorylation of Bfa1, which promotes mitotic exit [51], was abolished in bub2-Q132L cells, in agreement with the recent proposal that it requires Bub2 activity [52].
We then asked if Bub2-Q132L could still interact efficiently with Bfa1 and Tem1. Immunoprecipitations of Bub2 or Bub2-Q132L tagged with three HA epitopes showed that both proteins pulled down roughly the same amounts of GFP-tagged Bfa1. In contrast, Bub2-Q132L precipitated a higher amount of GFP-tagged Tem1 than wild type Bub2 (Fig. 1I), suggesting that abolishing the GAP catalytic activity of the Bub2/Bfa1 complex stabilizes the interaction between Tem1 and its GAP.
Because lack of Bub2 GAP activity through the bub2-R85A allele leads to increased symmetric localization of Bub2 to SPBs in anaphase [25], we analyzed the subcellular distribution of eGFP-tagged Bub2-Q132L. In contrast to wild type Bub2, which was almost exclusively present on the bud-directed SPB in 84% of anaphase cells, Bub2-Q132L-HA3 was found on both SPBs in 97% of cells in anaphase. In addition, Bfa1 and Tem1 were also more symmetrically localized on the SPBs of bub2-Q132L cells than they were in wild type cells during the same cell cycle stage (Fig. 1J). We therefore conclude that, consistent with previous data, interfering with Bub2 GAP activity affects the asymmetry of the Tem1/Bub2/Bfa1 complex on anaphase spindle poles.
Finally, we analysed the localization of the Tem1 effector kinase Cdc15 in bub2-Q132L cells. We found GFP-tagged Cdc15 on the SPBs of metaphase spindles in 55% of the cells. Deletion of BUB2 or its replacement with the bub2-Q132L allele increased both the total percentage of cells with Cdc15 at SPBs (83% and 80%, respectively, Fig. 1K) and the percentage of cells with symmetrically localized Cdc15 in metaphase (36% in bub2Δ and bub2-Q132L cells versus 11% of wild type cells). Thus, lack of Bub2 GAP activity leads to more efficient recruitment of Cdc15 at spindle poles.
Since spindle misalignment leads to persistent residence of Bub2/Bfa1 on both SPBs [40], we and others proposed that symmetric distribution of the GAP complex might lead to inhibition of Tem1 [25, 40]. This idea was further supported by our previous finding that a myc-tagged variant of Bub2 (Bub2-myc9) localizing mostly symmetrically on SPBs was lethal and prevented mitotic exit in sensitized backgrounds [25]. On the other hand, the Bub2/Bfa1 complex is required throughout most of the cell cycle for Tem1 association with SPB, which in turn triggers its activation [22, 40]. To assess the importance of Bub2/Bfa1 asymmetry at SPBs, we tethered Bfa1 or Bub2 to both SPBs by fusing them to the structural SPB component Spc72. We confirmed that in about 90% of the cells the Spc72-Bfa1 chimeric protein localised constitutively to the SPBs throughout the cell cycle (Fig. 2A) and was able to recruit Tem1 to both SPBs in 74% of anaphase cells, as opposed to 35% of wild type cells (Fig. 2B).
Both Spc72-Bfa1 and Spc72-Bub2 chimeric proteins were functional based on their ability to complement lack of endogenous BFA1 or BUB2, respectively, for what concerns the checkpoint response to microtubule depolymerisation. Indeed, in the presence of nocodazole, SPC72-BFA1 and SPC72-BUB2 cells arrested in mitosis with 2C DNA contents as well as wild type cells, whereas bub2Δ and bfa1Δ cells re-replicated their genome in the same conditions (Fig. 2C-D). Thus, the Spc72-Bfa1 and-Bub2 chimera are likely functional in that they retain their inhibitory properties towards Tem1.
Previously characterized Bub2 and Bfa1 chimeric proteins constitutively anchored to SPBs are SPOC-defective [36]. Similarly, our Spc72-Bfa1 and Spc72-Bub2 failed to activate the SPOC upon spindle mispositioning caused by DYN1 deletion (Fig. 2E). Indeed, dyn1Δ SPC72-BFA1 bfa1Δcells undergoing anaphase in the mother cell, which is symptomatic of spindle mispositioning, exited the cell cycle and re-budded, in contrast to dyn1Δ cells that arrested in mitosis as large budded cells (Fig. 2E). The SPOC failure of SPC72-BFA1 bfa1Δ cells was not worsened by deletion of BUB2 or KIN4 or both, consistent with the notion that Kin4 and Bub2/Bfa1 act in concert to inhibit Tem1. Similar results were obtained with the Spc72-Bub2chimera (S1 Fig.). Thus, constitutive targeting to both SPBs of the GAP Bub2/Bfa1, and of Tem1 as a consequence, leads to unscheduled Tem1 activation, consistent with a previous proposal [22].
We then asked if symmetric localization of Tem1 driven by the Spc72-Bfa1 chimera leads to more efficient recruitment of Cdc15 to SPBs in metaphase. This was indeed the case. Whereas GFP-tagged Cdc15 was present at the SPBs of 55% wild type metaphase cells, 90% of metaphase cells expressing the fusion SPC72-BFA1 displayed SPB-bound Cdc15 (Fig. 2F). Furthermore, Cdc15 was significantly more symmetric in SPC72-BFA1 than in wild type cells. Thus, stable tethering of Tem1 to SPBs by fusion to an SPB component [22] or by SPB recruitment via its inhibitory GAP ([36] and our data) leads in both cases to premature Tem1 activation.
We then asked if expression of the Spc72-Bfa1 chimera could have any phenotypic consequence for conditional mutants affecting the MEN. Remarkably, SPC72-BFA1 as the only source of Bfa1 in the cells suppressed the growth defects of several MEN mutants. In particular, it could partially rescue the temperature sensitivity of tem1–3 and mob1–77(Fig. 2G), as well as the cold-sensitivity of cdc15–2 and dbf2–2 mutant cells (S2A Fig.). A slight suppression, if any, was observed for the temperature-sensitivity of cdc5–2 cells, whereas the temperature-sensitivity of cdc14–3 cells was not suppressed at all (S2B-C Fig.).
Suppression of tem1–3 was recessive, as it was not observed when a wild type copy of BFA1 was present concomitant to SPC72-BFA1 in the cells (Fig. 2G). Importantly, suppression was not due to reduced GAP activity, as it could not be recapitulated by deletion of BFA1(Fig. 2G), or BUB2 or both (S2D Fig.). Since the mitotic exit defects of tem1–3 cells at high temperatures correlate with a loose interaction of the mutant Tem1 protein with SPBs [53], we conclude that Spc72-Bfa1 suppresses the temperature-sensitivity of tem1–3 cells likely by recruiting Tem1 to the SPBs. Consistent with this notion, the Spc72-Bfa1 and Spc72-Bub2 chimera suppressed the lethality caused by overexpression of the SPOC kinase Kin4 (Fig. 2H), which increases the turnover of Bub2/Bfa1, and by consequence of Tem1, at SPBs [36].
Thus, these data, together with those from a previous study [36], indicate that symmetric persistence of Bub2/Bfa1 at SPBs does not interfere with mitotic exit. Rather, in spite of being part of an inhibitory GAP complex, Bfa1 is a receptor of Tem1 at SPBs, where Tem1 promotes MEN signaling. Stable residence of Bub2/Bfa1 at SPBs causes unscheduled mitotic exit by decreasing Tem1 turnover at SPBs, as previously suggested [22].
The ability of Spc72-Bfa1 and-Bub2 tethers to efficiently prevent mitotic exit upon microtubule depolymerization, but not upon spindle mispositioning, was somewhat puzzling. A major difference between the two conditions lies in the activation of the spindle assembly checkpoint (SAC) after nocodazole treatment. Through inhibition of Cdc20/APC, SAC leads to securin stabilization, in turn preventing activation of separase and the FEAR pathway [15]. If inhibition of the FEAR pathway is the only reason for the failure of Spc72-Bfa1 and-Bub2 chimera to promote mitotic exit, premature FEAR activation should allow mitotic exit in cells expressing Spc72-Bfa1 and-Bub2 treated with nocodazole. Conversely, FEAR inactivation should prevent mitotic exit in the same cells undergoing spindle misalignment. To test this hypothesis, we prematurely activated the FEAR pathway and Cdc14 release from the nucleolus by either inactivation of the PP2ACdc55 phosphatase or ESP1 overexpression [16]. Remarkably, PP2ACdc55 inactivation through deletion of CDC55 had a synergistic effect with SPC72-BFA1 and SPC72-BUB2 on the kinetics of mitotic exit upon nocodazole treatment, as judged by the ability of cells to re-replicate their DNA (Fig. 3A-B). Similar data were obtained with ESP1 overexpression from the galactose-inducible GAL1 promoter (Fig. 3C-D). In contrast, reducing the levels of mitotic CDKs through CLB2 deletion did not accelerate mitotic exit in SPC72-BFA1 cells (Fig. 3E). Importantly, FEAR inactivation through deletion of both SPO12 and BNS1 reduced the unscheduled mitotic exit caused by SPC72-BFA1 in dyn1Δ cells (Fig. 3F).
Thus, constitutive recruitment of the Bub2/Bfa1 complex to SPBs leads to precocious Tem1 activation. Whether this translates into a premature mitotic exit depends on the activation state of the FEAR pathway.
To further investigate the links between Tem1 activity and the establishment of SPB asymmetry of the Tem1/Bub2/Bfa1 complex, we generated a TEM1-Q79L mutant allele, where the catalytic glutamine in the G domain (Q79, according to sequence comparison with Rab-like GTPases [27]), was replaced by leucine. We first tested the catalytic properties of Tem1-Q79L in in vitro GTPase assays in the presence of Bfa1 and Bub2. As shown in Fig. 4A, Tem1-Q79L was completely refractory to stimulation of GTP hydrolysis by the GAP Bub2/Bfa1 in vitro, suggesting that in vivo it is preferentially in its active GTP-bound form. Thus, Q79 of Tem1 likely participates directly to GTP hydrolysis along with R85 and Q132 of Bub2.
When expressed in yeast cells as the sole source of Tem1, Tem1-Q79L did not cause any detectable growth defect at any temperature. Furthermore, TEM1-Q79L mutant cells showed kinetics of cell cycle progression similar to those observed in wild type cells (Fig. 4B). The absence of obvious cell cycle phenotypes in unperturbed conditions was somewhat surprising, as a similar mutation in fission yeast spg1+, encoding the SIN counterpart of the Tem1 GTPase, leads to premature cytokinesis and formation of multiple septa [54]. We therefore analysed by time-lapse video microscopy the speed of actomyosin ring contraction, as a marker of cytokinesis [55], in wild type and TEM1-Q79L cells expressing GFP-tagged myosin II (Myo1). Strikingly, contraction of the actomyosin ring took place on average 2’ faster in TEM1-Q79L relative to wild type cells (i.e. 6’ and 8’, respectively, Fig. 4C), consistently with previous data on bub2Δ cells [56]. Thus, the TEM1-Q79L allele accelerates at least some aspects of cytokinesis without affecting cell viability.
To gain further insights into the factors allowing TEM1-Q79L cells to grow at normal rates, we carried out a synthetic genetic arrays (SGA) screen to find deletions of non-essential genes that become synthetically lethal/sick with TEM1-Q79L(Table 1). This screen uncovered several genes encoding proteins involved in spindle positioning and nuclear migration, such as Kar9, the dynein light chain (DYN2) and components of the dynactin complex that cooperates with dynein for spindle positioning [57]. In addition, this screen uncovered several genes implicated in microtubule biogenesis, which might indirectly influence spindle positioning. Thus, TEM1-Q79L aggravates the sickness of cells undergoing spindle mispositioning. Interestingly, the same deletion mutants were also identified in other SGA screens as synthetically sick or lethal with BUB2 or BFA1 deletion [58–60]. A number of additional non-essential genes whose deletion displayed synthetic interactions with TEM1-Q79L was also uncovered with this screen and will be described elsewhere, since the significance of these genetic interactions has not been further explored in this context.
We then tested the ability of TEM1-Q79L mutant cells to respond to microtubule depolymerization and spindle mispositioning. In the presence of nocodazole, whereas wild type cells arrested in mitosis with 2C DNA contents, TEM1-Q79L cells re-replicated their genome similar to cells lacking Bub2 (Fig. 4D). In addition, TEM1-Q79L dyn1Δ cells exited mitosis and re-budded in face of spindle position defects (Fig. 4E). Thus, the TEM1-Q79L mutant allele affects SPOC response. As expected, the checkpoint defect was dominant, as the TEM1-Q79L allowed mitotic exit and re-replication in the presence of nocodazole even when expressed from an episomal plasmid in cells carrying also the endogenous TEM1 gene (Fig. 4F). Consistent with its constitutive activation, the TEM1-Q79L allele was also able to suppress the lethality associated with overexpression of KIN4 from the galactose-inducible GAL1 promoter (Fig. 4G), which delays mitotic exit by keeping the Bub2/Bfa1 GAP active [34].
The “dual finger” model predicts that GTP hydrolysis is catalysed by the GAP and the so-called catalytic glutamine of the GTPase could stabilize the interaction between the GAP and the GTPase without directly contributing to the catalytic reaction [48]. We formally tested this idea by analysing the interaction between Tem1 or Tem1-Q79L with Bub2 and Bfa1 in co-immunoprecipitation experiments. Remarkably, HA-tagged Tem1-Q79L pulled down a higher, rather than a lower, amount of Bub2 and Bfa1 (tagged with 3PK epitopes and GFP, respectively) (Fig. 5A). Thus, constitutive binding to GTP seems to increase the affinity of Tem1 for its GAP. To further corroborate this conclusion we co-expressed Bfa1-GFP and Bub2-GFP with HA-tagged Tem1 or Tem1-Q79L. Immunoprecipitation of Tem1-Q79L-HA3 pulled down higher amounts of both Bfa1-GFP and Bub2-GFP than Tem1-HA3 (Fig. 5B), without affecting the relative proportion of Bfa1-GFP and Bub2-GFP in the immunoprecipitates. We therefore conclude that locking Tem1 in the GTP-bound form enhances its affinity for Bub2/Bfa1 without affecting the stoichiometry of the GAP complex.
Since our data suggest that proper regulation of Tem1 GTP hydrolysis is required for asymmetry of the Tem1/Bub2/Bfa1 complex at SPBs, we analysed the subcellular distribution of Tem1-Q79L tagged with eGFP. As expected, Tem1-eGFP was already asymmetric in 53% of metaphase cells, whereas Tem1-Q79L-eGFP was asymmetric in a lower fraction of cells (35%). In anaphase, whereas Tem1-eGFP localized asymmetrically to the bud-directed SPB in 90% of cells, Tem1-Q79L-eGFP was present symmetrically at SPBs in 70% of the cells (Fig. 5C). Thus, abolishing the GAP-stimulated GTPase activity of Tem1 through different kinds of mutations invariably leads to Tem1 symmetric localization at SPBs.
Interestingly, whereas BFA1 deletion markedly affected Tem1-eGFP recruitment to SPBs as previously reported [22, 40, 61], it had a less pronounced effect on the SPB localization of Tem1-Q79L-eGFP (Fig. 5D), suggesting that loading to SPBs of constitutively active Tem1 is partially GAP-independent.
Previous data suggested that loading of Bfa1 and Bub2 on SPBs and their asymmetry in anaphase still occur in cells lacking Tem1 [30, 40]. On the other hand, experimental evidence indicates that an increased residence time of Tem1 at the SPBs or its decreased GTPase activity can influence Bub2/Bfa1 localization [22, 25, 62]. Because our results indicate that the Q79L substitution affects Tem1 activity as well as its localization in anaphase, we asked if Tem1-Q79L had any impact on localization of Bfa1. Both wild type and TEM1-Q79L mutant cells showed a similar partially asymmetrical SPB localization of Bfa1-eGFP in metaphase (Fig. 5E). In contrast, at the onset of anaphase, while Bfa1 drastically dropped to hardly detectable levels on the mother-bound SPB in 95% of wild type cells, it remained completely symmetrical on SPBs in 22% of TEM1-Q79L cells and persisted to low but clearly detectable levels on the mother-bound SPB in 46.6% of the cells (Fig. 5F). The symmetric localization of Bfa1 in TEM1-Q79L cells did not depend on premature activation of downstream MEN kinases, as it was not affected by Cdc15 inactivation through the cdc15–2 temperature-sensitive allele (S3 Fig.).
Since Kin4 promotes Bfa1 turnover at SPBs upon SPOC activation [36], we analysed Bfa1 localization in wild type and TEM1-Q79L cells lacking KIN4(Fig. 5E-F). In agreement with previous data [36], deletion of KIN4 alone did not affect Bfa1 distribution on SPBs of wild type cells in unperturbed conditions. In stark contrast, it had a synergistic impact with the TEM1-Q79L mutant allele on Bfa1 localization at SPBs specifically in anaphase, making it completely symmetrical in 65% of the cells and partially asymmetric in 30% of the cells (Fig. 5F). Consistently, the ratio in Bfa1-GFP fluorescence intensity at the mother- versus the bud-directed SPB was close to 0 for wild type and kin4Δ cells, while it significantly increased in TEM1-Q79L and TEM1-Q79L kin4Δ cells (Fig. 5G). Thus, these data reveal an unanticipated role of Kin4 in actively dislodging Bfa1 from the mother-bound SPB during anaphase of the unperturbed cell cycle. Furthermore, they indicate that Tem1 GTP hydrolysis is a primary determinant of Bfa1 asymmetry in anaphase.
To investigate further if the TEM1-Q79L allele leads to premature MEN activation, we analysed the subcellular localization of downstream MEN components, such as Cdc15 and Mob1. We observed that 99% of TEM1-Q79L mutant cells recruited Cdc15 to SPBs in metaphase, as opposed to 55% in wild type cells (Fig. 6A). Furthermore, Cdc15 was significantly more symmetric in TEM1-Q79L than in wild type cells. Deletion of KIN4, either alone or in combination with the TEM1-Q79L allele did not have any impact on Cdc15 distribution (Fig. 6A). In spite of Cdc15 enhanced loading on SPBs, recruitment of Mob1 to SPBs in metaphase was only slightly increased in TEM1-Q79L relative to wild type cells (Fig. 6B), consistent with the notion that mechanisms other than the SPOC restrain MEN activity downstream of Tem1 until late anaphase.
Phosphorylation of Cdc15 and Mob1 by cyclin B/CDKs together with Bub2/Bfa1 GAP activity provides a dual inhibition of the MEN [63]. Consistently, we found that combining the TEM1-Q79L allele with deletion of CLB2, which encodes the main mitotic cyclin B, caused synthetic growth defects at 37°C (Fig. 6C). However, no synthetic lethality or detrimental synthetic defects were induced by the additional deletion of KIN4 or BFA1, suggesting that the components of the MEN downstream of Cdc15 are likely targets of negative regulators additional to Bub2/Bfa1 and Clb2-associated CDKs.
Partial asymmetry of Bub2/Bfa1 at SPBs begins already in metaphase [39]. Since during metaphase MEN induces asymmetric localization of Kar9 at spindle poles, which is in turn required for correct spindle positioning [18], we checked if symmetrical Bub2/Bfa1 and Tem1 might affect Kar9 localization and spindle positioning. To this end, we analysed the distribution of Kar9 tagged with eGFP on the metaphase spindles of wild type, SPC72-BFA1 bfa1Δ, SPC72-BUB2 bub2Δ, and TEM1-Q79L cells. Whereas 84.4% of wild type cells showed strongly asymmetric Kar9, this value dropped to 43.5%, 50.5% and 47.6% in SPC72-BFA1 bfa1Δ, SPC72-BUB2 bub2Δ and TEM1-Q79L cells, respectively, while the remaining fraction of cells displayed partial or complete symmetry (Fig. 7A).
Since our results indicate that the GAP activity of Bub2 and Bfa1 influences the localization of the GAP complex in the cell, we characterized Kar9 distribution in the GAP-dead Bub2 variants, bub2-Q132L and bub2-R85A cells, in cells expressing the symmetric GAP-inactive BUB2-myc9 construct [25]. All strains showed more symmetric localization of Kar9, with 33.9% of complete asymmetry for bub2-Q132L, 27.8% for bub2-R85A, 53.6% for BUB2-myc9, as opposed to 80.3% for wild type cells (Fig. 7B). Surprisingly, we found increased Kar9 symmetry also in bub2Δ and bfa1Δ mutant cells (45.7% and 41.9% of complete asymmetry, respectively), where Tem1 is present at SPBs at low levels [22], but symmetrically (Fig. 5C). Increased symmetry of Kar9 in the mutants (with the exception of bub2Δ and bfa1Δ that were not analysed) was accompanied by increased Bfa1 symmetry (Fig. 7B). Therefore, establishment of Bub2/Bfa1 and Tem1 asymmetry impacts on Kar9 asymmetry. Nonetheless, Bub2/Bfa1 symmetric distribution at SPBs is not sufficient to drive Kar9 symmetry. Indeed, when spindles were misaligned in dyn1Δ mutant cells Bub2 became increasingly more symmetric from metaphase to anaphase, whereas Kar9 symmetry increased only slightly (Fig. 8A-B), in agreement with recently published data [64, 65]. Thus, once Kar9 asymmetry has been established, it cannot be reversed by spindle misalignment, in contrast to that of Bub2/Bfa1.
We then asked if Kar9 mislocalization in our mutants affects spindle positioning. SPC72-BFA1 and TEM1-Q79L cells expressing Spc42-mCherry were synchronized to collect cells in metaphase and imaged. Measurements of spindle distances from the bud neck (Fig. 7C) and spindle angles relative to the mother-bud polarity axis (Fig. 7D) indicated that both TEM1-Q79L and SPC72-BFA1 cells significantly affected the position and the orientation of metaphase spindles. Thus, hyperactive Tem1 impaired Kar9-dependent spindle positioning.
In conclusion, although constitutive symmetric localization of Tem1 and its GAP Bub2/Bfa1 does not appear to affect mitotic exit, it does compromise asymmetry of Kar9 at spindle poles, thereby causing spindle positioning and orientation defects.
The mechanistic details of how GTPases switch between a GTP- and a GDP-bound state build on initial structural studies on Ras. In Ras a conserved glutamine in the switch II domain of the GTPase and a conserved arginine of the GAP both contribute to GTP hydrolysis [66, 67] and, consistently, mutations of either residue abolish GTP hydrolysis. Crystal structure of some Rab GTPases in complex with their TBC (Tre-2, Bub2 and Cdc16) GAP revealed that the conserved switch II glutamine (Q79 of Tem1) does not directly participate in GTP hydrolysis. Rather, catalysis is entirely brought about by a conserved arginine and a conserved glutamine of the TBC GAP through a mechanism referred to as “dual-finger” [48, 68]. Hence, the switch II glutamine of the GTPase was proposed to stabilize its interaction with the GAP [48]. Recently, however, Rab GTPases have been shown to be more plastic than originally anticipated in their activation/hydrolysis mechanisms. In particular, the contribution of the switch II glutamine in GTP hydrolysis is variable and for some GTPases it contributes, together with a conserved lysine in the P-loop, to activation of the GTPase by stabilizing its GEF-bound nucleotide-free form [69]. Therefore, the outcome of mutations of conserved catalytic residues varies depending on the GTPase, GEF and GAP, and is altogether unpredictable.
Here we have addressed the importance of Q79 in the switch II and the dual-finger mechanism in Tem1 GTP hydrolysis. First, we have established that the intrinsic rate of Tem1 GTP hydrolysis is negligible, and loss of GTP is mostly accounted for by nucleotide dissociation. In agreement with previous data [25, 26, 30], Bfa1 prevents nucleotide dissociation and therefore acts as guanine-nucleotide dissociation inhibitor (GDI). Second, we show for the first time that Tem1 Q79 is directly involved in the GAP-induced GTP hydrolysis without impairing its interaction with Bub2 and Bfa1. Mutation of Q79 into leucine generates a hyperactive, dominant Tem1 that is refractory to its GAP, recruits more efficiently Cdc15 to SPBs and leads to unscheduled mitotic exit in the presence of spindle positioning defects.
Finally, we show that the dual-finger mechanism applies also to GTP hydrolysis of the Tem1-Bub2-Bfa1 complex. Indeed, glutamine 132 of Bub2 is involved in GTP hydrolysis, in addition to arginine 85 that we previously showed [25]. Consistent with an important role of Q132 in Tem1 inhibition, bub2-Q132L mutant cells are SPOC-defective and undergo mitotic exit upon microtubule depolymerization.
Thus, we have defined Q79 of Tem1 together with R85 and Q132 of Bub2 as a catalytic triad for GAP-induced GTP hydrolysis.
The Bub2/Bfa1 complex is required for efficient Tem1 binding to SPBs throughout most of the cell cycle, except in late mitosis [22, 40, 61]. In contrast, SPB recruitment of Bub2/Bfa1 does not require Tem1 [30, 40]. The amount of Tem1 at SPBs depends on the turnover of Bub2/Bfa1 at SPBs, which in turn is accelerated by spindle mispositioning through Kin4-dependent phosphorylation of Bfa1 [36, 39]. Thus, as recently proposed [61], the GAP Bub2/Bfa1 is a major Tem1 receptor at SPBs and its regulation is instrumental for establishing Tem1 asymmetry. Critical regulators of Bub2/Bfa1 asymmetry are the polo kinase Cdc5 and the phosphatase PP2ACdc55. Cdc5 phosphorylates and inactivates the Bub2/Bfa1 complex leading to Tem1 activation [51, 70], whereas PP2ACdc55 dephosphorylates Bfa1 [71]. Phosphomimetic mutations in some Cdc5-dependent Bfa1 phosphorylation sites, as well as loss of PP2ACdc55, are sufficient to induce premature asymmetry of Bub2/Bfa1 at SPBs, whereas Cdc5 inactivation or phospho-ablating mutations in Bfa1 lead to its persistent symmetry [62, 71]. Similar mechanisms might be operational in fission yeast to establish SIN asymmetry. Indeed, polo kinase has been proposed to phosphorylate the Bfa1 homolog Byr4 and promote its dissociation from SPBs [72], thereby influencing the distribution of GTP-bound Spg1 and its effector kinase Cdc7. Furthermore, PP2A regulates Byr4 asymmetry through dephosphorylation of the SIN anchor at SPBs Cdc11 [73, 74]. Indeed, Byr4 binds more efficiently to dephosphorylated Cdc11 [73], whereas the Cdc15-like kinase Cdc7 binds preferentially phosphorylated Cdc11 [75]. Thus, S. cerevisiae and S. pombe might adopt common regulatory strategies to establish MEN and SIN asymmetry.
We previously proposed that Tem1 GTP hydrolysis promotes asymmetry of both Tem1 and its GAP Bub2/Bfa1 at SPBs in anaphase [25] and results from other studies [22, 62] support this conclusion. Consistent with this idea, we now show that mutating the second catalytic finger of Bub2 (Q132) or mutating the catalytic Q79 of Tem1 leads in both cases to increased symmetry of Bub2/Bfa1 and Tem1 at SPBs. At a first glance these results appear at odds with the finding that in the complete absence of Tem1 Bub2/Bfa1 is not only recruited to SPBs with normal kinetics, but becomes asymmetric in anaphase exactly like in wild type cells [30]. However, we now show that when GTP hydrolysis is abolished Bub2 and Bfa1 bind more avidly to Tem1. Thus, the increased symmetry of Bub2/Bfa1 at SPBs in these conditions likely reflects its stronger affinity for GTP-bound Tem1. The different affinity of Bub2/Bfa1 for GTP- versus GDP-bound Tem1 has important implications for the SPOC, where active Tem1 needs to be quickly inactivated in the presence of a mispositioned spindle.
Based on our previous results using a version of Bub2 tagged with nine myc epitopes at the C-terminus (Bub2-myc9), we proposed that removal of Bub2/Bfa1 from the mother SPB is important for timely mitotic exit [25]. Indeed, Bub2-myc9 is more symmetrically localized at SPBs than wild type Bub2 and is lethal for cdc5–2 and tem1–3 mutants because it prevents mitotic exit in these sensitized backgrounds. Now we further tested this idea by expressing chimeric proteins that constitutively recruit the GAP and Tem1 to both SPBs. Contrary to our predictions, these chimeric proteins partially rescued, instead of aggravating, the temperature-sensitive growth phenotype of several MEN mutants. In particular, Spc72-Bfa1 rescued the temperature-sensitivity of tem1–3 cells likely by suppressing the SPB-binding defects of the mutant Tem1–3 protein at high temperature [53]. Importantly, suppression of MEN mutants by our chimeric proteins is not accounted for by their possible impaired GAP activity, because it could not be recapitulated by BFA1 and/or BUB2 deletion.
The chimeric proteins Spc72-Bfa1 and-Bub2 caused also unscheduled mitotic exit in the presence of mispositioned spindles, similar to Bub2- and Bfa1-Cnm67 fusion proteins previously characterized [36]. Therefore, despite different parts of Bub2 and Bfa1 are fused to the SPB anchor (the N-terminus in our Spc72- fusions and the C-terminus in the-Cnm67 chimera), constitutive binding of the Bub2/Bfa1 complex to SPBs invariably leads to SPOC defects. Our data are totally consistent with the notion that Tem1 recruitment to SPBs is necessary for its MEN function [22]. In this scenario, SPB-locked Bub2/Bfa1 activates Tem1 by increasing its symmetry and residence time at SPBs, thereby causing unscheduled mitotic exit in the presence of mispositioned spindles. It is worth noting that symmetry of SIN components, such as the Cdc7 kinase, at SPBs also causes unscheduled cytokinesis and repeated rounds of septation [46, 47], suggesting that asymmetry is a common strategy in S. cerevisiae and S. pombe to restrain MEN/SIN signalling.
The molecular basis for SPB-driven Tem1 activation is not known. It is possible that the GAP Bub2/Bfa1 at SPBs is constitutively kept inactive by Cdc5-mediated phosphorylation, thereby making Tem1 at SPBs refractory to GAP-mediated inhibition. Upon spindle misalignment, Kin4-mediated dislodgement of Bub2/Bfa1 from SPBs becomes essential for Tem1 inhibition in the cytoplasm and SPOC response [36, 61]. An interesting non-mutually exclusive hypothesis is that a putative GEF for Tem1 localizes at SPBs [76]. However, as mentioned above the identity of Tem1 GEF(s) remains elusive.
The finding that the chimeric proteins Spc72-Bfa1 and-Bub2, similar to Bub2- and Bfa1-Cnm67 [36] and Cnm67-Tem1 [22], support a mitotic arrest after microtubule depolymerization, but not after spindle mispositioning, was somehow puzzling. One major difference between the SAC-mediated metaphase arrest and the SPOC-mediated anaphase arrest is that in the latter, but not in the former, the PP2ACdc55 phosphatase is inhibited by the FEAR pathway [15, 16]. We find that, indeed, CDC55 deletion or ESP1 overexpression in cells expressing Spc72-Bub2 or-Bfa1 drives unscheduled mitotic exit in the presence of nocodazole. In contrast, FEAR inhibition by deletion of SPO12 and BNS1 [15] prevents mitotic exit in cells expressing the same chimeric proteins and experiencing spindle position defects. Thus, whether the FEAR is activated or inhibited influences the impact of the chimeric proteins on mitotic exit. PP2ACdc55 has been recently shown to antagonize the Cdc5-dependent phosphorylation of Bfa1 [71], thereby providing a mechanistic explanation to our data. The FEAR-mediated partial release of Cdc14 from the nucleolus might also contribute to MEN activation by counteracting the CDK-dependent inhibitory phosphorylation of MEN components [63].
In conclusion, persistent symmetric localization of the GAP Bub2/Bfa1 does not interfere with mitotic exit. Most likely, the Bub2-myc9 protein that we described previously prevents timely MEN activation by a different mechanism. Consistently, tagging of Spc72-Bub2 at the C-terminus with nine myc epitopes causes synthetic sickness in combination with the cdc5–2 mutant allele affecting the polo kinase (S1 Table).
We show that symmetric localization of the Bub2/Bfa1/Tem1 complex, independently of whether it is driven by chimeric proteins or loss of GTPase activity, interferes with Kar9 asymmetry at spindle poles in metaphase, as well as with spindle positioning and orientation relative to the cell division axis. Although Tem1 inactivation causes similar phenotypes for what concerns Kar9 localization and spindle orientation, it does not affect spindle positioning at the bud neck [18], indicating that Tem1 hyperactivation and inactivation are not equivalent in this respect. The molecular bases of this difference remain to be established. Similarly, whether Tem1 hyperactivation primarily affects Kar9 localization and, as a consequence, spindle positioning or vice-versa remains to be investigated. Although the phenotypic analyses of our mutants did not reveal any apparent alterations of astral microtubules, at the moment we cannot exclude that subtle defects in microtubule dynamics could account for spindle mispositioning and, in turn, increased Kar9 symmetry.
Previous [64] and our data indicate that not all conditions leading to symmetric distribution of the Bub2/Bfa1 complex and Tem1 cause symmetric localization of Kar9 at spindle poles. Indeed, whereas Bub2, Bfa1, Tem1 and Kar9 are all asymmetric, to different extents, on metaphase spindles, Bub2, Bfa1 and Tem1 become increasingly more symmetric upon spindle misalignment, while Kar9 remains strongly asymmetric. These data suggest that establishment of Bub2/Bfa1/Tem1 symmetry on misaligned spindles is an active process and Kar9 asymmetry is so robust that once established it cannot be reversed by bringing the Bub2/Bfa1/Tem1 complex to both spindle poles. In contrast, in our mutants Bub2, Bfa1 and Tem1 are more symmetric already in metaphase and might therefore interfere with the establishment of Kar9 asymmetry. Furthermore, we speculate that in these mutants the residence of Tem1 and the GAP Bub2/Bfa1 at SPBs is relatively stable, while these proteins turn over very fast at the SPBs of misaligned spindles [36]. An alternative explanation to our data is that Tem1 hyperactivation, rather than symmetry, is responsible for Kar9 mislocalization and spindle misalignment. Indeed, while in our mutants Tem1 is likely in the active GTP-bound form, it is inactivated and GDP-bound when the SPOC is turned on.
Factors involved in cell polarity were implicated in the asymmetry of Bub2-Bfa1 at spindle poles [25, 39]. Thus, it is tempting to speculate that in budding yeast the role of cell polarity in spindle positioning might be partly exerted through asymmetric localization of the Bub2-Bfa1-Tem1 trimeric complex at spindle poles, which in turn influences Kar9 asymmetry. Remarkably, other eukaryotic cells (i.e. nematodes, flies and mammals) employ heterotrimeric G proteins for spindle positioning during both symmetric and asymmetric cell division (reviewed in [77]. A striking parallel can be drawn between the asymmetric enrichment of their GDIs GPR-1/2, which is controlled by polarity factors and necessary for proper spindle alignment (reviewed in [78], and asymmetric localization of Bub2-Bfa1. Future work will certainly shed new light onto possible additional similarities in the mechanisms adopted by different organisms to achieve correct spindle positioning.
All strains, except those used for Fig. 7B and 8 (derivatives of S288c) are derivatives of W303 (ade2–1, trp1–1, leu 2–3,112, his 3–11,15, ura3, ssd1) and listed in S2 Table. Cells were grown in YEP medium (1% yeast extract, 2% bactopeptone and 50mg/L adenine) supplemented with 2% glucose (YEPD) or 2% galactose (YEPG). Unless otherwise stated, α-factor, nocodazole and benomyl were used at 2, 15 and 12,5 μg/ml respectively. Synchronization experiments with α-factor were performed at 25°C. Bacterial cells were grown in LD broth (1% bactotryptone, 0,5% yeast extract and 0,5% NaCl pH7,25) supplemented with 50 µg/ml ampicillin and 34 μg/ml chloramphenicol.
The SPC72-BUB2 fusion was generated by triple ligation of a HindIII/XbaI PCR fragment containing the whole ORF of 340 bp of SPC72 promoter, a XbaI/EcoRI PCR fragment containing the ORF of BUB2 spanning codons 2–172, and the LEU2-based Yiplac128 vector linearized with HindIII and EcoRI. The generated plasmid (pSP275) was linearized with BamHI for integration at the BUB2 locus, thereby generating a gene fusion under the SPC72 promoter and containing the whole ORF of SPC72 fused in frame to the entire ORF of BUB2, as well as a truncated BUB2 gene lacking the last 134 codons. Single integration of the construct at the BUB2 locus was checked by Southern blot.
The SPC72-BFA1 fusion was generated by triple ligation of a HindIII/XbaI PCR fragment containing the whole ORF of 340 bp of SPC72 promoter, a XbaI/BglII PCR fragment containing the entire ORF of BFA1 starting from the 2nd codon and 330 bp of 3’ UTR, and the LEU2-based Yiplac128 vector linearized with HindIII and BamHI. The generated plasmid (pSP371) was linearized with PstI for integration at the SPC72 locus and single integration of the construct at the SPC72 locus was checked by Southern blot.
The ORF of TEM1 and about 1000 bp of promoter region was cloned in Yiplac128 (pSP596). A variant carrying the TEM1-Q79L mutation was generated by site-directed mutagenesis (pSP597). The TEM1-bearing plasmids have been integrated at the LEU2 locus by BstXI digestion and single integrations have been checked by Southern blot.
Gene deletions were generated by one-step gene replacement [79]. One-step tagging techniques [80, 81] were used to tag Tem1, Tem1-Q79L, Bfa1, Bub2, Spc72-Bfa1 and Kar9 with multiple HA tags or eGFP.
E. coli BL21 cells carrying pLysE plasmid (Novagen) and 6His-TEM1, 6xHis-TEM1-Q79L, MBP-BFA1, GST-BUB2 and GST-BUB2-Q132L expression plasmids were grown in LD broth containing ampicillin and chloramphenicol at 37°C for 3 h, transferred to 14°C for 1 h and induced with 0,1 mM isopropyl-1-thio-β-D-galactopyranoside for 15 h. Cells expressing MBP-Bfa1, 6His-Tem1 and GST-Bub2 fusions were resuspended, respectively, in the following cold lysis buffers: 50 mM Tris-HCl pH7.5, 200 mM NaCl and 2mM DTT supplemented with a cocktail of protease inhibitors (Complete; Boehringer); 50 mM Tris-HCl pH8, 300 mM NaCl, 2mM MgCl2 and 10mM imidazole supplemented with a cocktail of protease inhibitors; 50 mM Tris-HCl pH7.5, 200 mM NaCl supplemented with a cocktail of protease inhibitors. Cells were incubated with 1 mg/ml lysozyme in ice for 30 min, placed at 37°C for 5 min and sonicated at 4°C for 10 seconds. The extract was then clarified by centrifugation at 15,000 rpm for 30 min at 4°C. Tem1–6xHis and Tem1-Q79L-6xHis were purified by affinity chromatography with Ni-NTA columns (QUIAGEN). The MBP-Bfa1 fusion protein was purified using Amylose resin (New England Biolabs, Inc.), whereas GST-Bub2 and GST-Bub2-Q132L were purified with glutathione-Sepharose (GE Heathcare). After elution, the fusion proteins were dialyzed against 50 mM Tris-HCl pH7.5, 200 mM NaCl and stored at -80°C. For quantification, purified proteins were analyzed by Comassie staining and by Western blot with anti-GST polyclonal antibodies (Santa Cruz Biotechnology, Inc.), anti-MBP mAb (New England Biolabs, Inc.) and 6xHis mAb (CLONTECH Laboratories, Inc.).
GTPase assay were performed according to [25]. In brief, 240 nM of Tem1–6×His was incubated in 25 μl of loading buffer (20 mM Tris-HCl, pH 7.5, 25 mM NaCl, 5 mM MgCl2, and 0.1 mM DTT) containing 0.1 MBq of γ[32P]GTP or 0.03 MBq of γ[35S]GTP in the absence or presence of 150 nM of MBP-Bfa1 for 10 min at 30°C. The reaction was then put on ice, and 10 μl of reaction were added to 50 μl of reaction buffer (20 mM Tris-HCl, pH 7.5, 2 mM GTP, and 0.6 μg/μl BSA) containing 15 μM of GST-Bub2. The mixture was incubated at 30°C, and for each time point 10 μl of the reaction was diluted in 990 μl of cold washing buffer (20 mM Tris-HCl, pH 7.5, 50 mM NaCl, and 5 mM MgCl2). The samples were filtered through nitrocellulose filters, washed with 12 ml of cold washing buffer, and air dried, and the filter-bound radioactivity nucleotide was determined by scintillation counting.
Immunoprecipitations were performed as described in [82]. Bub2-HA3, Bub2-Q132L-HA3, Tem1-HA3 and Tem1-Q79L-HA3 were immunoprecipitated from 1 mg of total extract by a HA-affinity resin (Roche). For Western blot analysis, protein extracts were prepared according to [83]. Proteins transferred to Protran membranes (Schleicher & Schuell) were probed with anti-PK mouse monoclonal antibodies for PK-tagged Bub2, with anti-GFP rat monoclonal antibodies for GFP-tagged Tem1, Bub2 and Bfa1 (Chromotek) and with an anti-HA monoclonal antibody (12CA5). Secondary antibodies were purchased from GE Healthcare, and proteins were detected by an enhanced chemiluminescence system according to the manufacturer.
In situ immunofluorescence was performed according to [84]. Immunostaining of α-tubulin was performed with the YOL34 monoclonal antibody (Serotec) followed by indirect immunofluorescence using rhodamine-conjugated anti–rat antibody (1:100; Pierce Chemical Co.).
Cells expressing GFP and mCherry-tagged proteins were grown in minimum complete medium. Digital images of live cells, cells fixed with 3.7% formaldehyde or cold EtOH were taken with an oil 63X 1,4–1,6 HCX Plan-Apochromat objective (Zeiss) with a Coolsnap HQ2–1 charge device camera (Photometrics) mounted on a ZeissAxioimager Z1/Apotome fluorescence microscope controlled by the MetaMorph imaging system software. Z-stacks of 12 planes at 0.3 μm step size were acquired.
For analysis of actomyosin ring contraction (Fig. 4C) cells were mounted in SD medium on Fluorodishes and filmed at room temperature (~21°C) with a DeltaVision OMX microscope using a 63X 1.4 NA oil immersion objective and the softWoRx software (Applied Precision). Z stacks containing 31 planes were acquired every 1’ with a step size of 0.2 μm and a binning of 1. Z-stacks were deconvolved with Huygens (Scientific Volume Imaging) and max-projected.
For the analyses in Fig. 8, 120s time-lapse microscopy was performed using an Olympus BX51 microscope controlled by the TILLVision software (TILLPhotonics). For the localization of Kar9-YFP and Bub2-GFP, Z-stacks of four layers (step size 0.35 μm) and maximum intensity projections were used. Fluorescence microscopy was performed with a monochromator PolychromIV as light source and a CCD camera (Imago, TillPhotonics).
Fluorescence intensity measurements of max intensity-projected images were performed using the ImageJ software. The index of Bfa1 symmetric distribution (σ, Fig. 5G) was measured using the following equation: σ (0<σ<1) = I1/I2, where I1 is the fluorescence intensity of the brightest of the two SPBs, and I2 is the fluorescence intensity of the dimmest.
The position and the orientation of the spindle were measured on max intensity-projected images using ImageJ. The position was determined measuring the distance between the bud neck and the nearest SPB. The orientation was determined measuring the smaller of the two angles that the spindle forms intersecting the polarity axis of the cell.
Adobe Photoshop and ImageJ were used to mount the images and to produce merged color images. No manipulations other than contrast and brightness adjustments were used.
Student’s t-test or chi-square test was used to evaluate statistical significance of differences, depending on whether one (t-test) or more parameters (chi-square test) were compared for each experimental condition.
Nuclear division was scored with a fluorescence microscope on cells stained with propidium iodide (Sigma Aldrich). Flow cytometric DNA quantification was determined according to [84] on a Becton-Dickinson FACScalibur.
|
10.1371/journal.pntd.0004622 | Effects of Mother’s Illness and Breastfeeding on Risk of Ebola Virus Disease in a Cohort of Very Young Children | Young children who contract Ebola Virus Disease (EVD) have a high case fatality rate, but their sources of infection and the role of breastfeeding are unclear.
Household members of EVD survivors from the Kerry Town Ebola Treatment Centre in Sierra Leone were interviewed four to 10 months after discharge to establish exposure levels for all members of the household, whether or not they became ill, and including those who died. We analysed a cohort of children under three years to examine associations between maternal illness, survival and breastfeeding, and the child’s outcome. Of 77 children aged zero to two years in the households we surveyed, 43% contracted EVD. 64 children and mothers could be linked: 25/40 (63%) of those whose mother had EVD developed EVD, compared to 2/24 (8%) whose mother did not have EVD, relative risk adjusted for age, sex and other exposures (aRR) 7·6, 95%CI 2·0–29·1. Among those with mothers with EVD, the risk of EVD in the child was higher if the mother died (aRR 1·5, 0·99–2·4), but there was no increased risk associated with breast-feeding (aRR 0·75, 0·46–1·2). Excluding those breastfed by infected mothers, half (11/22) of the children with direct contact with EVD cases with wet symptoms (diarrhoea, vomiting or haemorrhage) remained well.
This is the largest study of mother-child pairs with EVD to date, and the first attempt at assessing excess risk from breastfeeding. For young children the key exposure associated with contracting EVD was mother’s illness with EVD, with a higher risk if the mother died. Breast feeding did not confer any additional risk in this study but high risk from proximity to a sick mother supports WHO recommendations for separation. This study also found that many children did not become ill despite high exposures.
| Our study is the first to quantify sources of infection and describe risk of transmission of Ebola to young children. We found that the risk of a child under three developing Ebola disease was low unless their mother had EVD, and that the risk was particularly high if their mother died of EVD. But we found no additional risk from breastfeeding. WHO recommends separating asymptomatic breast-fed infants from their mothers if they develop Ebola, and using formula feeding. We support the need for separation because of the high risk related to proximity, but more research is needed to more fully understand this, particularly given the importance of breast-feeding in preventing other childhood illnesses. We also found young children in Ebola-affected households whose mothers were not ill had a surprisingly low risk of developing EVD which was not all explained by low exposure to the virus. Many children stayed well despite having direct contact with EVD patients with diarrhoea, vomiting or bleeding who are considered the most infectious. We hope these findings will provide impetus for more detailed studies into age-related response to the Ebola virus.
| Young children experience a high case fatality rate from Ebola, but the incidence of Ebola Virus Disease (EVD) in children appears to be lower than in adults.[1–4] Young children may have limited exposure outside the home, but within the household maintaining hygiene in young children is difficult, although efforts may be made to keep children away from those who are sick. For very young children who need to be fed and held, contact with sick caregivers may be unavoidable.
Breastfeeding is a possible additional source of infection for young children: Ebola has been found in breast milk, but the risk to breastfed babies and the contribution of breastfeeding to transmission is poorly understood.[5,6] An investigation of household contacts following the Ebola outbreak in Gulu, Uganda in 2000 included five infants whose mother had EVD: three of four infants who were breastfed developed EVD.[7] The other infant was reported to have been separated from his mother early in the course of her illness and remained well; it is not clear if he was breastfed. Two recent systematic reviews of transmission of Ebola did not did not mention risks associated with breastfeeding.[8,9]
As part of a study of transmission patterns in Sierra Leone we collected data on exposure patterns and outcomes of all individuals present in the households of EVD survivors. In this analysis we sought to identify likely sources of infection and characterise risk of transmission to young children, including those breastfed by mothers with EVD.
In July-September 2015, interviews were sought with the household members of all individuals who were discharged from the Ebola Treatment Centre in Kerry Town, Sierra Leone (“Ebola survivors”) from November 2014 to March 2015. Contact was made through members of the survivor support team who were involved in their reintegration into the community. An initial approach was made to explain the study. If the household head agreed, an interview was arranged at a community centre or other meeting place and all who were in the household at the time that members of the households had Ebola were encouraged to attend.
At the interview, individual informed written consent to participate in the study was sought from all adults, and from parents or guardians for children (< 18 years), with assent from children of 12 years or older. An inventory was drawn up of all household members who had been present in the household at the time that one or more household members were ill with EVD, including any who had died or were not present at the interview. For each member we asked whether they had had Ebola. We asked relatives whether any deceased had died of Ebola.
Household members were asked to describe what happened when Ebola came to their household, including who became ill first, whether those with Ebola had any diarrhoea, vomiting or bleeding while they were at home, and who looked after them. They were encouraged to tell the narrative in their own words, with probing questions to clarify who had been exposed and how. For each household member (including those who had died, but excluding any absent members or those who refused consent) we sought to establish the highest-risk exposure. Reported exposures were ranked a priori from highest to lowest as: contact with the body of someone who died of Ebola; direct contact with body fluids of someone with Ebola, including breastfeeding, or other direct contact with “wet” cases (i.e. those with diarrhoea, vomiting or bleeding); direct contact with “dry” cases (i.e. those without diarrhoea, vomiting or bleeding); indirect contact with a wet case (e.g. washing their clothes); indirect contact with a dry case; minimal contact (e.g. shared utensils); and no known contact. For each mother-baby pair who both had EVD we attempted to ascertain from the narratives who was affected first.
All survivors from the Kerry Town Ebola Treatment Centre had EVD confirmed by PCR. We did not have laboratory data for those from other treatment centres or for those who died, so have relied on the families’ reports. For individuals who were not reported as having had Ebola we asked about symptoms at the time that Ebola was in the household. For the analysis they were classified as not having had Ebola if they were asymptomatic or had symptoms that did not fulfil the Sierra Leone Ministry of Health and Sanitation case definition for “probable” Ebola,[10] or had had a negative test; and as having had Ebola if they were symptomatic and fulfilled the case definition for probable Ebola and were not tested. The case definition was contact with a case plus fever or miscarriage or unexplained bleeding; or contact plus three or more symptoms (of fatigue, headache, loss of appetite, nausea or vomiting, abdominal pain, diarrhoea, muscle or joint pain, sore throat or pain on swallowing, hiccups).
In this analysis we concentrate on risks to children aged less than three years at the time Ebola reached their household in order to include all those who were breast fed, and examine attack rates, case fatality rates and the role of breast feeding. Proportions were compared using Χ2 or Fisher’s exact test. Analyses used multivariable logistic regression. Because the outcome is very common we have presented the results as risk ratios (RR) using marginal standardization to estimate RRs, and the delta method to estimate 95% confidence intervals (95%CI).[11–13] We repeated the analysis calculating risk ratios using Poisson regression with robust error variance.[14] Crowding (number of people per room) and sanitation (access to water, soap and latrine) were considered as possible confounders, in addition to age, sex and the exposure variables. The effects of clustering by household were explored using generalised estimation equations in logistic regression: the results were very similar to analyses ignoring clustering so clustering is not included in the models. Analyses used STATA 14.
The study was approved by the Sierra Leone Ethics and Scientific Review Committee and the Ethics Committee of the London School of Hygiene & Tropical Medicine. At the interview, individual written informed consent to participate in the study was sought from all adults, and from parents or guardians for children (< 18 years), with assent from children of 12 years or older.
One hundred and fifty one survivors were discharged from Kerry Town Ebola Treatment Centre from November 2014 through March 2015, of whom 138 were still living in the Western Area of Sierra Leone when sought for interview in July-September 2015. Twelve were uncontactable and a further two were known to have bad relationships with their households so were not approached. We contacted and interviewed 123 Kerry Town survivors, living in 94 households. Only one contacted survivor refused to be interviewed, and only two of 526 household members refused to participate. A further 37 members were not available to attend the interview. Some households also included survivors who had been treated in other facilities.
The households contained 77 children aged less than three years: 43% (33/77) got EVD, including four who fitted the case definition but were not diagnosed at the time. The risk of EVD was 54% (13/24) in those under one year; 40% (12/30) in those aged one year and 35% (8/23) in those aged two years (p-value for trend = 0.2). The risk was slightly higher in males than in females: 51·4% (18/35) vs 35·7% (15/42), p = 0.2. Three of the children were primary or co-primary cases in their household. Overall, 24 children under three years died of EVD, giving a case fatality rate of 73%: 85% (11/13), 75% (9/12) and 50% (4/8) at ages under-one, one, and two respectively (p-value for trend = 0·1).
Among the 77 children were 13 whose mothers were not present (including two mothers who had died in other households), or were not clearly identified: six (46%) of these children developed EVD and five died compared to 27 cases (42%) and 19 deaths among the 64 children who could be linked to their mothers.
Details of the mother-child pairs for whom the outcome of both mother and child are known are shown in Table 1 for the 40 whose mothers had EVD, in Table 2 for the 24 whose mothers had no symptoms, and in summary for all 64 in Table 3. The highest level of exposure is shown, in terms of direct or indirect exposure to those with EVD in the home or outside. None of the children had direct contact with dead bodies. Breastfeeding was taken as the highest exposure if the mother had EVD unless the child developed symptoms before or at the same time as the mother.
EVD in the children was much more likely among those whose mother had EVD (25/40, 63%) than among those whose mother did not get EVD (2/24, 8%, risk ratio (RR) 7·5, 95% confidence interval (CI) 1.9–28.9, p<0.0001, Table 3). The RR remained high after adjusting for age and sex of the child (RR 9·4, 95% CI 2·6–34·0), and after additionally adjusting for maximum exposure level (RR 7·6, 95%CI 2·0–29·1). Household crowding and sanitation were not associated with EVD in the child, and adjusting for them made little difference to the results. After adjusting for mother’s EVD status and exposure levels, the risk of EVD in the child decreased with age (Table 3). After adjusting for mother’s EVD, age, and sex, there was no effect of exposure level.
Among those whose mother had EVD, excluding the two pairs in which the children were ill first, the risk of EVD in the child was higher if the mother died (79% vs 50%, Table 3), giving a relative risk of 1·6 (95% CI 0·97–2·6). This association was similar after adjusting for the child’s age and sex and additionally for exposure level. Of the 13 children who did not get EVD whose mother survived, five had contact with the mother when she was a wet case and five only when she was a dry case (unknown for three).
As the only child over two years who was breastfed got ill at the same time as the mother and was therefore excluded, the analysis of breastfeeding was restricted to the under two’s. We also excluded the child who became ill first (Table 1), leaving 26 children. The proportion of children with EVD was very similar in those who were or were not breast fed (69% vs 70%, Table 3), RR 0·98, 0·58–1·7. There was no evidence of increased risk from breastfeeding after adjusting for age and sex (RR 0·76, 0·46–1·2) or for whether the mother died (Table 3).
The analyses were re-run excluding the six mother-child pairs for which either the mother or the child was classified as having EVD on the basis of symptoms (Table 1). The associations with having a mother with EVD (fully adjusted RR 6.5, 1·6–26·0) and with breastfeeding (fully adjusted RR 0·74 (0·47–1·2) were similar to the main analysis, but the effect of having a mother who died of Ebola was lost (fully adjusted RR 1·3, 0·76–2·1). The analyses were also rerun using Poisson regression. The results were similar to the main analysis.
Among the children under three years whose mother did not get EVD, only two children got EVD. Both were aged under one year, from households with many EVD cases (Table 2), and both were reported to have had close contact with wet cases in the household. Seven other children whose mother did not have EVD and 4 whose mother had EVD but were not breastfed, had direct contact with wet cases and did not get ill. Overall, excluding children breastfed by mothers with EVD, half (11/22) of the children who had direct contact with wet cases or fluids remained well. These contacts included sharing beds with and embracing close relatives who suffered from vomiting and/or diarrhoea.
Among the very young children in this study the risk of EVD depended largely on whether their mother developed EVD, with an additional risk for those whose mothers died of Ebola. The high risk in those with sick mothers is expected, and the higher risk in those with mothers who died may reflect higher viral loads and/or viral shedding in these mothers. The low risk in children in Ebola-affected households when the mother was not ill is surprising, and cannot all be explained by low exposure in the children. Overall, nearly two thirds of under-three year olds had direct contact with wet cases in the household or their body fluids. While the risk of disease decreased with decreasing exposure, half of the young children with direct exposure to wet cases remained well.
Only three children were deliberately sent out of the household to reduce exposure, and for all three there was some exposure before they left. The opportunities for households to protect children from exposure are limited, particularly as more and more cases arise, and young children share beds with sick relatives. While a ‘no touch policy’ may be understood by older children, it is impossible to explain to an infant.
Among children whose mothers had EVD, being breastfed did not appear to increase the risk. Numbers were small and risks were already high in this group so there was limited power to detect an association. Current WHO guidelines recommend that asymptomatic breastfed infants of Ebola-infected mothers should be separated from their mothers and replacement fed.[15] Although we found no excess risk from breastfeeding, further studies, ideally with larger, pooled datasets, are needed to assess this further before suggesting any changes to the recommendation. The high risk from proximity to a sick mother supports the need for separation.
The children in this study all came from households with at least one survivor. This may mean small households and households with fewer cases are underrepresented, as there would be a lower chance for small households to include a survivor, and households in which all cases of EVD died are missed. This might underestimate the case fatality rate and overestimate attack rates, but should not bias the relative risks by age and exposure.
This study shows the remarkable resilience of some young children despite apparent exposure to Ebola. This could be dose-related—we do not know the actual viral exposure through contact or breastfeeding—but in other contexts some people seem to be infected from minimal exposures. Relative resistance to Ebola could be influenced by genetic factors,[16] though the correlation between infections in mothers and children is more likely to reflect exposure patterns than shared genes. It is possible that there is some protection through maternal antibody from breastfeeding (perhaps more in mothers who survive) that counteracts any increased risk from transmission via breastmilk.
This is much the largest study of mother-child pairs with EVD to date, and the first attempt to assess any excess risk from breastfeeding. By visiting households after transmission had ceased and talking to all members we were able to determine exposure in much more detail than is possible in an acute epidemic situation. And because we included all children in these households, including those who were not sick, we have been able to calculate age and exposure-specific attack rates. In these households the risk to young children was largely dependent on whether their mother had EVD, regardless of whether they were breastfed.
|
10.1371/journal.ppat.1006827 | The fungal myosin I is essential for Fusarium toxisome formation | Myosin-I molecular motors are proposed to function as linkers between membranes and the actin cytoskeleton in several cellular processes, but their role in the biosynthesis of fungal secondary metabolites remain elusive. Here, we found that the myosin I of Fusarium graminearum (FgMyo1), the causal agent of Fusarium head blight, plays critical roles in mycotoxin biosynthesis. Inhibition of myosin I by the small molecule phenamacril leads to marked reduction in deoxynivalenol (DON) biosynthesis. FgMyo1 also governs translation of the DON biosynthetic enzyme Tri1 by interacting with the ribosome-associated protein FgAsc1. Disruption of the ATPase activity of FgMyo1 either by the mutation E420K, down-regulation of FgMyo1 expression or deletion of FgAsc1 results in reduced Tri1 translation. The DON biosynthetic enzymes Tri1 and Tri4 are mainly localized to subcellular structures known as toxisomes in response to mycotoxin induction and the FgMyo1-interacting protein, actin, participates in toxisome formation. The actin polymerization disruptor latrunculin A inhibits toxisome assembly. Consistent with this observation, deletion of the actin-associated proteins FgPrk1 and FgEnd3 also results in reduced toxisome formation. Unexpectedly, the FgMyo1-actin cytoskeleton is not involved in biosynthesis of another secondary metabolite tested. Taken together, this study uncovers a novel function of myosin I in regulating mycotoxin biosynthesis in filamentous fungi.
| The mycotoxin deoxynivalenol (DON) is the most frequently detected secondary metabolite produced by Fusarium graminearum and other Fusarium spp. To date, relatively few studies have addressed how mycotoxin biosynthesis occurs in fungal cells. Here we found that myosin I governs translation of DON biosynthetic enzyme Tri1 via interacting with the ribosome-associated protein FgAsc1. Moreover, the key DON biosynthetic enzymes Tri1 and Tri4 are mainly localized to the toxisomes derived from endoplasmic reticulum under toxin inducing conditions. We further found that the FgMyo1-actin cytoskeleton was involved in toxisome formation but not for the biosynthesis of another secondary metabolite tested. Taken together, these results indicate for the first time that myosin I plays critical roles in mycotoxin biosynthesis.
| Fusarium head blight (FHB) caused predominately by Fusarium graminearum is an economically devastating disease of small grain cereal crops [1]. This disease not only reduces yield and seed quality but also poses a great risk to human and animal health owing to its ability to contaminate grains with mycotoxins. The common mycotoxins associated with F. graminearum are deoxynivalenol (DON), nivalenol (NIV) and zearalenone (ZEA) [2]. Among them, DON is the most frequently detected mycotoxin in cereal grains throughout the world [3]. DON can inhibit protein synthesis by binding to the ribosome, and cause emetic effects, anorexia and immune dysregulation as well as growth, reproductive and teratogenic effects in mammals [4]. To minimize human and animal exposure to DON, regulatory organizations have established maximum permissible levels for DON in cereals and their products in many countries [5, 6]. However, DON contamination has become a challenging social issue because of the increased frequency and severity of FHB epidemics [7, 8].
DON contamination is closely linked to the severity of FHB disease in the field. The best way to prevent DON contamination would be to manage FHB in the field during crop cultivation. Currently, application of chemical fungicides is still a major approach against F. graminearum infection due to the lack of highly resistant wheat cultivars. However, application of several commercialized fungicides at sub-lethal concentrations could trigger DON biosynthesis [3, 9–11]. Recently, a novel cyanoacrylate fungicide phenamacril (JS399-19) has been marketed for FHB management and sale of phenamacril in China was approximately $40 million in 2016–2017. Interestingly, this small molecule compound (S1 Fig) exhibits highly specific antifungal activity against mycelial growth of a few Fusarium species including F. graminearum, F. asiaticum, F. verticillioides and F. oxysporum but not other fungal pathogens [12]. It shows excellent efficacy in controlling FHB in field trials [12, 13]. Combining inferences from genetic and biochemical results, we recently discovered that this compound acts on a novel target, the class I myosin (FgMyo1) in F. graminearum, which is homologous to Myo3p and Myo5p in Saccharomyces cerevisiae [12]. FgMyo1 is essential for F. graminearum growth. At the beginning of this study, we found that phenamacril not only suppressed the mycelial growth of F. graminearum, but also significantly inhibited DON production. These preliminary results suggested that the myosin I might also be involved in the secondary metabolism. Class I myosins are widely expressed, single headed and membrane-associated members of the myosin superfamily that participate in regulating membrane dynamics and structure in nearly all eukaryotic cells [14, 15]. However, the underlying function of myosin I in mycotoxin biosynthesis was totally unknown.
Enzymes for secondary metabolite synthesis may be compartmentalized at conserved sub-cellular sites in fungi, potentially channeling precursors, sequestering intermediates and products from the rest of the cell, thus promoting the efficiency of biosynthesis pathways [16]. In Penicillium chrysogenum, the major facilitator-type secondary transporter PenM promotes translocation of isopenicillin N from the cytosol to the peroxisomal lumen where it could be further metabolized to penicillin [17]. In Aspergillus, aflatoxin biosynthetic enzymes flow from peroxisomes to the motile vesicles termed aflatoxisomes in which aflatoxin biosynthesis takes place [18]. In F. graminearum, and other Fusarium spp, the biosynthetic pathway leading from the isoprenoid intermediate farnesyl pyrophosphate to DON involves 15 genes encoding the biosynthetic enzymes, a DON transporter and regulatory proteins, which are located on different chromosomes: the 25 kb Tri5 cluster containing 12 genes on chromosome 2, the Tri1-Tri16 locus with two genes on chromosome 1 and the single gene locus for Tri101 on chromosome 3 [19–21]. Recent studies suggested that there is a cellular compartmentalization of biosynthetic enzymes for DON biosynthesis in F. graminearum [22]. Hydroxymethylglutaryl (HMG) CoA reductase (Hmr1) is a key enzyme in the mevalonate pathway for generating farnesyl pyrophosphate and indispensable for DON production. Fluorescent labeled Hmr1-GFP localized to the reticulate peripheral and perinuclear endoplasmic reticulum (ER) in toxin non-inducing conditions, while the ER was remodeled to form spherical and ovoid structures in the trichothecene biosynthesis inducing (TBI) conditions [16, 22]. In addition, the enzymes trichodiene oxygenase (Tri4) and calonectrin oxygenase (Tri1) catalyzing the early and late steps in the DON biosynthetic pathway were co-localized and showed the same localization patterns as Hmr1 in TBI medium [22, 23]. These novel cellular structures containing DON biosynthesis enzymes were named "Fusarium toxisomes" (“toxisomes” in shorter form in this study) [16, 22, 23]. However, the molecular mechanism of toxisome formation remains elusive.
The object of this study was to uncover the underlying mechanism of a myosin I inhibitor in regulating DON biosynthesis. Our results showed that myosin I plays critical roles in the translation of a Tri enzyme and in toxisome formation in F. graminearum. The importance of myosin I in the development of the mycotoxin biosynthetic machinery in F. graminearum may apply to other toxigenic pathogens.
TRI1 encodes calonectrin oxygenase that catalyzes calonectrin to 7, 8-dihydroxycalonetrin, which is a late step of DON biosynthesis in F. graminearum [24]. To characterize expression patterns and the sub-cellular localization of Tri1 protein under various conditions, the TRI1 open reading frame tagged with GFP (green fluorescent protein) was introduced into a ΔTri1 F. graminearum PH-1 background, and the complemented strain expressing the Tri1-GFP (ΔTri1::Tri1-GFP) was used in the following study. In the toxin non-induction minimal (MM) or potato dextrose broth (PDB) media, Tri1-GFP displayed faint signals and was mainly associated with cell endomembrane (Fig 1A, left and middle panels). Tri1-GFP was highly induced and localized at the spherical structures (toxisomes) after 48 hours of incubation in the trichothecene biosynthesis induction (TBI) medium (Fig 1A right panel; S2 Fig) and in planta (Fig 1G, left panel). In addition, ER (endoplasmic reticulum)-tracker red staining indicated that Tri1-GFP was mainly localized to the ER in TBI cultures (Fig 1B), which is consistent with a previous finding that the toxisomes were identified as reorganization of the endoplasmic reticulum [22]. To determine whether the spherical structures were associated with the nucleus, we visualized nuclei by tagging the histone1 protein encoded by the FGSG_10800 locus with red fluorescent protein (RFP), which was designated as H1-RFP in the PH-1::Tri1-GFP strain. The H1-RFP/Tri1-GFP dual labeled strain was grown in TBI for 48 h, and localization of H1-RFP with Tri1-GFP was examined. As shown in Fig 1C, Tri1-GFP surrounded the H1-RFP labelled nuclei when the strain was cultured in the TBI medium. Moreover, the trichodiene oxygenase (Tri4) catalyzing the early step of DON biosynthesis had the same localization pattern as Tri1 (S3 Fig). Taken together, several lines of evidence suggested that trichothecene biosynthetic enzymes were clustered and localized to toxisomes derived from ER under the toxin inducing conditions.
Since the toxisomes are important for DON biosynthesis, a compound disrupting the toxisome formation may be very well effective against DON biosynthesis. To test this hypothesis, we established a "toxisome formation inhibitor screening" assay to quickly screen active compounds for their ability to restrict toxisome formation (see Material and methods). Briefly, the reporter strain expressing Tri1-GFP was grown in 24-wells plates supplemented with TBI medium. After 24 h incubation, individual compounds were added to wells. After incubation for another 24 h, the fluorescent intensity in each well was scanned with the plate-reader for the first round of screening. The wells with low or no fluorescent signals were further observed by microscopy. A total of 131 compounds including 11 commercial fungicides were tested for their activity against toxisome formation. Phenamacril was found to be the most efficient compound to inhibit toxisome formation and DON production (S4B and S4C Fig). The Tri1-GFP fluorescent signals were reduced dramatically and no typical toxisomes were observed in the mycelia treated with 0.5 μg/ml (approximately EC90 against mycelial growth) phenamacril for 6 h (S5B Fig) or 24 h (Fig 1D) in comparison with those in the non-treatment control. In addition, the beta-tubulin inhibitor, carbendazim, did not inhibit toxisome formation (Fig 1D). As shown in Fig 1E, the translation levels of Tri1-GFP were further verified by immunoblot assay using an anti-GFP antibody. Consistent with the microscopic observation, the intensity of the Tri1-GFP band from the strain treated with carbendazim increased more than 2-fold as compared with the non-treated control. In contrast, a faint immunoblot band was detected in the same strain treated with phenamacril (Fig 1E). Correspondingly, DON in the mycelia treated with phenamacril was below the level detectable by LC-MS (liquid chromatography-mass spectrometer) (Fig 1F). Furthermore, we tested the efficiency of phenamacril against DON production in planta and in the field. As shown in Fig 1G, phenamacril also clearly inhibited toxisome formation in hyphae of F. graminearum inoculated on wheat leaf. In the field trials, this antifungal compound was very effective against FHB and DON production in comparison with the control chemical carbendazim (Fig 1H). The class I myosin (named FgMyo1) of F. graminearum has been identified as the target of phenamacril [12]. Taken together, these results strongly indicated that the myosin I inhibitor phenamacril was able to inhibit DON biosynthesis in F. graminearum.
Given that the myosin I inhibitor significantly reduces DON biosynthesis, myosin I may be critical for toxisome formation. In order to verify this, we tagged FgMyo1 with RFP to determine its subcellular localization. In toxin non-induction media MM and PDB, FgMyo1-RFP protein was detected as diffuse fluorescent signal in the cytoplasm, mainly localized at hyphal tips (Fig 2A). However, in the TBI medium, most FgMyo1-RFP fluorescence accumulated in subapical spherical structures (Fig 2A, right panel). To determine whether myosin I was localized to the toxisomes, a strain labeled with FgMyo1-RFP and Tri1-GFP was constructed and cultured in TBI. As indicated in Fig 2B, both proteins were mainly co-localized at the toxisomes. Additionally, Co-IP and BiFC (Bimolecular Fluorescence Complementation) assays showed that FgMyo1 interacted with Tri1 in toxin inducing condition (Fig 2C and 2D). Affinity capture mass spectrometry (ACMS) was then used to identify interacting proteins upon toxin-induction conditions using the dual tagged protein ZZ-Tri1-Flag as the bait. In the ACMS assay, FgMyo1 was captured by Tri1 (S1 Table). Furthermore, ten of the 30 Tri1-interacting proteins were described previously [22] as components of the toxisome, including the three cytochrome P-450 enzymes Tri1, Tri4 and Tri11 as well as HMG-CoA reductase. These results indicated that FgMyo1 interacts with Tri1 and thus has the potential for involvement in toxisome formation.
To verify the role of FgMyo1 in toxisome formation, we used a knock-down approach because FgMYO1 is an essential gene in F. graminearum [12]. First, we took the advantage of the RNA interfering (RNAi) pathway to induce FgMYO1 silencing with hairpin RNA (hpRNA), which has been proven to be efficient in knockdown of mRNA expression for target genes in F. graminearum [25]. The recombinant plasmid pSilent-FgMYO1, designed for generating the hpRNA of an FgMYO1 fragment (540 bp), was introduced into the wild-type PH-1. Predicting that transformants with reduced expression of FgMYO1 may grow poorly on the medium supplemented with phenamacril, we screened for transformants with increased sensitivity towards this compound and then verified the FgMYO1 expression level by reverse transcription-PCR. Among the 20 transformants tested, four showed increased sensitivity to phenamacril, and the expression levels of FgMYO1 were decreased 65%-90% in these silencing transformants in comparison with the wild type. The FgMYO1-S2 transformant, having the lowest FgMYO1 expression (10% of the parent strain PH-1), was selected for further characterization. It had normal growth rate on PDA but failed to grow on PDA supplemented with phenamacril at 0.3 μg/ml (approximately EC50 against mycelial growth) (Fig 2E). As expected, the toxisome formation indicated by Tri1-GFP was significantly impaired and only faint fluorescent signals were observed in the mycelia of FgMYO1-S2 harboring Tri1-GFP (Fig 2F). Next, we replaced the native promoter of FgMYO1 with the zearalenone (ZEA)-inducible promoter (Pzear) [26] to generate a transformant that conditionally expressed FgMYO1. The resulting transformant (termed as Pzear-FgMYO1) without ZEA induction was unable to grow on PDA supplemented with 0.3 μg/ml phenamacril (Fig 2E). Consistently, this strain formed very faint toxisomes in TBI without the inducer as compared to the wild type (Fig 2F upper panel). The defects in mycelial growth and toxisome formation of Pzear-FgMYO1 were partially recovered by adding the inducer β-estradiol (Fig 2E and 2F, upper panel). In addition, translation levels of Tri1-GFP protein in above strains quantified by the western blotting assay were consistent with fluorescent signals (Fig 2F, lower panel). All of the above mutants, whether constructed by silencing or conditional expression, revealed significantly reduced DON production in TBI (Fig 2G). Since DON is a critical virulence factor and plays a significant role in the spread of pathogen within host tissues [27–29], it follows that each of these strains was severely attenuated in virulence toward flowering wheat heads (S6 Fig). These results confirmed that FgMyo1 plays an important role in toxisome formation.
To gain an insight into the function of FgMyo1 in toxisome formation, we further conducted an ACMS assay using the dual tagged protein ZZ-FgMyo1-Flag as the bait. In the ACMS assay, the ribosome-associated protein Asc1 (hereafter named FgAsc1,) was captured by FgMyo1. Unexpectedly, FgAsc1 was also pulled down by Tri1 (S1 Table). In addition, the interaction of FgMyo1 and FgAsc1 was confirmed by Co-IP assay (Fig 3A, left panel), while the directly interaction between these two proteins was not verified by BiFC. Given that the translation level of Tri1-GFP protein was inhibited dramatically by phenamacril (Fig 1D), and Asc1 is a conserved ribosomal protein and is required for efficient protein translation [30,31], we inferred that FgMyo1 might regulate Tri1 translation via interacting with Asc1. To test this hypothesis, we first examined co-localization of FgMyo1-GFP and FgAsc1 tagged with RFP. In toxin non-inducing conditions, FgAsc1-RFP was detected as diffuse fluorescent signal in the cytoplasm (Fig 3B). However, in the TBI medium, most FgAsc1-RFP accumulated in spherical structures and co-localized with FgMyo1 (Fig 3A, right panel) and also with Tri1 at the perinuclear positions (Fig 3B, lower panel). Since Asc1 is ribosome-associated protein, to further visualize localization of ribosomes, FgRpL25 (an essential component of 60S subunit of ribosome [32, 33]) was tagged with mCherry under the control of its own promoter, and transformed into the wild type. Confocal microscopic examination showed that most FgRpL25-mCherry accumulated at the perinuclear positions in the toxin inducing conditions (S7 Fig, bottom panel). In contrast, FgRpL25- mCherry was mainly localized in the cytoplasm in the toxin non-inducing conditions (S7 Fig, upper panel). These results indicated that FgMyo1 interacts with the ribosome protein FgAsc1 in toxin inducing conditions.
To further understand the role of FgAsc1 in Tri1 translation, we constructed a deletion mutant of FgAsc1. As expected, the mutant exhibited dramatically reduced hyphal growth (Fig 3C, upper panel). The translation level of Tri1-GFP in this mutant was decreased markedly in comparison with that in the wild type (Fig 3D, right panel), and subsequently, toxisome formation and DON production was not detected in this mutant cultured in TBI medium (Fig 3D, left panel; Fig 3E). It is very interesting that the translation level of FK506-binding protein Fg_Fkbp54 (FGSG_01059) was not altered in ΔFgAsc1 as compared to that in the wild type (Fig 3F, right panel), suggesting that FgAsc1 controls the translation of some proteins (at least Tri1) but not all proteins, which is agreement with a role for Asc1 in regulating the translation of specific mRNAs in S. cerevisiae [31, 34]. Taken together, these results indicated that FgMyo1 was indispensable for translation of Tri1 protein by interacting with the ribosome protein FgAsc1.
In a previous study, we found that the ATPase activity of FgMyo1 is dependent on actin. FgMyo1E420K bearing a mutation at the actin interacting domain of FgMyo1, which caused the actin-activated ATPase activity of FgMyo1E420K was reduced to 5% as that of the wild-type FgMyo1 [12]. Correspondingly, toxisome formation in this strain was markedly decreased in comparison with that of the wild type (Fig 2E, upper panel). Moreover, we found that components of the actin cytoskeleton were enriched in the ACMS with Tri1 and FgMyo1 as the bait (S1 Table), suggesting that actin cytoskeleton may be associated with toxisome formation in F. graminearum. To address this possibility, we further constructed a strain bearing actin-RFP and Tri1-GFP. Then, the interaction between actin-RFP and Tri1-GFP was further verified by Co-IP assay (Fig 4A). In S. cerevisiae, the myosin I interacts with the actin and is required for polarization of the actin cytoskeleton [35]. Consistent with what is known in S. cerevisiae, actin was also associated with FgMyo1 in the ACMS assay using FgMyo1 as the bait (S1 Table). In addition, the interaction of FgMyo1-GFP and actin-RFP was further confirmed by the Co-IP assay (Fig 4B).
Since actin is essential for F. graminearum growth, we were unable to obtain a knockout mutant of the ACTIN gene. Thus, to further investigate the function of actin in DON biosynthesis, the actin polymerization inhibitor latrunculin A was used to mimic impaired function of the actin cables. After treatment with latrunculin A at 0.1 μg/ml (approximately EC90 against mycelial growth of F. graminearum), the typical toxisome structures could not be observed, and Tri1-GFP was detected as diffuse fluorescent signal in the cytoplasm (Fig 4C, left panel). In addition, Tri1-GFP was noticeably decreased in the western blot assay upon latrunculin A treatment (Fig 4C, right panel). Subsequently, latruncunlin A showed strong inhibition of DON production (Fig 4D). These results indicated that the actin cytoskeleton is involved in toxisome formation in F. graminearum.
In S. cerevisiae, Prk1 and End3 are involved in the organization of the actin cytoskeleton [36, 37]. To better understand the roles of the myosin I-actin cytoskeleton in toxisome formation, we therefore were interested in constructing deletion mutants of their orthologs FgPrk1 (FGSG_05586) and FgEnd3 (FGSG_09721). Toxisome formation in mycelia of these two gene deletion mutants harboring the tagged Tri1-GFP was examined. The Tri1-GFP signals decreased noticeably in both ΔFgPrk1 and ΔFgEnd3 mutants (Fig 5A, left panel). In addition, western blot assays confirmed the amount of Tri1-GFP protein in these mutants was considerably lower than that of the wild type under the toxin inducing condition (Fig 5A, right panel). Furthermore, these mutants produced significantly less DON as compared with the wild type (Fig 5B) and both mutants showed increased sensitivity to the myosin I inhibitor phenamacril (Fig 5C). Taken together, these results strongly indicated that the myosin I-actin cytoskeleton is essential for the toxisome formation in F. graminearum.
To test whether or not the myosin I-actin cytoskeleton is also necessary for biosynthesis of other secondary metabolites (SM), we examined aurofusarin biosynthesis because aurofusarin is a red polyketide pigment and easily visualized. As shown in Fig 6A, the FgMyo1 point mutation (FgMyo1E420K) and FgMYO1 knockdown mutants had similar red pigmentation in comparisons with the wild type PH-1, as well as ΔTri1 and ΔTri4 mutants after incubation for 3 days on PDA or 5 days in liquid PDB. Consistent with these observations, phenamacril and latrunculin A did not inhibit aurofusarin biosynthesis in the wild type (Fig 6A). As controls, deletion mutants of aurofusarin biosynthesis genes AurJ and AurF did not produce the red pigment (Fig 6A). These results suggest that the myosinI-actin cytoskeleton is dispensable for aurofusarin pigmentation. To further confirm this finding, the aurofusarin biosynthesis gene AurJ was tagged with RFP and transformed into the wild type bearing Tri1-GFP or the peroxisomal structural protein FgPex3-GFP. As indicated in Fig 6B (left panel), AurJ-RFP was mainly located in the cytoplasm and presented in a punctuate pattern that was different from the Tri1-GFP localization. However, AurJ-RFP was clearly co-localized with FgPex3-GFP. These results indicate that aurofusarin might be synthesized in peroxisomes. In addition, the cellular localization and fluorescent intensity of AurJ-RFP was not discernibly affected by treatment with phenamacril or latrunculin A (Fig 6C). In summary, the myosinI-actin cytoskeleton is not involved in aurofusarin pigmentation in F. graminearum.
Trichothecenes are synthesized from acetyl-CoA as the basic precursor though the isoprenoid intermediate farnesyl pyrophosphate (FPP) and ultimately the trichothecene biosynthesis pathway [38]. The enzymes Tri1 and Tri4 are delivered to the specific cellular compartment known as the toxisome under the DON induction condition (Fig 1A, S2 Fig). This process is largely dependent on various environmental factors or stimuli, including nitrogen and carbon sources [10, 11, 39], amines [40], pH [41], light [42], and reactive oxygen species (ROS) [3]. Accumulating evidence indicates that some fungicides also stimulate DON biosynthesis. Milus and Parsons reported that propiconazole and tebuconazole treatments could result in a 50% increase in DON contamination in field trials [9]. Application of fluquinconazole or azoxystrobin reduced disease incidence on wheat spikes but led to a significant increase in DON production by F. culmorum or F. graminearum in the harvested grains [10]. The fungicides epoxyconazole and propiconazole could also stimulate DON production in vitro and in wheat grains [11]. Therefore, the effects for disease management by application of fungicides may not be consistent with the impacts on mycotoxin biosynthesis. In this study, we tested the effect of 131 antifungal compounds on DON biosynthesis and found that phenamacril showed significant inhibition against DON biosynthesis. In agreement with previous studies, other fungicides including the carbendazim and azoles at sub-lethal concentrations could stimulate DON biosynthesis. Therefore, the chemical fungicides for FHB management should be carefully considered to avoid stimulating mycotoxin biosynthesis.
In eukaryotic cells, myosins participate in a wide variety of cellular processes, including cytokinesis, organellar transport, cell polarization, transcriptional regulation, intracellular transport, and signal transduction [43, 44]. They bind to the filamentous actin or other binding partners, and produce physical forces by hydrolyzing ATP, therefore converting chemical energy into mechanical force [12–14, 44, 45]. The conserved head domain is accompanied by a broad diversity of N-terminal or C-terminal domains that bind to different molecular cargos, providing the functional specificity of myosin proteins [46]. A total of 31 defined myosin classes have been identified in eukaryotes based on genomic surveys and phylogenetic analyses [15,46]. Three myosins: an essential class II myosin FgMyo2 (FGSG_08719), a class V myosin FgMyo2B (FGSG_07469), and the essential class I myosin FgMyo1 (FGSG_01410) are recognized in F. graminearum [47]. FgMyo2 is specifically localized to the delimiting septum of phialides and conidia, and required for septation [48]. In addition, the expression levels of TRI5 and TRI6 were obviously higher in the FgMyo2B heterokaryotic disruption mutant than those in the wild type [16,49]. These studies indicated that FgMyo2 and FgMyo2B may not be involved in mycotoxin biosynthesis directly. In the current study, we found that FgMyo1 is necessary for toxisome formation. Moreover, we further proved that FgMyo1 was not essential for the biosynthesis of the polyketide secondary metabolite, aurofusarin. These data suggest that the myosin I, but not other myosin motors, participates in DON biosynthesis in F. graminearum.
The cellular compartmentalization (toxisome) for DON biosynthesis in F. graminearum was first described though the dynamic localization of fluorescent labeled Tri1 and Tri4 [23]. More recently, the toxisome was further identified as reorganization of the endoplasmic reticulum with pronounced expansion at perinuclear-and peripheral positions [22]. Consistent with that, results of the current study further confirmed that Tri1 and Tri4 are often localized in the perinuclear ER under the toxin inducing condition (Fig 1B), and that the ER was remodeled from thin reticulate ER (S8 Fig) in the toxin non-inducing conditions to thickened ER in the TBI conditions (Fig 1A). In addition, the ER remodeling is further supported by accumulation of the perinuclear ribosomes under the TBI conditions (S7 Fig) since ribosomes are often attached the rough ER.
The toxisome structures were predicted to confer multiple beneficial biological functions including clustering of DON biosynthetic enzymes, promoting the efficiency of DON biosynthesis, as well as serving as a self-protection system against the self-toxicity of the Tri products and reaction intermediates [9, 22]. To date, four proteins including Tri1, Tri4, Tri14 and Hmr1 were validated to be localized to toxisomes [16, 22, 23]. However, the molecular mechanism for the ER remodeling to toxisome remains unknown. In eukaryotic cells, structures and functions of ER are dynamically changed by various intercellular and extracellular stimuli. For example, the ER network of Arabidopsis undergoes extensive remodeling, which is critically depended on a myosin-actin cytoskeleton system [50]. The plant specific myosin XI provides the force to propel ER streaming and the dynamic rearrangement of the ER network depends on the propelling action of myosin-XI over actin coupled with a SYP73-mediated bridging [51]. Since F. graminearum doesn’t contain a myosin XI homologous protein, we infer that the FgMyo1-actin cytoskeleton may be involved in the ER remodeling for toxisome formation in F. graminearum. This inference is supported by multiple lines of evidence. First, FgMyo1 is comprised of the motor domain that binds to and interacts with actin [12, 18], an isoleucine and glutamine (IQ) motif, and a C-terminal tail. The tail domain contains a pleckstrin homology (PH) motif that is known to bind the anionic phospholipids in cellular membranes (S9 Fig) [52, 53]. The presence of a lipid-binding domain in the tail and an actin binding region in the motor domain equips the myosin I for cellular roles that link membranes to the actin cytoskeleton [54]. Second, dysfunction of FgMyo1 and actin by inhibitors disrupts the toxisome formation (Figs 1D and 4C). Third, knockdown expression of FgMyo1 or the deletion of actin cytoskeleton organization related genes FgPrk1 and FgEnd3 resulted in a defect in toxisome formation and a reduction in DON production (Figs 2F, 5A and 5B). Finally, the point mutation FgMyo1E420K allowing only 5% of the wild-type ATPase activity also affected toxisome formation (Fig 2F), which is in agreement with the interpretation that the hydrolysis of ATP in FgMyo1 coverts the chemical energy into mechanical force and might provide the physical forces for ER remodeling.
In addition to providing the force for membrane dynamics, the myosin I motors have also been suggested to function as anchors or tethers between membranes and other proteins. In opossum kidney epithelial cells, Myo1b was found to tether amino acid transporters to the apical plasma membrane, thereby facilitating neutral amino acid transport across the membrane [55]. Similarly, Myo1a is important for the retention and localization of sucrose isomaltase in the intestinal brush border membrane [56]. Furthermore, the spatial association of nuclear myosin I with the ribosome protein S6 plays an important role in the export of small ribosomal subunits through the nuclear pores [57]. In current study, we found that FgMyo1 interacts with the ribosome-associated protein Asc1, thereby facilitating translation of toxin biosynthesis enzymes, and further contributing to toxisome formation in the toxin inducing conditions.
In eukaryotic cells, the myosin-actin system also plays important roles in endocytosis [58–60]. Consistent with that, deletion mutants of actin cytoskeleton organizing gene orthologs, Prk1 and End3 resulted in the defects in both endocytosis and toxisome formation in F. graminearum. However, the mutants of two conserved endocytic components (Apm4 and Abp1) still formed typical toxisomes in TBI (S10C Fig). Importantly, the FgMyo1E420K mutant that exhibits the defect in toxisome formation (Fig 2F) retains the capability of endocytosis (S10A Fig), while the actin-activated ATPase activity of FgMyo1E420K is very low (circa 5% as that of the wild-type FgMyo1) [12]. This finding is similar to a previous report that the Myo1 mutants of Aspergillus nidulans with no more than 1% of the actin-activated ATPase activity of wild-type Myo1 in vitro and no detectable in vitro motility activity can support fungal cell growth, albeit with a delay in germination time and a reduction in hyphal elongation [61]. Therefore, the myosin I mediated endocytosis process is not connected with the toxisome formation in F. graminearum.
The myosin-actin system also involves in the movement of organelles within cells, including the organelles for secondary metabolites organization. For instance, the short transportation of melanosomes for the skin pigment melanin biosynthesis at the peripheral region of the mammalian cell is largely dependent on the Rab27a, melanophilin, myosinV-actin filament complex [62]. In Fusarium spp. Tri12 is suggested to play a role in export of trichothecene mycotoxins, which forms vacuoles and vesicles during the mycotoxin inducing condition [20, 21]. A previous study suggested that Tri12 interacted with toxisomes and may transfer the trichothecenes from toxisomes into the vesicles and vacuoles for further export [23]. The motility of vesicles containing Tri12 was reversibly inhibited by latrunculin A, indicating that movement was dependent upon the filamentous actin [21, 23]. The motor proteins are needed for the cellular motility of Tri12 by mechanical driving force on the filamentous actin. There are three major super-families of motor proteins: kinesins, dyneins, and myosins. The first two act as motors on microtubule filaments, while myosins function on actin [63]. Thus, it would be interesting to further study the functions of myosins in the transport of toxins that may accumulate in Tri12-linked vacuoles and vesicles in F. graminearum and in other toxigenic fungi. Taken together, our data support a model in which FgMyo1 is essential for toxisome formation under the DON induction conditions in F. graminearum by interacting with FgAsc1 indirectly for regulating the Tri protein biosynthesis and by directly participating in the endoplasmic reticulum (ER) remodeling via the myosin-actin cytoskeleton system. In addition, the small molecule phenamacril is able to suppress the toxisome formation by inhibiting the ATPase activity of FgMyo1 (Fig 7).
The F. graminearum wild-type strain PH-1 (NRRL 31084) was used as a parental strain. The wild-type strain and transformants generated in this study were grown on potato dextrose agar (PDA) and minimal medium (MM) for hyphal examination. The carboxymethyl cellulose (CMC) liquid medium was used for conidiation assays [64]. For toxisome observation and trichothecene production analysis, each strain was grown in liquid trichothecene biosynthesis inducing (TBI) medium [38] at 28 °C in a shaker (150 rpm) in the dark. Each experiment was repeated three times.
The strains ΔFgPrk1, ΔFgEnd3, ΔFgTri1, ΔFgTri4, ΔFgAsc1, ΔFgAurJ and ΔFgAurF were constructed using the protocol described previously [65]. Briefly, the open reading frame (ORF) of each gene was replaced with hygromycin resistance cassette (HPH) and subsequent deletion mutants were identified by PCR assays with relevant primers (S2 Table). For complementation, each ORF fused with a tag and geneticin resistance gene was introduced into corresponding mutant, and transformants were selected with geneticin. To construct FgMyo1 silenced mutants, a 540 bp fragment was amplified and inserted forward and reverse into the pSilent-1 plasmid, and the recombination hairpin RNA silencing plasmid was introduced into PH-1 as previous described [25]. To replace the FgMYO1 promoter with Pzear, the HPH and Pzear fragments were amplified respectively and fused by overlap PCR. Subsequently, the “HPH-Pzear” fragment was further fused with the 5′ and 3′ flanking regions of the FgMYO1 gene. The resulting fusion fragment was purified and transformed into PH-1. To induce the Pzear replacement, the inducer β-estradiol at 30 μM was added to the medium during the regeneration and mutant selection processes [66].
To construct the FgTri1-GFP fusion cassette, the FgTri1 fragment containing the native promoter and ORF (without stop codon) was amplified with primers A15 + A16 (S1 Table). The resulting PCR products were co-transformed with Xho1-digested pYF11 into XK1-25. The alkali-cation yeast transformation kit (MP Biomedicals, Solon, USA) was used to generate the recombined FgTri1-GFP fusion vector. Subsequently, the FgTri1-GFP fusion vector was recovered from the yeast transformant by using the yeast plasmid extract kit (Solarbio, Beijing, China) and then transferred into E.coli strain DH5α for amplification. Using the same strategy, other GFP or RFP fusion cassettes were also constructed. Each recombination plasmid was transformed into PH-1 or the corresponding mutant for generating fluorescent label strains.
The strain expressing the FgTri1-GFP in the ΔTri1 background was used as the fluorescent reporter strain for anti-toxisome formation screening. The TBI medium supplemented with 104 conidia/mL was added into a 24-well plate (2.0 mL/well). After 24 h static incubation at 28 °C, each tested compound was added into a well and the plate was incubated for another 48 h. Then, the fluorescent intensity in each well was scanned with the Varioskan Flash Multimode Reader (Thermo Scientific, MA, USA) for first round screening. The wells with lower or no fluorescent signals compared with that of the control treatment (the same volume of solvent dimethyl sulfoxide, DMSO) were further observed by a confocal microscopy. A total of 131 antifungal compounds including 11 commercialized fungicides were tested for the activity against toxisome formation. For each compound, there were three-well replicates, and the experiment was repeated three times.
The fluorescent intensity and localization of tagged proteins were observed with a Zeiss LSM780 confocal microscopy (Gottingen, Niedersachsen, Germany). For observation of toxisome formation patterns in PH-1 and derived mutants, each strain labeled with FgTri1-GFP was cultured in TBI for 48 h before examination. All samples were mounted on glass slides and sealed with cover glasses. The following parameter sets of the confocal microscopy were used: Plan-Neofluar 40x/1.30 Oil DIC M27 objective; laser: at 488 nm at 50% power for green fluorescence; dimension of X = 70.78 μm, Y = 70.78 μm; pinhole: 90 μm; digital gain: 1.00. To observe toxisomes in planta, fresh mycelial plugs of the fluorescent reporter strain were inoculated on the leaves of wheat seedlings of a susceptible cultivar Jimai 22. After incubation at 25°C and 100% RH (relative humidity) for 5 days, the infected leaves were taken for toxisome examination observed under Plan-Neofluar 20x/0.50 M27 objective.
The following filter sets were used for other fluorescent or dye staining: the laser excitation wavelength was set at 405 nm for DAPI (blue fluorescence), at 561 nm for FM4-64 or RFP/mCherry (red fluorescence), at 514 nm for YFP (yellow fluorescence). The endoplasmic reticulum (ER) was stained with ER-Tracker Red (Beyotime technology Co., Ltd), and laser was set at 587 nm for red fluorescence. The intensity of fluorescence was acquired using the Zeiss ZEN 2010 software.
For BiFC assays, the final plasmid constructs of pYFPN-FgTri1 and pFgMyo1-YFPC were verified by sequencing and then co-transformed into the protoplasts of PH-1 in pairs. Transformants resistant to both hygromycin and neomycin were isolated and confirmed by PCR. The recombination plasmid pYFPN-FgTri1 or pFgMyo1-YFPC was transformed into PH-1, and resulted transformants were used as negative controls. YFP signals in the mycelia grown in TBI for 48 h were examined under a Zeiss LSM780 confocal microscope (Gottingen, Niedersachsen, Germany).
The protein isolation was performed as described previously [67]. The resulting proteins were separated by 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to Immobilon-P transfer membrane (Millipore, Billerica, MA, USA). The polyclonal anti-Flag A9044 (Sigma, St. Louis, MO) and monoclonal anti-GFP ab32146 (Abcam, Cambridge, UK) antibodies were used at a 1:5000 to 1:10 000 dilution for immunoblot analyses. The samples were also detected with monoclonal anti-GAPDH antibody EM1101 (Hangzhou HuaAn Biotechnology Co., Ltd.) as a reference. The intensity of immunoblot bands were quantified using the ImageQuantTL software.
To quantify the mycotoxin production, each strain was grown in TBI medium or inoculated on wheat kernels. DON was extracted, and then purified, and quantified using the LC-MS/MS system as described previously [5, 68].
The bait protein FgMyo1 was dual labeled with ZZ tag and 3×Flag at its N-terminus and C-terminus, respectively. The resulting fusion cassette was transferred into PH-1. The resulting transformant (PH-1::ZZ-FgMyo1-3×Flag) was used for protein extraction as previous described previously [65] and the affinity capture was conducted by the following procedures. After protein extraction, supernatant (25 ml) was transferred into a sterilized tube. The first run affinity capture was conducted using rabbit IgG agarose beads (Haoran Biotech Co., Shanghai, China), which was immuno-interacted with the ZZ tag. A total of 500 μl IgG agarose beads were added into the above supernatant to capture ZZ-FgMyo1-3×Flag interacting proteins, following the manufacturer’s instructions (General Electric Company, GA, USA). Then, the washed beads were subjected for the second run capture with anti-Flag agarose beads according to the manufacturer’s instructions (Abmart, NJ, USA). The final ZZ-FgMyo1-3×Flag interacting proteins captured by the anti-Flag agarose beads were eluted with TBS supplemented with 10% SDS. In addition, the ZZ-FgTri1-3×Flag was constructed and the interacting proteins were captured using the same strategy. The captured proteins were digested with trypsin and further analyzed by mass spectrometry using a previous published protocol [69]. Enrichment for proteins assigned to particular functional categories (FunCat) was calculated as described previously [20, 28].
The GFP, RFP, 3× Flag, or mCherry-fusion constructs were verified by DNA sequencing and transformed in pairs into PH-1. Transformants expressing pairs of fusion constructs were confirmed by western blot analysis. In addition, the transformants expressing a single fusion construct were used as references. For Co-IP assays, total proteins were extracted and incubated with the anti-GFP (ChromoTek, Martinsried, Germany) or anti-Flag (Abmart, Shanghai, China) agarose as described above. Proteins eluted from agarose were analyzed by western blot detection with a polyclonal anti-Flag A9044 (Sigma, St. Louis, MO), or an anit-GFP antibody (Abcam, Cambridge, UK). The protein samples were also detected with monoclonal anti-GAPDH antibody EM1101 (Hangzhou Huaan Biotechnology Co., Ltd.) as a reference. Each experiment was repeated twice.
|
10.1371/journal.pntd.0006344 | Experimental Chagas disease-induced perturbations of the fecal microbiome and metabolome | Trypanosoma cruzi parasites are the causative agents of Chagas disease. These parasites infect cardiac and gastrointestinal tissues, leading to local inflammation and tissue damage. Digestive Chagas disease is associated with perturbations in food absorption, intestinal traffic and defecation. However, the impact of T. cruzi infection on the gut microbiota and metabolome have yet to be characterized. In this study, we applied mass spectrometry-based metabolomics and 16S rRNA sequencing to profile infection-associated alterations in fecal bacterial composition and fecal metabolome through the acute-stage and into the chronic stage of infection, in a murine model of Chagas disease. We observed joint microbial and chemical perturbations associated with T. cruzi infection. These included alterations in conjugated linoleic acid (CLA) derivatives and in specific members of families Ruminococcaceae and Lachnospiraceae, as well as alterations in secondary bile acids and members of order Clostridiales. These results highlight the importance of multi-‘omics’ and poly-microbial studies in understanding parasitic diseases in general, and Chagas disease in particular.
| Host-parasite interactions are usually studied as a binary system, without considering the role of the host microbiota. This work integrates microbiome research into the study of gastrointestinal Chagas disease. We show that T. cruzi infection perturbs the fecal microbiome and metabolome, indicating functional changes affecting the gastrointestinal lumen. Our results support further investigation into the role of the microbiota-parasite interaction in gastrointestinal Chagas disease pathogenesis.
| Trypanosoma cruzi are protozoan parasites endemic to Central and South America. They cause a range of cardiac and gastrointestinal manifestations collectively known as Chagas disease. With increasing travel and immigration, infected individuals are also now found worldwide. Six to seven million people are T. cruzi-positive, thirty to forty percent of which will develop symptomatic disease decades after their initial exposure to the parasite. Cardiac symptoms are the most common; these include conduction abnormalities, arrhythmias, aneurysms, and heart failure leading to death. Clinically apparent gastrointestinal Chagas disease is less prevalent; gastrointestinal Chagas disease is associated with enlargement of the esophagus and/or colon (megaesophagus, megacolon), leading to pain, dysphagia, altered intestinal transit, altered nutrient intake, and constipation [1].
Research on cardiac Chagas disease progression has focused mainly on heart tissue. However, studies in murine models using luminescent T. cruzi cell lines showed recirculation of parasites from gastrointestinal tissues to the heart and propose a model in which gastrointestinal sites function as a reservoir for parasites to re-invade heart tissue and cause cardiac damage [2, 3]. These suggest an important role for intestinal T. cruzi infection beyond megasyndrome pathogenesis. Gastrointestinal sites may also be a major source of parasites during post-treatment recrudescence [4].
Gastrointestinal Chagas disease has a strong geographic association; most cases represent infections acquired in Bolivia, Brazil, Argentina and Chile. Disease tropism has been strongly tied to T. cruzi strain [5], but diet may also play a role [6]. T. cruzi infection is associated with parasite dose-dependent recruitment of inflammatory cells to the colon and colon damage [7], all of which could perturb the intestinal microbiota. Conflicting results comparing infection outcomes in germ-free and conventional mice have been reported, with one study showing similar survival [8], and another study showing differential survival [9]. The impact of T. cruzi infection on the gut microbiome and metabolome composition in immunocompetent animals has yet to be assessed. Such a system is more representative of human infection than germ-free models that show significant immunological defects [10]. This work applies 16S amplicon sequencing and mass spectrometry-based metabolomics on fecal pellets to characterize the functional bacterial changes associated with T. cruzi infection, in an immunocompetent murine model of Chagas disease. This joint approach enabled us to identify correlated microbiome and metabolome changes, and paves the way for further investigation of the T. cruzi-microbiota interaction in the context of Chagas disease pathogenesis.
All vertebrate animal studies were performed in accordance with the USDA Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Euthanasia was performed by isoflurane overdose followed by cervical dislocation. The protocol was approved by the University of California San Diego Institutional Animal Care and Use Committee (protocol S14187).
Male C3H/HeJ mice were purchased from Jackson laboratories and allowed to acclimatize to our vivarium for 2 weeks before the start of experimentation. At day 0, mice were infected by intraperitoneal injection with 1,000 red-shifted luciferase-expressing T. cruzi strain CL Brener culture-derived trypomastigotes [2] (20 mice across four cages) or left uninfected (injected with DMEM media only, 20 mice divided in four cages), and initial fecal pellets collected. Parasite burden was measured bi-weekly during the acute stage of infection by bioluminescence imaging following D-luciferin injection using an In vivo Imaging System (IVIS) Lumina LT Series III (Perkin Elmer). Total body luminescence, cardiac region luminescence, and abdominal luminescence were determined using Living Image 4.5 software. Fecal pellets were collected by monitoring the mice until they defecated naturally, at which point the freshly excreted pellets were immediately collected and snap-frozen in liquid nitrogen. Fecal pellets were collected bi-weekly in the acute stage of disease; imaging and fecal collection were performed every 2–3 weeks during the chronic stage of disease. Each time point was analyzed individually; no samples were pooled.
No visual changes were observed at any time point for fecal pellets from infected mice compared to fecal pellets from uninfected mice. Infected mice showed no overt disease symptoms except slight decrease in weight at the last two collection timepoints (days 64 and 90, p<0.05, Mann-Whitney, FDR-corrected) (S1A Fig), although four mice were found dead over the course of the experiment (days 20, 63, 79 and 90 post-infection) (S1B Fig). Hematoxylin-eosin (H&E) staining of colon samples did not show any apparent tissue damage or inflammatory infiltrate in infected mice compared to uninfected mice (S1D Fig). However, parasite distribution through the gastrointestinal tract is highly localized during chronic stage of infection with luminescent CL Brener [2], and we cannot rule out the possibility that other colon regions were altered by infection.
Weighed fecal pellets were homogenized in 50% methanol spiked with 2 μM sulfachloropyridazine using a Qiagen TissueLyzer at 25 Hz for 5 min [11], at a constant concentration of 50 mg feces / 1000 μL of extraction solvent, followed by overnight incubation at 4°C. Samples were then centrifuged at 16,000g for 10 min. Equal volumes of centrifugation supernatant were dried in a vacuum concentrator and frozen at -80°C. For LC-MS/MS analysis, samples were resuspended in 50% methanol spiked with 2 μM sulfadimethoxine and analyzed on a Maxis Impact HD QTOF mass spectrometer (Bruker Daltonics) coupled to an UltiMate 3000 UHPLC system (Thermo Scientific). A given infected or uninfected mouse was randomly assigned to one of eight 96 well plates, alternating infected and uninfected samples. Time-course samples were plated left to right in the 96 well plates, while run order was top to bottom. Controls included blanks (resuspension solvent) and pooled QC controls every 16 samples, and a standard mix of six compounds (sulfamethazine, sulfamethizide, sulfachloropyridazine, sulfadimethoxine, amitryptiline and coumarin) with known retention time at the beginning of the run and between each plate, to monitor for retention time shifts.
Liquid chromatography separation was performed on a 1.7 μm C18 (50 × 2.1 mm) UHPLC column (Phenomenex) heated to 40°C, with water + 0.1% formic acid as mobile phase A and acetonitrile + 0.1% formic acid as mobile phase B, at a constant flow rate of 0.5 mL/min. The LC gradient was: 0–1 min, 5% B; 1–9 min linear ramp up to 100% B; 9–11 min hold at 100% B; 11–11.5 min ramp down to 5% B; 11.5–12.5 min hold at 5% B.
Ions were generated by electrospray ionization and MS spectra acquired in positive ion mode with the following instrument parameters: nebulizer gas pressure, 2 Bar; Capillary voltage, 3,500 V; ion source temperature, 200°C; dry gas flow, 9.0 L/min; spectra rate acquisition, 3 spectra/s. MS/MS data was collected by fragmentation of the five most intense ions, in mass range 50–1,500 m/z, with active exclusion after 2 spectra and release after 30s. Mass ranges representing common contaminants and the lock masses were also excluded (exclusion list 144.49–145.49, 621.00–624.10, 643.80–646.00, 659.78–662.00, 921.00–925.00, 943.80–946.00, 959.80–962.00). Ramped collision-induced dissociation energy parameters ranged from 10–50 eV. Daily calibration was performed with ESI-L Low Concentration Tuning Mix (Agilent Technologies). Hexakis(1H,1H,3H-tetrafluoropropoxy)phosphazene (Synquest Laboratories), m/z 922.009798, was present throughout the run and used as internal calibrant (lock mass).
LC-MS/MS raw data files were lock mass-corrected and converted to mzxml format using Compass Data analysis software (Bruker Daltonics). MS1 feature identification was performed using an OpenMS-based [12] workflow (Optimus version 1.1.0 https://github.com/alexandrovteam/Optimus, see S1 Table for parameters), restricting to features with MS2 data available. Feature abundance was normalized to the sulfachloropyridazine extraction control. Principle coordinates analysis (PCoA) was performed on the normalized data with our in-house tool ClusterApp using the Bray-Curtis-Faith dissimilarity metric [13, 14], and visualized in EMPeror [15]. Molecular networking was performed on the Global Natural Products Social Molecular Networking platform (GNPS) [16], with the following parameters: parent mass tolerance 0.02 Da, MS/MS fragment ion tolerance 0.02 Da, cosine score 0.6 or greater, at least 4 matched peaks, maximum analog mass shift, 200 Da. Molecular networks and correlation networks were visualized with Cytoscape 3.4.0 [17]. Most metabolites were identified to levels 2/3 according to the 2007 metabolomics standards initiative (putatively annotated compounds or compound classes [18]). Additional putative annotations were performed using the LIPID MAPS m/z search tool [19]. Linoleic acid/conjugated linoleic acid (LA/CLA) were identified with higher confidence by retention time and spectral matching to authentic standards (Spectrum Chemical/Sigma Aldrich; level one annotation [18]). Random forest analysis over 5,000 trees was performed in R [20].
DNA extraction, 16S library preparation and sequencing were performed according to standard protocols from the Earth Microbiome project (http://www.earthmicrobiome.org/protocols-and-standards/ [21]). Briefly, DNA extraction was performed using the MO BIO PowerSoil DNA Isolation Kit (MoBio Laboratories). PCR amplification targeting the V4 region of the 16S rRNA bacterial gene was performed with barcoded primers 515F/806R as described in [22]. Equal amounts of amplicons from each sample were pooled in equal concentration and cleaned with the MoBio UltraClean PCR Clean-Up Kit. Library was PhiX-spiked and sequenced on the UC San Diego Institute for Genomic Medicine Illumina MiSeq2000 platform.
Raw FASTQ data files were demultiplexed using Qiita (https://qiita.ucsd.edu, study ID 10767) with the following parameters: maximum barcode errors: 1.5; sequence maximal ambiguous bases: 0; maximal bad run length: 3; Phred quality threshold: 3. This resulted in 12,307,767 high-quality reads with a median of 24,578 sequences per non-blank sample. Closed-reference Operational Taxonomic Unit (OTU) picking was performed in Qiita with 97% sequence identity using sortmeRNA [23] as the clustering algorithm. Subsequent data analysis was performed using the QIIME1 pipeline [24], rarefying to 8,500 reads per sample. PCoA plots were generated using the weighted UniFrac distance metric [25] and visualized in EMPeror [15]. Random forest analysis over 5,000 trees [20] was performed in R using jupyter notebooks [26].
Procrustes analyses [27, 28] were performed using the QIIME1 [24] scripts beta_diversity.py to generate the weighted UniFrac distance matrix (16S data) or Bray-Curtis-Faith distance matrix (LC-MS data), followed by principal_coordinates.py to perform principal coordinates analysis. PCoA outputs were used as input for transform_coordinate_matrices.py (Procrustes), with 1000 random permutations. The output of this analysis was visualized EMPeror [15].
Groups of bacteria and metabolites correlated with infection status were identified by Weighted Correlation Network Analysis (WGCNA) analysis. Average hierarchical clustering using the WGCNA R package in combination with soft-thresholded Pearson correlation was performed to independently cluster highly correlated microbes and metabolites into modules [29]. Data was pre-filtered using the goodSampleGenes function of the WGCNA package to remove metabolites or OTUs with >50% missing values. Remaining outlier samples were removed using the cutreeStatic function, with a minimum size of 10. Soft thresholding power was determined using the pickSoftThreshold function and set to 4 for metabolites and for 30 microbiome data. Minimum module size was 30 for OTUs and 10 for metabolite features; threshold for merging modules was 0.25. Using this approach, 49 metabolite modules and three microbial modules were obtained. Microbial and chemical modules were independently correlated with parasite burden using Pearson correlation. Since we were interested in identifying the changes in gut ecosystem due to parasite infection specifically, only the modules that were significantly correlated with parasite burden were retained for downstream analysis (Student asymptotic p-value <0.01; positive correlation coefficient). This represented nine metabolite modules and one microbial module. We performed pairwise Pearson correlation between these modules, which yielded six positively correlated microbe-metabolite module pairs (Student asymptotic p-value <0.01). Finally, we performed pairwise Pearson correlations between microbial and chemical components of these strongly correlated module pairs to obtain candidate microbial-metabolite associations relevant to T. cruzi infection in mice (positive correlation, p<0.05). These metabolites were then compared with molecular networking results to identify common members of chemical families. Correlations between the members of these families and bacterial OTUs were plotted using Cytoscape 3.4.0 [17].
T. cruzi infection perturbs the fecal microbiome and metabolome
To determine the impact of T. cruzi infection on the fecal metabolome and microbiome, we followed C3H/HeJ mice infected with a bioluminescent T. cruzi strain for 3 months post-infection. In this system, abdominal parasite burden peaked at 35 days post-infection (Fig 1A). Fecal samples were collected twice a week during the acute stage and every 2–3 weeks during the chronic stage of disease. Fecal bacterial operational taxonomic units (OTUs) were identified by sequencing of the V4 hypervariable region of the 16S rRNA genes [30]. Untargeted mass spectrometric analysis of the collected fecal pellets was performed by liquid chromatography-mass spectrometry followed by molecular networking for metabolite identification [16]. Detected identifiable metabolites include known products specifically from gastrointestinal microbes (secondary bile acids, tryptophan metabolites…), metabolites that can be found in the diet and/or modified by gut microbes (e.g. conjugated linoleic acid and derivatives generated from dietary linoleic acid) as well as common host and microbial metabolites (amino acids, phospholipids…). Overall bacterial composition was strongly affected by parasite burden (Fig 1B), as was the overall fecal metabolome (Fig 1C). Interestingly, infected to uninfected average distances reached their maximum before peak parasite burden (S2 Fig), with the best discriminatory ability between infected and uninfected samples at day 21 post-infection for both metabolome and microbiome (Fig 1D and 1E).
Gut microenvironments are influenced by dietary components and by bacterial and host metabolism, all of which could affect parasite nutritional availability and antiparasitic immune responses [31]. Likewise, chemical changes in the gut microenvironment would influence bacterial growth and composition [32]. We therefore investigated the integration between the microbial and chemical changes we observed during experimental T. cruzi infection by performing Procrustes analysis [27, 28]. Separation between infected and uninfected fecal microbiome and metabolome samples jointly was observed at days 21 and 90 post-infection but not at day 0 (Fig 2A, S2 Table).
To determine the nature of these joint changes, we performed weighted gene co-expression network analysis (WGCNA) [29, 33] on microbial and chemical data. Metabolites and microbes were individually clustered into modules, and microbial and chemical modules correlated with abdominal parasite burden (abdominal luminescence) were identified (significance cutoffs: Student asymptotic p-value <0.01; correlation coefficients > 0). Only one module of 1954 bacterial OTUs (out of three bacterial modules) was correlated with parasite burden (Pearson correlation coefficient, 0.19; Student asymptotic p-value, 3e-05), S3 Fig). Nine metabolite modules (out of 49 metabolite modules) were correlated with parasite burden (Student asymptotic p-value <0.01, Pearson correlation coefficient 0.13–0.33, S4 Fig). Pair-wise correlation was then performed between these burden-correlated microbial and chemical modules, six of which showed statistically significant correlation (Student asymptotic p-value<0.01, Pearson correlation coefficient 0.13–0.52, S5 Fig). Metabolite feature to OTU pair-wise comparisons were then performed within each metabolite-microbe module pair (cutoffs: positive correlation, p<0.05).
Within these six correlated module pairs, almost all the metabolite features positively correlated with parasite burden were from different molecular subnetworks, suggesting that they are part of different chemical families [16]. Strikingly however, eleven metabolite features from the most strongly correlated metabolite module (Pearson correlation coefficient 0.33, p-value, 3e-13) were from the same molecular subnetwork of linoleic acid derivatives (Table 1, S6A and S7 Figs). Dietary linoleic acid (LA) is modified in the gut environment by bacteria from the genera Lactobacillus, Bifidobacterium and Enterococcus into conjugated linoleic acid (CLA) and further derivatives [34, 35]. Conjugated linoleic acid can also be taken up in the diet and further modified in the gastrointestinal tract [34–36]. m/z 281.251 RT 485s was confirmed as LA or CLA by retention time and accurate mass matching to authentic LA/CLA standards (level one annotation according to the 2007 metabolomics standards initiative [18]; S6B Fig). Our chromatography conditions do not enable clear differentiation of LA and CLA. Specific members of the orders Bacteroidales and Clostridiales, including members of the families Ruminococcaceae and Lachnospiraceae can hydrogenate CLA [37, 38], and indeed we observed the strongest correlation (Pearson correlation coefficient >0.4) between members of the order Clostridiales and m/z 283.266 RT 435s, putatively identified as vaccenic acid (Fig 2B, S3 Table). Microbial hydration of linoleic acid by members of the Pediococcus and Lactobacillus genera has also been reported [35, 37, 39–41]. Molecular networking indicates that m/z 299.261 RT 336s and m/z 317.271 RT 336s could represent single and double hydration products of linoleic or conjugated linoleic acid; they were correlated with specific Ruminococcaceae and Lachnospiraceae family members (Fig 2B, S3 Table). CLA absorption in the colon is limited; bacterial metabolites of linoleic acid therefore primarily exert their effects locally [42]. Linoleic acid metabolism products alter gut inflammatory responses, by promoting regulatory T cell recruitment [43], decreasing TNF receptor expression [39] and TNFα production [44], and increasing the anti-inflammatory cytokine TGFβ in the colon [44]. These metabolites could therefore promote gut microenvironments favoring T. cruzi persistence and gastrointestinal reservoir function.
An additional group of 5 co-modulated features networked with cholic acid (Table 2, S8 Fig). m/z 357.281 RT 337s, m/z 357.281 RT 371s and m/z 375.291 RT 393s are identified as different close isomers or adducts of deoxycholic acid (level two annotation according to the 2007 metabolomics standards initiative [18]). Host-produced primary bile acids such as cholic acid are conjugated to taurine or glycine in the liver. Further modifications of primary bile salts are specifically performed in the gastrointestinal environment: members of the gut microbiota deconjugate primary bile salts and remove the 7-hydroxy group to form secondary bile acids such as deoxycholic acid. Bacteroides, Bifidobacterium, Clostridium, Lactobacillus and Listeria genera deconjugate bile acids, which are then dehydroxylated by Clostridium and Eubacterium genera [45]. Indeed, one member of the Clostridium genus, Clostridium celatum (OTU ID 4315688) was correlated with m/z 357.281 RT 337s (Pearson correlation coefficient 0.21028, p-value = 0.00000426), and weakly correlated with m/z 357.281 RT 371s and m/z 358.285 RT 371s (respective correlation coefficient, 0.09079 and 0.09770; respective p-values, 0.049161208 and 0.034207271) (S4 Table). Likewise, two members of the Bifidobacterium genus were correlated with m/z 357.281 RT 371s and m/z 375.291 RT 393s, and five members of the Lactobacillus genus were correlated with m/z 357.281 RT 371s, m/z 358.285 RT 371s and m/z 375.291 RT 393s (S4 Table). Further modifications can be performed by these genera and by Escherichia, Egghertella, Fusobacterium, Peptococcus, Peptostreptococcus, Ruminococcus genera [45], several of which were also correlated with our infection-modulated secondary bile acids (Fig 2C). The OTUs most strongly correlated with these secondary bile acids in our experiment (correlation coefficient >0.4) were also members of the order Clostridiales, either from the genus Oscillospira or from unidentified genera (S4 Table). Bile acid metabolism by the gut microbiota has been tied to local colon inflammation and general health [45], all of which could affect Chagas disease pathogenesis.
Several of these microbiome changes have been associated with other gastrointestinal diseases. Lactobacillus genus in particular is increased in obese individuals, while genus Bifidobacterium is decreased [46]. Members of the Lactobacillus genus and some Bifidobacterium species are increased in ileal Crohn’s disease, while members of order Clostridiales and family Lachnospiraceae are decreased [46]. Large-scale perturbations are also observed in these diseases, such as for example a trend for increased Firmicutes to Bacteroidetes ratio in obese individuals compared to lean individuals [46]. The observed microbial and metabolic perturbations in T. cruzi-infected animals may be a consequence of parasite-mediated modulations of local gastrointestinal microenvironments, such as nutrient depletion, or an off-target effect of anti-parasitic immune responses. Parasite control is associated with reactive oxygen and nitrogen species [47], which are known to affect the gut microbiome composition by killing bacterial species sensitive to oxidative stress while promoting the growth of species that use nitrate as a terminal electron acceptor for respiration [48]. Significant bacterial and metabolic changes become apparent by day 14 post-infection (Figs 1D and 1E and S2), which coincides with induction of adaptive immune responses to T. cruzi [49], suggesting an immune-mediated role in this disruption.
Given the anti-inflammatory roles of the hydrated linoleic acid metabolites we found altered by infection [39, 43, 44], the gut microbiome and metabolome changes we observed may be promoting long-term parasite gastrointestinal persistence and enabling the gastrointestinal tract to serve as a parasite reservoir. Microbiota perturbation may also contribute to the nutrient malabsorption and constipation observed in megasyndromes [50]. Modulating the infection-associated changes in the gut microbiome and its metabolism may prove to be an effective way to mitigate disease symptoms, nifurtimox gastrointestinal side effects or prevent parasite dissemination from the gastrointestinal tract to the heart. Modifying the levels of anti-inflammatory conjugated linoleic acid metabolites may be particularly useful in this context. Finally, although production of bile acid metabolites is performed in the gut environment by the local microbiota, these metabolites can be re-absorbed and circulate throughout the body, with far-ranging effects [51]. Bile acid metabolites may therefore also affect cardiac Chagas disease pathogenesis. Future work will directly investigate these possibilities, by testing whether the gut microbiome perturbations and the metabolites identified in this study are associated with Chagas disease severity, and assessing whether microbiome perturbation affects Chagas disease progression.
Research on Chagas disease pathogenesis has focused on the interaction between the mammalian host and the parasite. Our results indicate that infection modulates the fecal microbiome, suggesting that host-microbe interaction research in the context of Chagas disease should also include the microbiota and not just T. cruzi. By integrating microbiome with metabolome data, we show that these microbial alterations are associated with functional changes in the gut chemical environment that could be affecting host inflammatory responses. These results support additional investigation into the T. cruzi-microbiota connection and into the role of the microbiota in Chagas disease pathogenesis. Given new evidence on the role of gastrointestinal persistence in parasite recrudescence [4], and our limited understanding of gastrointestinal Chagas disease compared to cardiac Chagas disease, such studies are essential to identify treatment strategies able to achieve sterile cure. Microbiota- and microbial metabolism-modulating therapies are now actively being developed for other cardiovascular diseases [52, 53]. Our results demonstrate that such approaches are likely to be beneficial in cardiovascular Chagas disease. Modulation of the gut microbiota or its metabolism may also be a promising strategy for megasyndrome patient management, or to slow progression of asymptomatic individuals to symptomatic disease.
|
10.1371/journal.ppat.1000147 | The Cysteine-Rich Interdomain Region from the Highly Variable Plasmodium falciparum Erythrocyte Membrane Protein-1 Exhibits a Conserved Structure | Plasmodium falciparum malaria parasites, living in red blood cells, express proteins of the erythrocyte membrane protein-1 (PfEMP1) family on the red blood cell surface. The binding of PfEMP1 molecules to human cell surface receptors mediates the adherence of infected red blood cells to human tissues. The sequences of the 60 PfEMP1 genes in each parasite genome vary greatly from parasite to parasite, yet the variant PfEMP1 proteins maintain receptor binding. Almost all parasites isolated directly from patients bind the human CD36 receptor. Of the several kinds of highly polymorphic cysteine-rich interdomain region (CIDR) domains classified by sequence, only the CIDR1α domains bind CD36. Here we describe the CD36-binding portion of a CIDR1α domain, MC179, as a bundle of three α-helices that are connected by a loop and three additional helices. The MC179 structure, containing seven conserved cysteines and 10 conserved hydrophobic residues, predicts similar structures for the hundreds of CIDR sequences from the many genome sequences now known. Comparison of MC179 with the CIDR domains in the genome of the P. falciparum 3D7 strain provides insights into CIDR domain structure. The CIDR1α three-helix bundle exhibits less than 20% sequence identity with the three-helix bundles of Duffy-binding like (DBL) domains, but the two kinds of bundles are almost identical. Despite the enormous diversity of PfEMP1 sequences, the CIDR1α and DBL protein structures, taken together, predict that a PfEMP1 molecule is a polymer of three-helix bundles elaborated by a variety of connecting helices and loops. From the structures also comes the insight that DBL1α domains are approximately 100 residues larger and that CIDR1α domains are approximately 100 residues smaller than sequence alignments predict. This new understanding of PfEMP1 structure will allow the use of better-defined PfEMP1 domains for functional studies, for the design of candidate vaccines, and for understanding the molecular basis of cytoadherence.
| Malaria parasites express proteins of the erythrocyte membrane protein-1 family (PfEMP1) on the surfaces of the human red blood cells that they infect. These large proteins vary in sequence extensively, yet bind to host receptors to allow infected cells to adhere to host tissues. PfEMP1 proteins help parasites evade the immune system, as the 60 PfEMP1 genes are expressed one at a time. Sequence comparisons predict that PfEMP1 molecules are modular, made up of Duffy binding-like (DBL) and cysteine-rich interdomain region (CIDR) domains. Many CIDR domains bind to the human receptor CD36. We have analyzed the structure of the CD36-binding portion, known as MC179, of a CIDR domain. The MC179 protein is composed of a bundle of three helices connected by a loop and three additional helices. Based on the structure and sequence similarities, MC179 is a good model for the hundreds of known CIDR sequences. In addition, the MC179 three-helix bundle is remarkably similar to subdomain 3 of the known DBL structures. MC179 provides insight into the relatedness of both kinds of PfEMP1 domains and predicts that the large PfEMP1 molecules are polymers of three-helix bundles and their connecting polypeptides.
| Cycles of exponential parasite growth inside infected red blood cells (iRBCs), followed by lysis, and immediate re-invasion of uninfected RBCs are the hallmarks of malaria caused by P. falciparum. Though seemingly hidden from the human immune system inside the iRBC, P. falciparum expresses a number of proteins on the iRBC surface, extensively remodeling the iRBC, and exposing parasite antigens to the immune system (for reviews [1]–[6]). Major antigens on the iRBC surface include members of the P. falciparum erythrocyte membrane protein-1 (PfEMP1) family, highly polymorphic and modular proteins composed of DBL (Duffy binding-like) and CIDR (cysteine-rich interdomain region) domains [7]. On each iRBC, a single PfEMP1 is expressed from one of ∼sixty var genes [8],[9] in the parasite genome. A different var gene can be activated during each cycle of infection, resulting in a new surface PfEMP1 molecule, and the parasite's evasion of antibodies against the previously expressed PfEMP1 [10].
Var genes undergo frequent recombination and gene conversion giving rise in the parasite population to a vast number of variant PfEMP1 proteins [11] that are under immune system selective pressure. PfEMP1 molecules are thought to be constrained in sequence only to maintain binding to host receptors and thus to mediate cytoadherence [12]. Cytoadherence prevents iRBCs from passing through the spleen and thus being cleared from the circulation [13]. Adherence of iRBCs to uninfected RBC (known as rosetting) and adherence to the microvasculature (known as sequestration) are thought to be primary causes of morbidity, especially in cerebral and placental malaria. iRBCs adhere due to the binding of PfEMP1 to a variety of endothelial receptors among which are CD36, intercellular adhesion molecule-1 (ICAM-1), and chondroitin sulfate A (CSA). Almost all parasites isolated directly from patients have the capacity to bind to CD36 [14]. iRBC appear to modulate immune system responses by their binding to CD36 and to other receptors on immune cells in the blood [15]–[17].
With the availability of genomic sequences from several P. falciparum strains, the numbers of DBL and CIDR domains and the extents of their diversity are becoming better known [18],[19]. Within the ∼sixty PfEMP1 sequences of the 3D7 strain of P. falciparum, there are 130 DBL and 110 CIDR domains [8],[20]. Though sequence identities among DBL and among CIDR domains are low, short blocks of conserved residues enable the classification of domains into families [21]. PfEMP1 domain families have been named DBLα, DBLβ, DBLγ, DBLδ, DBLε and CIDRα, CIDRβ, CIDRγ [18] and are also consistent with the groupings that arise from sequence similarities as performed by ClustalW [19]. Family groupings are used together with the position of the domain within its particular PfEMP1 to yield domain designations such as DBL1α, CIDR1α, CIDR2β, and so on. As an example, Figure 1A depicts the PfEMP1 molecule from the Malayan Camp (MC) strain of P. falciparum that consists of four DBL and two CIDR domains [22]. Correlations of domain families with specific ligand binding have been determined [23],[24]. In the case of the CIDR families, many domains have been tested for their binding to CD36. From these experiments, only CIDR1α domains bind CD36; other CIDR domains, even the closely related CIDR1α1 domains, do not [21],[25],[26].
Binding to CD36 by RBCs infected with the MC strain of P. falciparum has been studied [27],[28]. Experiments using labeled PfEMP1 proteins showed that a proteolytically-cleaved portion of PfEMP1 is capable of binding to CD36. Antibodies against a recombinant protein corresponding to residues 576 to 808 of the CIDR1α domain of MC PfEMP1 block adherence of iRBCs to CD36 [29]. Further analysis of recombinant portions of the MC PfEMP1 molecule identified a 179-residue polypeptide, MC179 (residues 576 to 754), as a minimal portion of the MC CIDR1α domain that retained binding to CD36 [30]. Recombinant MC179 competes with iRBCs for binding to CD36 under blood flow conditions, resulting in the release of bound RBCs [31],[32]. CIDR1α domains have been divided into three subdomains: M1, M2, and M3 [33]. MC179 corresponds to the M2 subdomain and M1 and M3 are the respective N-terminal and C-terminal portions of the CIDR1α domain. Vaccination with MC179 protects Aotus monkeys from severe infection upon challenge with the MC strain [34] and prevents severe anemia after infection with a heterologous strain [35].
Three-dimensional structures of DBL, but not CIDR domains, have been available, with sequence comparisons revealing a number of conserved hydrophobic residues and cysteines, yet with low overall sequence identities among PfEMP1 domains [7]. We present the structure and analysis of MC179 in the context of the broad sequence diversity of the CIDR domains of the P. falciparum genomes and indicate the insights into overall DBL and PfEMP1 structure that MC179 provides.
MC179 is composed of a bundle of three helices connected by loops and three additional helices (Figure 1). The first helix (H1) of the bundle is joined to the second helix (H2) through a disordered fourteen-residue loop. At the end of the H2 helix there are seventy residues in three smaller helices, a, b, and c, that form the connection to the third helix (H3). The “V-shape” of MC179 is made up of the three-helix bundle and the three helices between the H2 and H3 helices, in the order H1-H2-a-b-c-H3. The H1 and H2 helices make many contacts with each other and are linked by a conserved disulfide bond, Cys 32-Cys 45, that was not visible in the electron density, but inferred to be present. The H2 and H3 helices also make many contacts with each other and are linked by a conserved disulfide bond, Cys 49-Cys 159, that is observed in the electron density. The H1 and H3 helices, however, make only a few contacts within 4 Å distance and are not linked by a disulfide. The cysteine-rich area from which the domain gets its name is located near the end of the bundle where three disulfide bonds and one free cysteine are located (Figure 1D).
Using mammalian cells transfected to express human CD36, we confirmed by flow cytometry that the MC179 portion of the CIDR1α domain that had been crystallized was capable of binding CD36. Adding MC179 to cells expressing CD36 revealed robust binding of MC179 (Figure 2A and 2B, blue hatched peaks). Preincubation of MC179 with soluble human CD36 produced in insect cells diminished the binding of MC179 to the CD36-expressing cells (Figure 2A, orange peak). This indicated that CD36 in solution could compete with cell surface CD36 for the binding of MC179. Preincubation of the CD36-expressing cells with anti-CD36 antibodies that are known to block the cytoadherence of parasitized RBC [36] prevented the binding of MC179 to CD36 (Figure 2B, green, red, violet peaks). This result signified that the binding of cytoadherence-blocking antibodies to CD36 could interfere with the binding of MC179 to CD36.
The cysteine-richness for which the domain is named stems from the conserved cysteines located on the H1-H2 loop, at the N-terminus of H2, and at the C-terminus of H3 (Figure 1D). We directly observed four of the seven cysteines and have evidence that six of the seven cysteines in MC179 are in disulfide bonds. The disulfide bond Cys 49-Cys 159 between the H2 and H3 helices was observed in electron density. Two other cysteines (Cys 45 and Cys 51) at the N-terminal end of the H2 helix were also observed in electron density. Difference electron density based on the sulfur anomalous signal in the native X-ray dataset confirmed the observation of each of the four cysteines. The sulfur atoms of cysteines 32 and 41 on the loop between the H1 and H2 helices and of cysteine 168 at the C-terminus of MC179 were not visible.
During the preparation of MC179 for crystallization, mass spectrometry detected an adduct of the size of a cysteamine molecule on MC179. We reasoned that one cysteine had reacted with cystamine (the oxidized form of cysteamine) during protein refolding, which implied that there was one free cysteine in MC179. From comparisons with DBL domains, we concluded that Cys 51 is the single unpaired cysteine in MC179 (see below). In experiments to determine which cysteine possessed the adduct, we deleted Cys 168 by truncating twelve residues from the MC179 C-terminus, resulting in the shorter MC167 construct. After refolding, mass spectrometry of MC167 showed that two cysteines had reacted with cystamine, indicating that the deletion of Cys 168 had generated a second free cysteine. This implied that Cys 168 in MC179 participates in a disulfide bond. Reports that MC179 with Cys 168 substituted by serine and a MC167-like truncated molecule still bind CD36 [30],[37] encouraged us to produce MC167 crystals and to use them interchangeably with MC179 during the structure determination. As residues 168–179 and the loop between H1 and H2 were not visible in the MC179 structure (see Methods), the observed X-ray structures of MC167 and MC179 are identical.
How do MC179 and thus the CIDR1α domains of PfEMP1 molecules bind CD36? The effects of mutations chosen based on sequence similarities in several regions of MC179 have been reported and these mutations can now be located and defined structurally. The substitution of Ser for Cys 45 or a double substitution of Ser for both Cys 159 and Cys 168 abolished CD36 binding [30]. Except for those cysteines, point mutations have little effect on CD36 binding, but changing many residues at the same time has a large effect on CD36-binding. CD36-binding was not affected when point mutations within helices H1 and H2 of the three-helix bundle were made or when the residues between Cys 32 and Cys 41 on the loop connecting the H1 and H2 helices were replaced with residues from a non-binding CIDR domain [33]. When several point mutations were placed into the same molecule, CD36 binding diminished from 100% to 63% [21]. CD36-binding was lost when chimeric proteins were made between the MC CIDR1α domain and CIDR domains that do not bind CD36. When the H1-H2, a-b-c, and a-b-c-H3 regions of MC179 were substituted in turn with the corresponding sequences from the non-binding CIDR domains, no binding to CD36 was detected [21],[33]. The MC179 domain appears to function as a whole in binding CD36 and only large changes in the domain have succeeded in affecting binding to CD36.
Three point mutations in the same molecule affected the binding of several CIDR1α domains to CD36. The wild type CIDR1α domain from the ItG2-CS2 parasite strain does not bind to CD36, but spontaneous mutations that arose during in vitro culture changed each of the residues in the sequence Gly-His-Arg, resulting in CD36-binding [33]. Sequence comparison aligns Gly-His-Arg with the MC179 sequence, Glu 108-Ile 109-Lys 110, which is located at the beginning of the b helix and makes contact with the end of the a helix (Figure 2D). Glu 108 lies just above a hydrophobic patch in the right-hand panel of Figure 2D. Replacing Glu-Ile-Lys with the wild type ItG2-CS2 Gly-His-Arg sequence decreased MC179 binding to CD36 by 50% [33]. This suggests that part of the CD36-binding region of MC179 lies within or near the a and b connecting helices.
Recently, a chimeric protein (“1640-f”) between a non-CD36-binding CIDR1α1 and a CD36-binding CIDR1α molecule exhibited CD36 binding [25]. Based on sequence alignments (Figure S2) and the MC179 structure, the 1640-f chimera contained the H1, H2, a, and b helices from the CIDR1α1 (PFE1640w) domain and the b, c, and H3 helices from the CIDR1α (PF10_0406) domain. It appears that the 1640-f chimera contained two copies of the b helix. Antibodies raised against the CIDR1α portion of the chimera inhibited CD36 binding by erythrocytes infected with the 3D7, HB3, or FCR3 parasite strains, which indicated that there are similar epitopes on CIDR domains that otherwise differ from each other [25]. Where might these cross-reactive epitopes be located? The c helix has few conserved residues and the conserved residues of the b helix point inward toward the a helix, though they still could be contacted by antibodies. We suggest that the H3 helix may be the location of cross-reactive epitopes since the side chains of some conserved residues of the H3 helix, such as Asp 146 and His 151, extend out into solvent.
Sequences and lengths of the a and b helices in CD36-binding CIDR1α domains vary substantially (Figure S1), which is difficult to rationalize with CIDR1α binding to the non-polymorphic CD36 protein. In addition, mutations and chimeric molecules have implicated several areas of the MC179 molecule in binding. An understanding of the mechanism of binding CD36 has not been obtained through mutational analysis of conserved residues. That hundreds of CIDR1α sequences still bind CD36 while containing multiple “natural” mutations implies that CD36-binding may derive from the overall structure of the CIDR1α domain.
We aligned the MC179-like sequences of the 110 CIDR domains of the P. falciparum 3D7 genome (Figure S2). For each position in the MC179 sequence, the percent of aligned sequences having the identical residue at that position was calculated [38]. We plotted the percent sequence identity as a gradient of color (blue) on a surface depiction of MC179 (Figure 3). This revealed a conserved region found in all CIDR domains (Figure 3B). This region is located between the H1 and H2 helices near to the N- and C-termini of MC179 and contains the unpaired cysteine, Cys 51 (Figure 3B). Cys 51 is located on the H2 helix and is conserved in all CIDR domains whether CIDR1 or CIDR2 (Figure S2).
The conserved surface shown in Figure 3B likely becomes part of the interior of the CIDR domain in the presence of the M1 subdomain of CIDR1α that is located N-terminal to the MC179 or M2 subdomain [37]. In other words, the blue surface in Figure 3B is predicted to be the interface between the M1 and M2 subdomains because it is conserved in all CIDR domains, is proximal to the MC179 N-terminus, and contains the unpaired, conserved Cys 51. In the intact CIDR domain, Cys 51 is likely disulfide-bonded to one of the cysteines of M1. The N-terminal part of MC179 was shown to be important in binding CD36 [30] and MC179-analogous protein constructs from some P. falciparum strains bound CD36 only after the polypeptides were lengthened to include M1 [37].
Examination of the MC179 molecule reveals that the surface on the opposite face of MC179 is little conserved (Figure 3A). The surfaces of the a, b, and c helices are not conserved (Figure 3) and the connecting helices in CIDR1α sequences vary in length. Therefore, a large part of the MC179 surface is not conserved, as most of the residues making up the surface of MC179 are not conserved in other CIDR1α domains. We propose that much of MC179 is exposed at the surface of the PfEMP1 molecule. Of course, MC179 must be exposed at the surface to bind CD36. How much of it? Perhaps 1000 Å2 would be a minimum estimate in light of other known protein-protein interactions. MC179 has a total solvent accessible area of 10,000 Å2. Therefore, at least 10% of MC179 is exposed. We detected conserved patches that we attribute to the interface with the M1 subdomain of CIDR1α and to potential CD36 binding (see below). We did not detect a conserved surface that might be an interface with DBL domains, for instance, DBL1α or DBL2, although the M1 portion may provide a conserved contact surface with DBL1α. It seems likely to us that the CIDR1α domain has only a small surface area in contact with the DBL domains and that more than 10% of the area of MC179 is exposed to the bloodstream. MC179 likely extends out from the PfEMP1 molecule, consistent with the proximity of the MC179 N- and C-termini to the conserved patch predicted to interface with the M1 portion. Persons living in malaria endemic regions only slowly acquire effective antibodies against iRBCs [39]. One or more variable MC179-like surfaces on PfEMP1 molecules would provide a high diversity of antibody epitopes, leading to the slow development of immunity and to the parasite's ability to evade the host immune response.
The majority of PfEMP1 molecules from the 3D7 parasite strain contain one CIDR1α domain that binds CD36 and one CIDR2β domain that does not bind CD36 [21]. We mapped the percent identity of the MC179 sequence with the 3D7 CIDR1α domains (Figure S1) and with the 3D7 CIDR2β domains (Figure S3) on separate surface models of MC179 (Figure 4). This revealed a conserved region located between the H1 and H3 helices that is present in both kinds of domains, but is larger in CIDR1α domains. Every CIDR domain has some identity with MC179 here, accounted for by conserved residues that make up the interior of the three-helix bundle, but are partially exposed at the surface of the molecule. Examples of these are three cysteines (45, 49, and 159) and two leucines (19 and 148) (Figure 4).
Contributing to this surface in CIDR1α domains are a conserved serine and a conserved glutamic acid. In MC179, Ser 22 and the charged Glu 152 are both nearly buried and form a hydrogen bond. Except in one PfEMP1 (PFL0020w) where Ser 22 is replaced by a cysteine residue, Ser 22 and Glu 152 are completely conserved in CIDR1α domains, but one or both of the residues are absent in other CIDR domains. Ser 22 becomes a tyrosine in most other CIDR domains and Glu 152 becomes a leucine in all other 3D7 CIDR domains (Figure 4, middle). Although this conserved surface in CIDR1α domains with a nearly buried glutamic acid residue might be a potential interaction surface with CD36, mutations of Ser 22 to threonine [33] and Glu 152 to leucine [21] have not affected CD36 binding. Additionally, Lys 147 is present in most CIDR1α sequences, within the conserved region shown in Figure 4. In almost all CIDR2β domains, the residue at the Lys 147 position is a cysteine (Figure 4).
During the first cloning and sequencing of PfEMP1 genes, conserved hydrophobic residues and cysteines observed in the midst of the enormous sequence diversity of these large proteins were used to define the DBL and CIDR domains and to suggest their relatedness [7]. Recent crystal structures of non-PfEMP1 DBL domains [40],[41] from the P. falciparum erythrocyte-binding antigen (EBA)-175, containing tandem DBL domains F1 and F2, and the P. knowlesi erythrocyte binding protein (Pkα-DBL), containing a single DBL domain, showed that conserved residues are within DBL helices. We superimposed the MC179 structure on the F1 DBL domain from EBA-175 (Figure 5A), the F2 DBL domain from EBA-175, and the DBL domain from Pkα (Figure 6). The three-helix bundle of MC179 structurally matched each DBL C-terminal three-helix bundle, a region that has also been named “subdomain 3” [41]. MC179 and all three DBL structures closely overlaid with root mean squared (r.m.s.) deviations between 1.6–2.0 Å over the 74–77 α-carbon atoms of the three helices. The overlay is best among the H1 and H2 helices of all four proteins and between the regular H3 helices of MC179 and the EBA-175 F1 DBL domain. The H3 helices of EBA-175 F2 and of Pkα-DBL are mixed with irregular secondary structure and take a meandering route along their respective H1 and H2 helices. This structurally conserved three-helix “bundle” is a feature of the DBL and CIDR protein folds and is evidence beyond their sequence similarities for the common ancestry between the two domain families [7].
More than main chain atoms match among the structures, however, as even the conserved side chains in the H1, H2, H3 helices of MC179 and in the corresponding helices of the DBL domains are in identical positions. In Figure 5B, six conserved Trp, Phe, and Tyr side chains in the H1 and H2 helices closely overlay between MC179 and the F1 DBL domain. In addition, the Cys 49-Cys 159 disulfide and Cys 45 side chain of MC179 overlay to within 3.0 Å of the analogous cysteines of the EBA-175 and Pkα-DBL domains. It is clear that the MC179 and thus the CIDR1α disulfide bonding is the same as in the DBL domains (Figure 1D and 5B). In the MC179 numbering used in the structure, the cysteines involved in disulfide bonds are 32–45, 41–168, and 49–159 (Figure 5B). Cys 51 extends into solvent from the H2 helix and is conserved in all CIDR, but not DBL domains. Cysteines 49 and 51 form the Cys-X-Cys sequence motif that is found in most CIDR sequences [18]. Sequence alignments and more recently, detailed modeling studies based on the EBA-175 and Pkα crystal structures, have predicted that the interiors of DBL and CIDR domains would be held together by conserved hydrophobic and cysteine residues and that loops would contain the most polymorphic sequences [42],[43].
The structural overlay of MC179 on the available DBL domains also reveals the differences in the connections between the helices of the bundle (Figure 6). A loop of approximately similar length connects the H1 and H2 helices in both MC179 and in the structurally known DBL domains, but the connection between the H2 and H3 helices differs considerably between the two kinds of domains. In the known DBL domains, the H2-H3 connections are about 30 residues in length consisting of a small loop and helix that vary among the three DBL domains. In MC179, the H2-H3 connection is about 70 residues in length and ranges from 30 to 100 residues in the many known CIDR sequences.
While examining the similarities between MC179 and the known DBL domains, it became evident that DBL domains are composed of two three-helix bundles (Figure 6). As described above, MC179 structurally matches the DBL C-terminal three-helix subdomain, but the DBL subdomain nearer to the N-terminus is also made up of three helices that are connected like the MC179 and C-terminal DBL bundles. The sequence extents of the N-terminal and C-terminal three-helix bundles in the structurally determined DBL domains are: residues 58–159 and 174–282 in EBA-175 F1, residues 365–463 and 481–596 in EBA-175 F2, residues 64–165 and 190–304 in Pkα-DBL (PDB codes 1ZRL and 2C6J). In Figure 6, MC179 and the three DBL domains are colored to highlight bundle helices and connections between helices. MC179 structurally overlays on the DBL C-terminal three-helix bundle, but does not overlay on the N-terminal bundle. However, the helices and the connections between the helices of the N-terminal bundle (h1, h2, h3) are similar to those of the C-terminal bundle: the h1 helix (dark blue), a short connecting loop (gray), the h2 helix (light blue), a longer connecting helix (yellow), and the h3 helix (red) (Figure 6). Like the C-terminal bundle, the N-terminal bundle contains a disulfide that connects the h2 helix at its N-terminus with the h3 helix at its C-terminus. The cysteines that compose this disulfide in each DBL domain are: Cys 88-Cys 164 in EBA-175 F1; Cys 396-Cys 471 in EBA-175 F2; Cys 99-Cys 176 in Pkα-DBL. These disulfides pin the h2 and h3 helices to each other and correspond to the Cys 49-Cys 159 disulfide of MC179 and the analogous disulfides of the C-terminal three-helix bundles of DBL domains.
Based on structural similarities and conserved residues among the CIDR and DBL domains, we developed sequence expressions to detect the H1 and H2 helices of the three-helix bundles that can be used to locate CIDR and DBL domains in PfEMP1 sequences. These sequence patterns describe the conserved residues that are unique to the H1 and H2 helices. The H1 helix sequence of CIDR and DBL domains satisfies the expression: (Leu, Phe, Trp, or Tyr)-X3-Trp-X17, 18, or 19-Cys, where “Xsubscript” indicates the number of residues of unspecified type occurring between conserved residues. Several possible hydrophobic residues are indicated in parentheses at some positions. As examples, the H1 helix of MC179 is located with the pattern Phe-X3-Trp-X17-Cys and the predicted H1 helix of the Malayan Camp DBL1α is found with Trp-X3-Trp-X18-Cys (Figure 7).
Different patterns of conserved residues distinguish the CIDR H2 helices and the DBL H2 helices. The H2 helix is found in CIDR domains by the expression: Cys-X3-Trp-X7 or 8-Trp-X6-(Phe or Tyr). This motif is preceeded by an additional Cys-X1- in many CIDR domains. In DBL domains, H2 helices are predicted with this algorithm: Cys-X3-Cys-X2-(Phe or Tyr)-X2-(Leu, Lys, Trp, Tyr)-X7-(Phe, Trp, or Tyr)-X6-(Phe or Tyr).
We used the CIDR algorithms for H1 and H2 to detect a CIDR-like domain in the var2csa protein, the product of a relatively conserved gene found in all P. falciparum strains and expressed by most parasites that bind CSA [44],[45]. For example, the var2csa family member in the 3D7 genome, PFL0030c, contains a CIDR-like domain between the DBL2x and DBL3x domains, two of the six DBL domains of PFL0030c. The H1 pattern of this CIDR-like domain is Leu-X3-Trp-X18-Cys and the H2 pattern is Cys-X3-Trp-X7-Trp-X6-Tyr. We predict a three-helix bundle in this CIDR-like domain that extends from residues 1025 to 1212 of the PFL0030c protein. This CIDR-like domain, termed the ID (interdomain)-2 domain, has recently also been predicted through sequence analysis and modeling [43].
In the MC179 crystal, pairs of molecules interact and bury 4300 Å2 of solvent accessible surface area between them (Figure 8). A crystallographic two-fold symmetry axis relates the two MC179 molecules. Much of the contact between the molecules is made by the helices that connect the H2 and H3 helices including the hydrophobic patch seen in Figure 2D. Helix a residues (Leu 94, Leu 95, Leu 96, Ile 98, Ile 99) and helix b residues (Ile 112, Leu 115, Leu 116) constitute this hydrophobic area on one molecule, interacting with H1 helix residues (Trp 12, Val 15, Leu 19, Ile 20, Ile 23) and H3 helix residues (Thr 144, Lys 147, His 151) of the symmetry-related MC179 molecule. Hydrophobic areas in the third connecting helix, helix c, also contribute to this interaction between molecules in the crystal. Opposite to the hydrophobic patch on the a, b, and c helices is a region made up of negative charges from helix a residues (Glu 97, Asp 101), helix b residues (Glu 113, Glu 117), and a helix c residue (Asp 125) (Figure 2E, left side in red). These charges are on the surface of the crystallographic dimer. The molecular surface of the crystalline dimer shows the intertwining of the two molecules in a “handshake”-like interaction (Figure 8B). The surface complementarity (SC) index of the monomer-monomer interface is high at 0.70 [46] and 55 residues (220 atoms) from each molecule take part in the interaction. Residues that when mutated have affected CD36 binding (His 69, Glu 108, Ile 109, Lys 110) contribute to the surface of the MC179 dimer and, except for His 69, are located near, but not in, the dimer interface (Figure 8B). His 69 is distant from the dimer interface.
The interaction between MC179 molecules in the crystal resembles a biological interaction with its large amount of buried surface area and high surface complementarity. In solution, when refolded MC179 was prepared by size exclusion chromatography, we separated molecular species with the mobility of dimers from the desired species with the mobility of monomers and used the monomers for crystallization. In addition, chemical cross-linking experiments indicated the presence of MC179 dimers in solution (data not shown). We conclude that MC179 has the tendency to dimerize in solution. How these dimers of a recombinant fragment of PfEMP1, if biologically relevant, might function in a PfEMP1 molecule is unclear. Based on the enormous contact area observed in the MC179 “dimer” in the crystal, our hypothesis is that the connecting helices of the MC179 portion of CIDR1α domains bind another molecule, either a PfEMP1 domain or a ligand. This hypothesis that the CIDR connecting helices are involved in binding another molecule extends to our prediction of binding by the proposed connecting helices of the DBL1α domain as described below.
Using BLAST with the MC179 amino acid sequence as probe, we aligned the MC179 portions of the approximately 300 CIDR1α sequences from the 3D7 strain, the Ghanaian clinical isolate, the IT strain, and P. reichenowi genomes from the Sanger sequencing center (http://www.sanger.ac.uk/Projects/Protozoa/). In the alignment, semi-conserved hydrophobic residues implied the presence of the MC179 connecting helices a, b, and c. By their sequences, all CIDR1α domains appear to retain approximately the first two connecting helices, the a and b helices, as predicted by alignment with the MC179 sequence, but the third connecting helix c varies greatly in length (see 3D7 CIDR1α alignment in Figure S1). These differing CIDR1α sequences and lengths must result in changes in the size of the V-like opening of the CIDR1α domain as seen in the MC179 structure. The sequence composition and length of the three connecting helices (a, b, and c) vary even among CIDR1α molecules that bind CD36 (Figure S1). In some CIDR1α sequences there are two additional cysteine residues, one located about ten residues after the end of the H2 helix and the other located about ten residues before the start of the H3 helix. In the MC179 structure, these two Cα atom positions are separated by about 8 Å, which could be spanned by a disulfide bond. For example, PF10_0001 (Figure 7), PFD0630c, PFD0005w from the 3D7 strain have such cysteines. We predict that these cysteines form a disulfide bond that could act to stabilize the MC179 structure, by linking the a and b helices. Such a disulfide would prevent movement of helices a and b relative to each other, which would need to be considered in models of how CIDR1α binds CD36, as PF10_0001, PFD0630c, and PFD0005w are CD36 binders [21].
Comparisons of the MC179 and DBL X-ray structures with DBL1α domain sequences indicate that DBL1α domains have a longer H1-H2 connecting loop than do the MC179 and DBL domains. Figure 7 presents the DBL1α domain of the MC PfEMP1 as an example. DBL1α domains also have an additional cysteine in the H1 helix and two additional cysteines within the connecting loop to H2 (Figure 7). The presence of extra cysteines in PfEMP1 DBL domains has complicated the identification of conserved cysteines. The MC179 structure will aid in the comparison of PfEMP1 DBL, non-PfEMP1 DBL, and CIDR domains.
From the current knowledge of the MC179 and DBL X-ray structures, it is clear that the C-terminal boundary of DBL1α domains has sometimes been predicted to be at the end of the H2 helix of the three-helix bundle, as determined from conserved residues and the number of residues between them (Figure 7). This is understandable since after the end of the H2 helix there is little sequence similarity among DBL domains to guide prediction. The MC179 and DBL structures show that a third helix, degenerate helix, or, conceivably, an irregular H3 structure must be present in all DBL and CIDR domains to complete the “bundle”. The structures also predict that there are at least two cysteines at the C-terminus of the third helix that make conserved bonds with Cys 41 and Cys 49 (MC179) or their homologs in the DBL domains (Figures 5B and 7A).
This need for two cysteines at the end of the DBL domain can be used to predict the C-terminus of the DBL1α domain and leads to the prediction of MC179-like connecting helices in DBL1α. For example, in the Malayan Camp sequence shown in Figure 7A (line 8, DBL1α MC), if one proceeds along the DBL1α MC sequence from the end of the H2 helix at conserved residue Tyr 394 (Figure 7A, red dot), a single cysteine is encountered at residue 445, but a cluster of three cysteines is found at residue 475 (Figure 7A, DBL1α MC). Since two cysteines are required to form disulfide bonds with two conserved cysteines near the start of the H2 helix, the third helix H3 of the DBL1α must include the multiple cysteines around residue 475 in the sequence -Cys-Glu-Ala-Cys-Pro-Trp-Cys- (-CEACPWC-). If the DBL1α domain has a three-helix bundle that ends around position 475, then the length of the three-helix region is about 70–80 residues longer than in the EBA-175 and Pkα-DBL domains. This leads to the prediction that the DBL1α domain likely has connecting helices between H2 and H3 of about the same length (70–80 residues) as does MC179 (Figure 7).
The definition of a longer DBL1α domain affects the predicted start of the following CIDR1α domain. Due to the longer DBL1α domain, the immediately following CIDR1α domain and its M1 region are predicted to be 70–80 residues shorter. This reasoning implies that the DBL1α C-terminal region has been included in some predictions of the CIDR1α domain. Locating the DBL domain C-terminal boundary in this way by finding the cysteines that end the third helix is applicable to the DBL sequence classes, α, β, γ, δ, ε, and to the analysis of the semi-conserved domain pairings of DBL1α-CIDR1α, DBL2β-C2, and DBL2δ-CIDR2β that frequently appear in PfEMP1 sequences [18].
The DBL1α-CIDR1α domain pairing or “conserved head structure” is present at the N-terminal end of a majority of PfEMP1 molecules [7],[8]. As the sequences that connect H2 to H3 in the DBL1α and CIDR1α domains of the MC strain appeared to be similar in length, we examined the lengths of connecting regions in all of the DBL1α and CIDR1α domains that are pairs in the 3D7 genome. We counted the number of residues that connect H2 and H3 for each DBL1α and its associated CIDR1α domain and produced a list of forty-seven pairs of numbers, one pair of numbers from each PfEMP1 with a conserved head structure. Each number is the length of the connecting sequence in a DBL1α or a CIDR1α domain. The lengths in the list are correlated with a 0.58 correlation coefficient (Spearman rank-order) between them (p<0.0001) (Figure S4). This positive length correlation between DBL1α and CIDR1α implies that the connecting sequences of the DBL1α domains are similar in length to the CIDR1α connecting regions. This may simply reflect the evolutionary relatedness of the two domains, but may indicate similar binding or other function.
The H2-H3 connecting helices of DBL1α may extend away from the PfEMP1 molecule, as is the prediction for MC179 based on the present structural analysis. The connecting helices in MC179 extend out from the H1-H3 side of the bundle. In DBL1α, we predict that the connecting helices would also extend out from the H1-H3 side of the bundle. This is distinct from the H1-H2 side that interacts with the N-terminal bundle (“subdomain 2” [41]) (Figures 5A, 6). The overlay of the DBL domain and MC179 in Figure 5A also serves as a model for DBL1α, if one looks at the connecting helices of MC179 as modeling the predicted connecting helices of DBL1α. In addition, it is possible that the DBL1α and CIDR1α domains interact with each other in a manner resembling the MC179 dimer in the crystal.
Two additional points support the DBL1α-CIDR1α domain boundary as described here. First, Plasmodium proteins frequently have regions of low sequence complexity inserted in loops and between domains [47],[48]. In the PFL1960w 3D7 PfEMP1 sequence, there is an insertion of 110 glycine and serine residues after the multiple cysteines that mark the end of the DBL1α H3 helix and before the start of the CIDR1α domain. Binding studies with the CIDR1α domain from PFL1960w showed that it does bind CD36 [21], indicating that the insertion did not interfere with CD36 binding. This is consistent with the locations of the end of the DBL1α and the beginning of the CIDR1α domain as proposed here. Second, although MC179 is the minimal portion of the MC CIDR1α that binds CD36, two other recombinant CIDR1α domains bind CD36 only after the domains are lengthened in the N-terminal direction to be longer than MC179 [37]. The minimal lengthening needed to enable these two other CIDR1α domains to bind CD36 extended to just before the end of the DBL1α domains as predicted in this work. This is consistent with the boundary between the DBL1α and CIDR1α domains that does not include the DBL1α H3 helix in the CD36-binding CIDR1α domain.
In addition to DBL and CIDR domains, a third type of domain, C2, is found in PfEMP1 molecules. There are nineteen C2 domains in the 3D7 genome and each is located C-terminal to a DBLβ domain. The A4var PfEMP1, containing a typical C2 domain, is known to bind ICAM-1, an endothelial protein that is a ligand of PfEMP1 [49]. When each of the A4var DBL and CIDR domains were expressed using domain boundaries based on sequence conservation, none of the expressed domains bound ICAM-1 [26]. The PfEMP1 domain that bound ICAM-1 was identified by combining DBL2β with the following C2 domain [50]. In subsequent studies to identify the minimal DBL2β-C2 domain that allowed ICAM-1 binding, the C-terminus of the C2 domain was progressively truncated [51]. ICAM-1 binding was lost when a truncation removed, as would be predicted from the current work, the third helix of the three-helix bundle at the C-terminus of the DBL2β-C2 domain. Work with a P. falciparum parasite that binds ICAM-1 also demonstrated that both DBL2β and C2 domains are required for ICAM-1 binding, as neither DBL2β or C2 expressed alone was able to bind ICAM-1 [52]. From these results, we suggest that the C2 domain completes the subdomain 3 of the DBLβ domain, by providing the connecting residues and the third helix of the three-helix bundle that has the cysteines that make disulfides to conserved cysteines in the H2 helix. That the C2 domain contains the third helix of the DBLβ subdomain 3 has been predicted recently from homology modeling [53].
Sequences of PfEMP1 proteins vary enormously to evade the host immune response, while maintaining their binding to endothelial receptors. Conservation of binding to CD36 in the midst of tremendous variation in CIDR1α sequences implies an essential role for binding in parasite survival and in disease outcome for the host. Based on the MC179 structure and the DBL structures known at present, the highly conserved helices and disulfide bonds of the three-helix motif comprise the individual domains of the molecules in the PfEMP1 and EBA families and serve as the scaffolding on which sequence variation takes place. The two helical bundles in DBL domains and the single bundle in CIDR domains imply that a PfEMP1 molecule is not only modular in domain organization, but is a polymer of helical bundles, elaborated by connecting loops and helices. These loops and helices exhibit extraordinary sequence polymorphism and may be regions exposed to the bloodstream and its antibodies. Focusing on the binding of host ligands by PfEMP1 molecules will give insights into vaccine or drug strategies that will affect cytoadherence, help elucidate the ways that the parasite adjusts the functioning of the immune system, and should provide tools to lessen the impact of malaria in endemic areas of the world.
MC179 (residues 576–754 of GenBank U27338) and MC167 (residues 576–742) expression plasmids were transformed into BL21 DE3-(RIL) cells, grown to an OD600 of ∼1, and induced with IPTG to produce insoluble protein (inclusion bodies). To initiate refolding of either MC167 or MC179, guanidine-solubilized inclusion bodies (30 mg) were then diluted to 45 ml with 3 M guanidine-HCl and pumped (0.05 ml/min) into 1 L of refolding buffer containing 400 mM arginine-HCl, 100 mM Tris-Cl, 5 mM cystamine-HCl, 2 mM DTT, and 2 mM NaEDTA. After 24 h, cystamine-HCl (50 mM) was added and allowed to react for 1 h to block free cysteines. The solution was then dialyzed against 10 L of water for 24 h and then against 10 L of 10 mM Tris-HCl pH 8.0 for 24 h. Protein was concentrated on CM52 cation exchange resin using 10 mM MES-NaOH at pH 6.0 and purified on a Superdex 75 size exclusion column. Protein fractions were concentrated to 6 mg/ml with CentriPrep concentrators. Using the hanging drop method, bipyramidal hexagonal MC167 or MC179 crystals grew in 27% polyethylene glycol (PEG) 400, 100 mM NaCl, 50 mM sodium citrate, pH 4.2, and diffracted to 2.7 Å using X-rays from a rotating anode generator. PEG400 (30% v/v) cryoprotected the frozen crystals. Seleno-methionine-containing inclusion bodies were produced in minimal medium (Athena Enzyme Systems, FL) using the BL21-(DE3)-X strain that is auxotrophic for methionine. Selenomet-MC167 or MC179 was refolded and purified similarly to native and the presence of selenium was verified by mass spectrometry. Selenomethionine crystals grew in 15% PEG400, 100 mM NaCl, and 50 mM sodium citrate, pH 4.2.
Native, selenomethionine, and heavy atom derivative X-ray data were integrated and scaled with the XDS package [54] (Table 1). Seleno-methionine positions were found with SHELXD [55]. The native, selenium, ytterbium, and osmium datasets were used in SHARP [56] to find and refine additional heavy atom sites and to estimate protein phases for electron density maps. After density modification with RESOLVE [57], clear helical segments were observed in the electron density. Using the selenium positions as landmarks, the model was built, and then refined in the CNS package [58] to an Rwork value of 0.25 and an Rfree value of 0.29. No dependable model could be built for one N-terminal residue, loop residues 30–43, and C-terminal residues 168–179. Using model phases and highly redundant data collected with 1.54 Å X-rays, anomalous difference electron density confirmed the positions of each sulfur atom in the model. Models were superimposed with LSQMAN [59] and the SSM server [60]. Figures were produced with PyMOL software [61] and Jalview [62]. Surface area calculations were performed with the PISA server [63]. Coordinates and structure factors have been deposited with accession code 3C64 in the RCSB Protein Data Bank.
The binding of recombinant MC179 with a hexa-His tag at its C-terminus to human CD36, expressed on stably transfected Chinese hamster ovary cells (CHO-CD36), was assayed with a flow cytometry protocol similar to one described [64]. Briefly, cells were suspended at 106 cells/ml in phosphate-buffered saline (PBS) containing 0.5% bovine serum albumin and 0.1% (w/v) sodium azide (FCA buffer). Cells (100 µl) in V-bottomed, 96-well plates were pelleted, then resuspended in 100 µl containing 100 ng/ml MC179 and incubated for 1 h. Following washing with PBS, an Alexa 488-labeled anti-pentahis monoclonal antibody (100 µl) (Novagen, San Diego, CA) or an anti-MC179 serum from an immunized Aotus monkey were added at a final concentration of 0.8 µg/ml and 1∶100, respectively. IgG1 isotype mAb (Becton Dickinson, San Jose, CA) and Aotus pre-immune serum were included as controls. After incubation for 30 min, cells were washed twice in PBS and Aotus antibodies were detected with Alexa 488-labeled goat anti-human IgG (Invitrogen, Carlsbad, CA) (100 µl) diluted 1∶250 in FCA buffer. The cells were then incubated for 30 min, washed, resuspended in 150 µl and analyzed by FACSort (Becton Dickinson). After staining with 0.5 µg/ml propidium iodide (PI), PI+ cells were excluded from the analysis. Anti-CD36 antibody clones used were 185-1G2 (Lab Vision, Fremont, CA), CLB-IVC7 (Sanquin, Amsterdam, Netherlands), and 8A6 [36] (gift from Dr. J. W. Barnwell, Center for Disease Control, Atlanta, GA). All the washing steps and incubations were performed at 4°C. Events (10,000) were acquired using CELLQuest software (version 3.3; Becton Dickinson) and the data from the live cell gate were analyzed by FlowJo software (version 6.4.1; Tree Star, San Carlos, CA).
|
10.1371/journal.ppat.0030029 | TBK1 Protects Vacuolar Integrity during Intracellular Bacterial Infection | TANK-binding kinase-1 (TBK1) is an integral component of Type I interferon induction by microbial infection. The importance of TBK1 and Type I interferon in antiviral immunity is well established, but the function of TBK1 in bacterial infection is unclear. Upon infection of murine embryonic fibroblasts with Salmonella enterica serovar Typhimurium (Salmonella), more extensive bacterial proliferation was observed in tbk1−/− than tbk1+/+ cells. TBK1 kinase activity was required for restriction of bacterial infection, but interferon regulatory factor-3 or Type I interferon did not contribute to this TBK1-dependent function. In tbk1−/−cells, Salmonella, enteropathogenic Escherichia coli, and Streptococcus pyogenes escaped from vacuoles into the cytosol where increased replication occurred, which suggests that TBK1 regulates the integrity of pathogen-containing vacuoles. Knockdown of tbk1 in macrophages and epithelial cells also resulted in increased bacterial localization in the cytosol, indicating that the role of TBK1 in maintaining vacuolar integrity is relevant in different cell types. Taken together, these data demonstrate a requirement for TBK1 in control of bacterial infection distinct from its established role in antiviral immunity.
| Early control of invading microbial pathogens is an essential function of the host response to infection. Previous studies have shown that upon viral infection, a protein called TANK-binding kinase-1(TBK1) signals the induction of a program of protection that results in inhibition of viral replication. During infection of mammalian cells by bacteria, a different type of microbe than a virus, TBK1 also sends signals, but the functional contribution of TBK1 to controlling bacterial infection was unknown. Here, we show that TBK1 does protect host cells from bacterial infection; however, the TBK1-dependent mechanisms that inhibit viral infection were not effective against bacterial growth. Instead, TBK1 maintained the integrity of the vacuolar compartment, consisting of small membrane-bound vesicles, where the invading bacteria were trapped. In the absence of TBK1, pathogens such as Salmonella, enteropathogenic Escherichia coli, and Group A Streptococcus were able to escape from the confining host vacuoles and grow to high levels within the host cytosol. Thus, TBK1 plays an important role in the cellular response to bacterial infection, distinct from its function in antiviral immunity.
| Host organisms employ a multitude of innate defense mechanisms against invading microbial pathogens. Functions of the innate immune system include control and destruction of pathogens and instruction of the developing adaptive immune response through expression of cytokines, chemokines, and other proinflammatory molecules [1,2]. Innate recognition of invading pathogens can occur through pattern recognition receptors, such as Toll-like receptors (TLRs), which bind to molecules characteristic of microbial organisms like lipopolysaccharide (LPS) [3]. Two well-characterized signaling pathways associated with TLR stimulation are the MyD88-dependent pathway, which primarily results in NFκB activation, and the TANK-binding-kinase-1 (TBK1)-dependent pathway that induces transcription of Type I interferon genes [4]. The importance of the MyD88-dependent pathway in antibacterial immunity is well established, but the functional contribution of TBK1-dependent signaling in protecting against bacterial infection is unknown [3].
TBK1, also termed NFκB-activating kinase or TRAF2-associated kinase, is a ubiquitous member of the IκB kinase (IKK) family that is required for embryonic development [5–7]. Upon microbial infection or LPS treatment, TBK1 phosphorylates the transcription factor interferon regulatory factor-3 (IRF3), resulting in IRF3 translocation into the nucleus and transcription of target genes, such as ip10 (IFN-gamma inducible protein 10), ifnb (interferon beta; a Type I interferon), and subsequent interferon response genes such as mx1 [8–10]. TBK1 is required for optimal induction of ifnb and Type I interferon-dependent antimicrobial effector mechanisms during viral infection [11]. Induction of TBK1-dependent signaling by viruses has been extensively studied and occurs upon viral recognition by TLRs or upon cytosolic binding of viral double-stranded RNA by the DEAD-box helicases, RIG-I and MDA5 [12–16]. Gram-negative bacterial infections can also trigger TBK1-dependent interferon induction by association of LPS with the TLR4 signaling complex [10,17]. In contrast, extracellular Gram-positive bacteria do not activate expression of interferon genes, although cytosolic Gram-positive bacteria can upregulate ifnb transcription in an IRF3-dependent manner [18,19].
To investigate the requirement for TBK1 in response to bacterial infection, we used Salmonella enterica serovar Typhimurium as a model Gram-negative bacterium. Salmonella is a facultative intracellular pathogen that replicates within macrophages and non-phagocytic cells [20,21]. Salmonella invades non-phagocytic cells by secreting effector proteins through a syringe-like Type III secretion system (T3SS) encoded on Salmonella pathogenicity island-1 (SPI-1) to induce membrane ruffling and bacterial uptake [22]. After entry into the host cell, Salmonella resides in a membrane-bound compartment termed the Salmonella-containing vacuole (SCV). Subsequent to invasion, the SCV progresses through early stages of endocytic maturation and acquires the late endosomal marker, lysosomal-associated membrane protein 1 (LAMP-1) [23]. At this stage, the SCV can diverge from the endocytic pathway by avoiding lysosome fusion and can localize to the perinuclear region [24]. Many Salmonella remain in the SCV and replicate; however, a small percentage of bacteria escape into the host cytosol where they acquire host-derived ubiquitin [25].
Like viruses, intracellular bacterial pathogens exploit the host cell as a replicative niche. By analogy to known antiviral immune mechanisms, we hypothesized that TBK1 would also protect host cells from infection by intracellular bacterial pathogens. However, specific effector mechanisms of antiviral innate immunity, such as Type I interferon-dependent induction of 2–5′ oligoadenylate synthase, which facilitates viral RNA degradation, would not necessarily be effective against intracellular bacteria [26]. Since loss of TBK1 results in embryonic lethality, we used tbk1+/+ and tbk1−/− mouse embryonic fibroblasts (MEFs) or RNAi to test the requirement for TBK1 during bacterial infection. Here, we show that TBK1 mediates an early cellular response to infection by Salmonella and other bacteria by maintaining these pathogens in a restrictive vacuolar compartment.
Strains used in this study are described in Table S1. Salmonella strains expressing green fluorescent protein (GFP) were constructed by electroporating the bacteria with pFPV25.1 obtained from C. Detweiler [27]. The Salmonella strain expressing red fluorescent protein (RFP) was constructed by amplifying the RFP gene from pGEM:RFP (J. A. Bauer, University of Michigan Medical School), ligating the sequence into PCR-II-Topo (Invitrogen, http://www.invitrogen.com), and then electroporation into Salmonella. MEFs were obtained from W. C. Yeh (tbk1+/+ and tbk1−/−) and K. Mossman and B. Williams (irf3+/+ and irf3−/−). MEFs and RAW264.7 cells (American Type Culture Collection) were grown in RPMI medium supplemented with 10% fetal bovine serum and 1% L-glutamine at 37 °C in 5% CO2. Experiments in tbk1+/+ and tbk1−/− MEFs were reproduced both with primary MEFs in early passage and with MEFs that were immortalized by continuous culture and clonally derived. HeLa cells (ATCC) were grown in MEMα medium supplemented with 10% fetal bovine serum, 1% L-glutamine, 1% nonessential amino acids, and 1% sodium pyruvate at 37 °C in 5% CO2.
All Salmonella strains were grown overnight in LB medium at 37 °C, shaking and back-diluted 1:100. When the bacteria reached exponential phase, they were washed twice with Dulbecco's PBS (D-PBS) and used to infect MEFs at a multiplicity of infection (m.o.i.) of 10 for 1 h, whereupon infected cells were washed three times with D-PBS and incubated in medium containing 100 μg/ml gentamicin for 2 h. The cells were then washed three times with D-PBS and fresh medium containing 5 μg/ml gentamicin was added. For intracellular growth curves, at indicated times post-infection (p.i.), three cover slips containing infected MEFs were removed and individually lysed in 5 ml sterile water; a fraction of the lysate was plated on LB agar to enumerate colony-forming units. St SPI-1− required bystander infection with wild-type Salmonella; coinfections were performed with Salmonella grown and infected under the same conditions as stated for monotypic Salmonella infection in MEFs, except that the St SPI-1− mutant was added at an m.o.i. of 100. Enteropathogenic Escherichia coli (EPEC) was grown statically overnight in LB at 25 °C, then back-diluted 1:100, and incubated shaking at 37 °C in serum-free DMEM for 1 h. EPEC were then used to infect MEF cells at an m.o.i. of 25 for 1 h; washing and gentamicin treatment were carried out as described for Salmonella. Streptococcus pyogenes were statically grown overnight at room temperature in BHI. The stationary phase bacteria were washed two times with PBS, vigorously vortexed, and used to infect MEFs at an m.o.i. of 25. The remainder of the infection was done under the same conditions as described for Salmonella infection in MEFs. Bacterial infections in transfected HeLa cells were performed as described for MEFs, except that Salmonella and S. pyogenes were incubated with the host cells for 30 min at an m.o.i. of 100 and EPEC at an m.o.i. of 50 for 1 h; the HeLa cells were washed three times with D-PBS and incubated with fresh medium for 30 min before gentamicin was added. Infections of RAW264.7 macrophages were performed with Salmonella at an m.o.i. of 10 for 30 min, EPEC at an m.o.i. of 50 for 45 min, and S. pyogenes at an m.o.i. of 25 for 30 min. The macrophages were washed three times with D-PBS, and medium with gentamicin added as described for MEFs. 2-μm beads were incubated with anti-streptavidin antibody in D-PBS + 1% BSA for 30 min, then washed three times in D-PBS + 1% BSA, added to the macrophage culture, and spun down onto the macrophage monolayer. Beads were incubated with the macrophages for 30 min and then the cells were washed three times with D-PBS before adding fresh medium.
HeLa cells or MEFs were re-suspended in the appropriate growth medium as described above and plated in a 6-well plate at a concentration of 5 × 105 cells/plate. Cells were transfected for 24 h with either 4.2 μg of plasmid DNA or 40 μM of the indicated short interfering RNA (siRNA) (Ambion, http://www.ambion.com) complexed with Lipofectamine 2000 (Invitrogen, http://www.invitrogen.com). For siRNA experiments, transfected cells were trypsinized, split 1:2, and subjected to a second round of siRNA transfection as above, until time of harvest or infection with Salmonella. For transfection of RAW264.7 macrophages, cells were re-suspended in growth medium as described above and plated in a 6-well plate at a concentration of 1 × 106cells/plate. Macrophages were transfected as described for HeLa and MEF cells.
Cells were grown in appropriate medium in a 6-well plate at a concentration of 7 × 105 cells/plate and infected as described above. At indicated times p.i., cover slips were removed and fixed in 3.7% paraformaldehyde in D-PBS. Cells were then washed three times with 0.1% Triton X-100 in PBS and blocked for 10 min with TBS-TX (25 mM Tris-HCl [pH 8.0], 150 mM NaCl, 0.1% Triton X100, 1% BSA). TBS-TX with primary antibody was added to the cells for 1 h. Cells were then rinsed three times with TBS-TX and incubated for 30 min with secondary antibody. Cover slips were rinsed with TBS-TX and mounted with Pro-Long Gold Antifade (Invitrogen). Samples were analyzed with the Olympus Fluoview FV-500 (http://www.olympus.com) confocal microscope using a 100× objective, unless otherwise stated, and Fluoview software. Quantitation of bacteria colocalized with LAMP-1 or ubiquitin was performed by scoring 150 randomly chosen bacteria per experiment for the presence of LAMP-1 or ubiquitin; only bacteria completely surrounded by LAMP-1 or ubiquitin as indicated by antibody staining were scored as colocalized. Transmission electron microscopy was performed on a Philips CM-100 transmission electron microscope equipped with automated compustage and Kodak 1.6 Megaplus (http://www.kodak.com) high-resolution digital camera. Samples were prepared as previously described [28].
The LAMP-1 (1D4B) rat monoclonal antibody was obtained from Santa Cruz Biotechnology (http://www.scbt.com), the anti-ubiquitin monoclonal antibody (FK2) from BIOMOL International (http://www.biomol.com), the anti-TBK1 monoclonal antibody from Imgenex (http://www.imgenex.com), and anti-streptavidin antibody from Molecular Probes (http://probes.invitrogen.com). TRITC-phalloidin was purchased from Invitrogen and 4′,6-Diamidino-2 phenylindole dihydrochloride (DAPI) from BioChemika (http://www.sigmaaldrich.com). Recombinant mouse interferon-β (PBL Biomedical Laboratories, http://www.pblbio.com) was used at 100 U/ml to treat cells overnight and throughout the course of infection. Lysotracker Red DND-99 (Invitrogen) was a gift from K. Collins (University of Michigan Medical School). Alpha-amanitin (BioChemika) was used at 50 μg/ml to pre-treat cells for 1 h prior to and during the course of infection. All siRNA reagents were obtained from Ambion. The EGFP-LC3 was obtained from Addgene, Incorporated (plasmid 11546) (http://www.addgene.com) and constructed in the laboratory of K. Kirkegaard. Beads and anti-streptavidin antibody were a gift from J. Swanson (2.01-μm streptavidin silicon oxide microspheres; Corpuscular, Cold Spring, New York, United States).
The tbk1 cDNA amplified from tbk1+/+ MEF RNA was cloned into the expression vector pcDNA3 (Invitrogen). After sequence validation, QuikChange mutagenesis [29] was performed to replace lysine-38 with an alanine, which was validated by sequencing. TBK1:pcDNA3 and TBK1-KD:pcDNA3 both encoded a full-length TBK1 protein as determined by TNT Quick Coupled Transcription/Translation Systems (Promega, http://www.promega.com) and Western blot probed with anti-TBK1 antibody.
Total RNA was isolated from cells using the RNAeasy kit (Qiagen, http://www1.qiagen.com) and cDNA synthesis carried out using 2.5 μg of total RNA (M-MLV Reverse Transcriptase; Invitrogen). Real time RT-PCR analysis was performed with the MX3000p (Stratagene, http://www.stratagene.com) and Brilliant SYBR Green MasterMix (Stratagene). Relative amounts of cDNA were normalized to actin cDNA levels in each sample. The following primers were used for amplification: ip10 (F- 5′ATGAGGGCCATAGG GAAGCTTGAA; R- 5′ACCAAGGGCAATTAGGACTAGCCA), mx1 (F-5′ TTGTCTA CTGCCAGGACCAGGTTT; R-5′ TTTCAGGTGCTGGGTCATCTCAGT), actin (F-5′AGGTGTGATGGTGGGAATGG; R-5′GCCTCGTCACCCACATAGGA).
To investigate the role of TBK1 in the cellular response to bacterial infection, we infected tbk1+/+ and tbk1−/− MEFs with Salmonella expressing GFP. Salmonella invaded the wild-type and mutant MEFs similarly; however, at 8 h p.i., the monolayer of tbk1−/− MEFs contained approximately 10-fold more bacteria than the wild-type MEFs (Figure 1A and 1B). Robust bacterial proliferation was observed in 35%–40% of infected tbk1−/− MEFs, which was consistent with immunofluorescence analysis of individual cells showing a more pronounced phenotype than that observed by measuring net increase in bacterial numbers. By immunofluorescence analysis, the remainder of the tbk1−/− infected MEFs appeared similar to tbk1+/+ infected MEFs (Figure 1A; 40× magnification). Infected tbk1−/− MEFs that did not exhibit greater bacterial proliferation might have undergone an unproductive infection, since at 1 h p.i., 20%–30% of the bacteria in either wild-type or mutant MEFs were found in autophagosomes, and an additional 25%–30% in lysosomes, which are likely nonreplicative compartments (Figure S1A and S1B). The increased bacterial growth observed in tbk1−/− MEFs was not suppressed by addition of exogenous Type 1 interferon, nor was a similar phenotype observed when Salmonella infection was compared in irf3+/+ and irf3−/− MEFs (Figures 1B, 1C, and S2A). Inhibition of de novo transcription and translation also had no effect on the phenotype (Figure S2B and S2C; unpublished data). However, the robust bacterial replication in tbk1−/− MEFs was substantially decreased by transient transfection with a plasmid expressing wild-type TBK1, but not a kinase dead mutant protein (TBK1 KD) [6] (Figure 1D). Together, these data indicate that TBK1 kinase activity limits intracellular infection of Salmonella independently of the IRF3-Type I interferon axis.
Previous studies have shown that shortly after invasion, SCV colocalize with the late endosomal marker, LAMP-1 [24]. To determine the nature of the Salmonella-containing compartment in TBK1-deficient cells, we infected MEFs with Salmonella-GFP and analyzed the samples by confocal immunofluorescence microscopy using an anti-LAMP1 antibody (Figures 2A and S3). SCV in tbk1+/+ cells were colocalized with LAMP-1 throughout the entire course of infection, with 94.5% colocalization at 2 h p.i. In contrast, as early as 90 min p.i. in tbk1−/− cells, individual Salmonella lost association with LAMP-1, and at 2 h p.i., only 63.0% of bacteria exhibited colocalization. Loss of LAMP-1 colocalization by individual bacteria was commonly observed early in infection, suggesting that replication per se was not required for this abnormal phenotype. Thus, in the absence of TBK1, many SCV lose the late endosomal marker, LAMP-1, and deviate from the characterized Salmonella endocytic trafficking pathway.
Because Salmonella in TBK1-deficient cells were found in a LAMP-1 negative compartment dispersed throughout infected cells, we reasoned that the bacteria might be in the host cytosol. It was recently reported that under circumstances when Salmonella was found in the cytosol, the bacteria associated with host ubiquitin [25]. Therefore, we infected MEFs with Salmonella-GFP and analyzed the cells by confocal immunofluorescence microscopy to determine whether bacteria were colocalized with ubiquitin (Figure 2B). In tbk1+/+ cells, 0.7% of Salmonella colocalized with ubiquitin at 4 h p.i.; almost all of the bacteria remained in LAMP1+ SCV during the course of infection. In contrast, substantial numbers of Salmonella in tbk1−/− MEFs associated with ubiquitin (37.1% by 4 h p.i.), suggesting that Salmonella was released from the SCV into the cytosol. However, the possibility remained that the SCV was perforated; allowing access to cytosolic ubiquitin, but the vacuolar membrane remained around each bacterium. We directly visualized Salmonella in infected MEFs by transmission electron microscopy and found that by 1 h p.i. in tbk1−/− cells, 66.7% of bacteria (n = 30) were surrounded by vacuolar space, compared to 87.5% of bacteria (n = 24) in tbk1+/+ cells (Figure 2C). From these data, it was not clear whether TBK1 regulated general integrity of the endocytic pathway, or whether infection specifically triggered a TBK1-dependent process. To test the general function of the endocytic compartment, we measured internalization and degradation of I125-labeled epidermal growth factor (EGF) in the absence of infection (Figure 2D). We would predict (based on earlier studies) that general loss of integrity of the endocytic pathway would affect lumenal pH and therefore the ability to degrade proteins taken up by endocytosis, such as EGF [30]. Endocytic uptake and processing of the radiolabeled EGF appeared similar in both tbk1+/+ and tbk1−/− MEFs, demonstrating that escape of Salmonella into the cytosol of TBK1-deficient cells was not the result of general destabilization of the endocytic compartment. Therefore, TBK1 controls an early response to Salmonella infection that maintains integrity of the pathogen-containing vacuole.
Since known mediators of TBK1-dependent signaling were not required to suppress intracellular bacterial growth, and endocytic function was not generally compromised, we hypothesized that Salmonella might be triggering a cellular process that requires TBK1. Salmonella contains two Type III secretion systems encoded on SPI-1 (termed SPI-1 T3SS) and SPI-2 (SPI-2 T3SS) that enable the bacterium to secrete proteins directly into the host cell cytosol [22,31]. The SPI-1 T3SS is required for entry into non-phagocytic cells and modulation of endosomal trafficking; later in infection, SPI-2 T3SS-dependent effectors act to regulate membrane dynamics. We tested the possibility that Type III secretion might contribute to triggering the phenotype observed in TBK1-deficient MEFs. Salmonella strains deficient in either the SPI-1 (St SPI-1−) or SPI-2 (St SPI-2−) encoded T3SS were assessed for their ability to replicate within MEFs and access the host cytosol (Figure 3A and 3B). The SPI-2 deficient bacteria proliferated similarly to wild-type Salmonella in both tbk1+/+ and tbk1−/− MEFs. We also analyzed a Salmonella mutant lacking the SPI-2-dependent effector SifA, which exhibits defects in SCV integrity [32–34]. If SifA and TBK1 were acting in concert, we would expect StΔsifA to proliferate equally in tbk1+/+ and tbk1−/−cells; however, we still observed increased cytosolic localization and replication by the mutant bacteria in tbk1−/− MEFs (Figure S4A and S4B). In contrast, the SPI-1-deficient bacteria (tetR), induced to enter independent vacuoles through bystander infection with wild-type Salmonella (tetS), as measured by tetR colony-forming units or immunofluorescence, were unable to replicate in host cells of either genotype and were never released in the cytosol (Figures 3A, 3B, and S5) [35]. A double SPI-1− SPI-2− mutant behaved similarly to the SPI-1− single mutant (unpublished data). These data suggest that a function associated with the Salmonella SPI-1 T3SS stimulates TBK1-dependent modulation of the pathogen-containing vacuole.
To determine whether vacuolar escape in tbk1−/−cells required a Salmonella-specific process, we used an invasive strain of EPEC, another Gram-negative bacterium, to infect MEFs and determined subcellular localization by quantitating colocalization with LAMP-1 (Figure 3C) [36]. LAMP-1 colocalization of EPEC was substantially decreased in tbk1−/− cells, similar to the phenotype we previously observed in Salmonella infection, and increased bacterial replication was observed compared to tbk1+/+ MEFs (Figures 3C and S4C). We also investigated subcellular localization of S. pyogenes, an invasive Gram-positive bacterium, and found reproducibly that the streptococci were found less often in a LAMP-1+ compartment in TBK1-deficient cells, although the difference between tbk1−/− and tbk1+/+ MEFs was not as striking as that observed for Gram-negative bacteria (Figure 3C). In addition, transmission electron microscopy revealed that at 1 h p.i., fewer S. pyogenes were contained in vacuoles in tbk1−/− MEFs (50%; n = 30) than in tbk1+/+ MEFs (91.5%; n = 47) (Figure 3D). These results demonstrate that TBK1 is necessary for restricting Gram-negative bacteria to the endocytic compartment during infection and may also play a similar role during cellular invasion by Gram-positive bacteria.
MEFs represent an amenable genetic system with which to test the role of TBK1 during bacterial infection in the absence of a gene-deficient live animal model; however, they are not a cell type that would be present during a physiological infection. To establish that loss of TBK1 caused a specific defect in response to bacterial infection, we used an RNAi approach to knock down TBK1 expression in HeLa cells, an epithelial cell line commonly used to study Salmonella pathogenesis, and used the treated cells to examine bacterial growth and compartmentalization. HeLa cells were transfected with tbk1 siRNA, gapdh siRNA, or a nonspecific siRNA control for 72 h prior to infection; knockdown of TBK1 and GAPDH was confirmed by immunoblot analysis (Figure 4A; unpublished data). HeLa cells treated with tbk1 siRNA supported increased replication of Salmonella compared to control treated cells, as observed in tbk1−/− MEFs (Figure 4A). Consistent with our previous observations in MEFs, 27.8% of bacteria in tbk1 siRNA-treated HeLa cells were associated with ubiquitin compared to 2.2% of bacteria in cells transfected with control siRNA (Figure 4B and 4C). No significant increase in ubiquitin-associated bacteria was observed with knockdown of GAPDH. IRF3 was also knocked down by siRNA in HeLa cells; the validated knockdown had no effect on Salmonella growth or ubiquitin association (unpublished data). At 4 h p.i., HeLa cells transfected with tbk1 siRNA had significantly lower numbers of Salmonella, EPEC, or S. pyogenes associated with LAMP-1 compared to the control transfection (Figure 4D). These data demonstrate that TBK1-dependent maintenance of vacuolar integrity is not a MEF-specific phenomenon, but also protects epithelial cells during bacterial infection.
We additionally sought to determine if there was a requirement for TBK1 in immune effector cells such as macrophages. We used RNAi to knock down TBK1 expression in the RAW264.7 macrophage cell line (Figure 5A). It was previously reported that Salmonella grow poorly in macrophage cytosol, so LAMP-1 colocalization was assessed to reflect escape of bacteria from the SCV [32]. At 4 h p.i., in macrophages treated with tbk1 siRNA, only 62.0% of Salmonella colocalized with LAMP-1 compared to 94.7% colocalization in control siRNA-treated cells (Figure 5B and 5C). Tbk1 siRNA treatment of RAW264.7 cells also resulted in decreased LAMP-1 colocalization with bacteria during infection by EPEC or S. pyogenes. In contrast, 2-μm beads taken up by phagocytosis remained completely associated with LAMP-1 in both tbk1 and control siRNA-treated cells. From these observations, we conclude that TBK1 regulates the integrity of pathogen-containing vacuoles in multiple cell types.
We have shown here that the IKK-like kinase, TBK1, mediates an early cellular response to bacterial infection. In the absence of TBK1, Salmonella replicated rapidly and to high levels in MEFs. The IRF3-Type I inteferon axis which contributes to antiviral immunity was not required for the growth restrictive function of TBK1, nor was de novo transcription or translation. After entry into tbk1−/−cells, Salmonella escaped into the cytosol where proliferation occurred. Loss of vacuolar integrity in TBK1-deficient cells was not specific to Salmonella infection, but occurred during infection by Gram-negative and Gram-positive pathogenic bacteria. Thus, TBK1 protects cells during bacterial infection by confining invading pathogens to a membrane-bound compartment.
The best studied TBK-dependent signaling pathway triggered by bacterial infection is LPS-mediated induction of Type I interferons through TLR4, TBK1, and IRF3 [9,11,37,38]. However, our data showed that TBK1 did not require IRF3 or Type I interferon to exert a protective effect on host cells during bacterial infection. It is yet unclear whether TLR4 activation contributes to TBK1-dependent maintenance of vacuolar integrity. We still observed a protective effect by TBK1 on pathogen-containing vacuoles in HeLa cells, which do not express surface TLR4 due to lack of the accessory protein, MD2 [39]. This observation suggests that TLR4 signaling is not absolutely required for the restrictive function of TBK1 in bacterial infection, but further studies will be necessary to definitively determine the role of TLR4 or other TLRs in modulation of vacuolar integrity by TBK1.
The vacuolar compartment is a restrictive antimicrobial environment because of its ability to decrease pH, produce degradative enzymes, and in some cell types, to generate harmful reactive oxygen species in a confined environment [40–42]. In contrast, there are few cytosolic antimicrobial mechanisms, possibly to minimize damage to cytosolic host machinery. Given the more permissive nature of the host cytosol, it is surprising that relatively few bacterial pathogens exploit this intracellular niche [43]. TBK1-dependent mechanisms may prevent many bacterial pathogens from accessing the cytosol by modulating integrity or function of the endocytic compartment during infection. Our data suggest two nonexclusive models by which TBK1 may contribute to the cellular response to bacterial invasion. First, TBK1 could function at the post-transcriptional level in response to an infection by phosphorylating target proteins. In the case of bacterial infection, there may be TBK1 kinase substrates whose function directly or indirectly modifies the pathogen-containing vacuole. Secondly, TBK1 may act prior to infection to establish a state of immune competence, perhaps by regulating expression of gene products that are important for the immediate response to bacterial invasion. This model requires that TBK1 have some constitutive activity prior to infection. Since there is a known requirement for TBK1 in embryonic development in the absence of infection, it is likely that TBK1 may function in normal adult animals in the absence of infection as well [5–7]. Indeed, two recent studies have identified TBK1 as a regulator of angiogenesis and oncogenesis [44,45]. Furthermore, microarray analysis of uninfected tbk1+/+ and tbk1−/− MEFs identified over 400 genes that were differentially expressed, some of which have known associations with innate immune function (ALR and MXDO, unpublished data).
At the molecular level, there are at least two mechanisms by which TBK1 may ultimately regulate vacuolar integrity. In response to bacterial invasion, TBK1 could control initiation of autophagy, which can capture bacteria in damaged vacuoles [46,47]. This possibility is less likely, as colocalization with LC3, a marker of autophagosomes, was slightly higher in infected tbk−/− cells than wild-type cells. However, it is notable that all of the bacterial species, for which we showed cytosolic localization in tbk1−/− cells, have mechanisms by which vacuolar membranes are damaged during infection, i.e., Salmonella and EPEC contain Type III secretion machinery, and S. pyogenes encodes the pore-forming toxin Streptolysin O [48–50]. Alternatively, a TBK1-dependent target could modulate influx/efflux of ions or water into the pathogen-containing vacuole to maintain its physical continuity or otherwise alter membrane dynamics in response to infection. Previous studies have demonstrated that host cells activate repair mechanisms in response to membrane damage [51,52]. Our findings are consistent with damage to the vacuolar membrane as a possible trigger for TBK1-dependent function during bacterial infection.
The innate immune system is required both for controlling pathogen replication and for communicating with other cell types. In viral infections, TBK1 clearly acts as a regulator of innate immunity by communicating with other cells through Type I interferon, which contributes to control of viral replication. However, we have demonstrated that in bacterial infections, TBK1 plays an important role in limiting pathogen replication by protecting the integrity of the pathogen-containing vacuole, independently of Type I interferon. Our findings do not preclude an additional requirement for TBK1 in antibacterial immunity through stimulation of cytokine or chemokine expression, but suggest an early TBK1-dependent mechanism by which host cells can achieve innate control of bacterial infection.
The DNA sequences used for primer design in this study from the NCBI Entrez Nucleotide sequence database (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=Nucleotide) are actin (NM_007393), ip10 (NC_00071), mx1 (NM_010846), and tbk1 (NC_00076.4). |
10.1371/journal.ppat.1000483 | Connecting Quorum Sensing, c-di-GMP, Pel Polysaccharide, and Biofilm Formation in Pseudomonas aeruginosa through Tyrosine Phosphatase TpbA (PA3885) | With the opportunistic pathogen Pseudomonas aeruginosa, quorum sensing based on homoserine lactones was found to influence biofilm formation. Here we discern a mechanism by which quorum sensing controls biofilm formation by screening 5850 transposon mutants of P. aeruginosa PA14 for altered biofilm formation. This screen identified the PA3885 mutant, which had 147-fold more biofilm than the wild-type strain. Loss of PA3885 decreased swimming, abolished swarming, and increased attachment, although this did not affect production of rhamnolipids. The PA3885 mutant also had a wrinkly colony phenotype, formed pronounced pellicles, had substantially more aggregation, and had 28-fold more exopolysaccharide production. Expression of PA3885 in trans reduced biofilm formation and abolished aggregation. Whole transcriptome analysis showed that loss of PA3885 activated expression of the pel locus, an operon that encodes for the synthesis of extracellular matrix polysaccharide. Genetic screening identified that loss of PelABDEG and the PA1120 protein (which contains a GGDEF-motif) suppressed the phenotypes of the PA3885 mutant, suggesting that the function of the PA3885 protein is to regulate 3,5-cyclic diguanylic acid (c-di-GMP) concentrations as a phosphatase since c-di-GMP enhances biofilm formation by activating PelD, and c-di-GMP inhibits swarming. Loss of PA3885 protein increased cellular c-di-GMP concentrations; hence, PA3885 protein is a negative regulator of c-di-GMP production. Purified PA3885 protein has phosphatase activity against phosphotyrosine peptides and is translocated to the periplasm. Las-mediated quorum sensing positively regulates expression of the PA3885 gene. These results show that the PA3885 protein responds to AHL signals and likely dephosphorylates PA1120, which leads to reduced c-di-GMP production. This inhibits matrix exopolysaccharide formation, which leads to reduced biofilm formation; hence, we provide a mechanism for quorum sensing control of biofilm formation through the pel locus and suggest PA3885 should be named TpbA for tyrosine phosphatase related to biofilm formation and PA1120 should be TpbB.
| Most bacteria live in biofilms, which are complex communities of microorganisms attached to a surface via polysaccharides; these biofilms are responsible for most human bacterial diseases. The pathogen Pseudomonas aeruginosa is best-studied for biofilm formation. Currently, it is recognized that cell communication or quorum sensing is important for biofilm formation, but how these external signals are converted into internal signals to regulate the networks of genes that result in biofilm formation is not well understood. Here, by studying 5850 bacterial strains, each of which lacks a single protein, we identify a new enzyme of P. aeruginosa, a tyrosine phosphatase (TpbA), that links extracellular quorum sensing signals to polysaccharide production and biofilm formation. We find that TpbA is subject to control by quorum sensing signals, that it is in the periplasm, and that it controls the level of the intracellular secondary messenger 3,5-cyclic diguanylic acid (c-di-GMP). By controlling c-di-GMP concentrations, TpbA serves to regulate biofilm formation, rapid cell movement on the surface, colony morphology, cell aggregation, and polysaccharide production. The importance of our study is that it shows the secondary messenger c-di-GMP may be regulated by tyrosine phosphorylation; hence, it provides a new target for controlling bacterial social behavior.
| Pseudomonas aeruginosa, an opportunistic pathogen, is often used to elucidate how biofilms form because persistence of this bacterium is linked to its ability to form biofilms [1]. Biofilms are formed by the attachment of bacteria to submerged surfaces in aquatic environments through their production of microbial products including polysaccharides, proteins, and nucleic acids [1]. In P. aeruginosa PA14, the glucose-rich extracellular polysaccharide (EPS) of the biofilm matrix is formed by proteins encoded by the pel operon; note the related strain P. aeruginosa PAO1 has two EPS production loci, pel and psl [2],[3]. Mutations in the pel locus of P. aeruginosa PA14 dramatically decrease biofilm formation as well as pellicle formation; pellicles are formed at the interface between the air and liquid medium [3].
Regulation of Pel polysaccharide involves 3,5-cyclic diguanylic acid (c-di-GMP) which is formed by diguanylate cyclases with GGDEF motifs that synthesize this second messenger; phosphodiesterases with EAL motifs catabolize c-di-GMP. Many proteins with GGDEF motifs enhance biofilm formation [4]; for example, c-di-GMP increases cellulose biosynthesis in Acetobacter xylinus [5], and c-di-GMP enhances EPS production by binding the PelD protein that is a c-di-GMP receptor in P. aeruginosa PA14 [6]. Thus, biofilm formation is controlled by a signal cascade mediated by a complex of c-di-GMP and PelD in P. aeruginosa PA14; however, the upstream portions of this cascade have not been elucidated [7].
Quorum sensing (QS) is bacterial communication using diffusible molecules known as autoinducers to regulate population behavior and is related to both polysaccharide production and biofilm formation. To date, three QS systems have been identified in P. aeruginosa. Las-based QS is regulated by N-(3-oxododecanoyl)-L-homoserine lactone, produced by LasRI [8], and Rhl-based QS is regulated by N-butyryl homoserine lactone, produced by RhlRI [9]. The third QS molecule, 2-heptyl-3-hydroxy-4-quinolone (PQS), was identified as a regulator for both Las- and Rhl-QS [10]. These cell communication signals regulate several phenotypes including virulence and antibiotic resistance [11]. Although the relationship between QS and biofilm formation has not been fully elucidated, some lines of evidence show the importance of QS for biofilm formation. Cells lacking Las-QS in P. aeruginosa form flat biofilms, and this structural abnormality makes bacteria in biofilms more sensitive to antibiotic treatment [12]. Biofilm architecture is regulated by rhamnolipids whose synthesis is controlled by Rhl-QS [13]. Thus, P. aeruginosa QS seems to participate in the development of biofilm architecture rather than initiation of biofilm formation. In addition, LasR- and RhlR-QS have been shown to influence the pel operons indirectly and another transcriptional regulator that controls pel has been predicted [7].
Protein phosphorylation and dephosphorylation are well-conserved posttranslational modifications in both prokaryotes and eukaryotes [14]. Protein kinases and phosphatases modulate cellular activity by adding and removing phosphate groups at Ser, Thr, or Tyr residues. Phosphorylation also occurs at His and Asp residues by histidine kinases and response regulators in two-component regulatory systems. Although the discovery of protein phosphorylation was delayed in prokaryotes compared to eukaryotes [14], many genome sequences predict the existence of phosphorylation/dephosphorylation systems in prokaryotes. The P. aeruginosa genome encodes an extraordinary number of the genes for two-component regulatory systems [15], and diverse cellular functions are regulated by His-Asp phosphorylation including chemotaxis, iron acquisition, alginate production, and virulence factors [16]. In contrast, phosphorylation at Ser, Thr, and Tyr residues has not been studied well in P. aeruginosa; although, Fha1 of the Type VI secretion system is posttranslationally regulated through Thr phosphorylation by the protein kinase PpkA and dephosphorylated by the phosphatase PppA [17]. In Bacillus subtilis, mutations in prkC, a Ser/Thr kinase, and prpC, a phosphatase, decrease sporulation and biofilm formation [18]. Mutations in stk1, a Ser/Thr kinase, and stp1, a phosphatase of Stk1, decrease virulence in Streptococcus agalactiae [19]. These findings show that posttranslational modification via protein phosphorylation at Ser, Thr, and Tyr residues regulates various cellular functions.
In this study, our goal was to explore the complex regulatory cascade that includes detection of QS signals, Pel polysaccharide production, and biofilm formation. By screening 5850 transposon mutants for altered biofilm formation, we identified and characterized a novel protein tyrosine phosphatase, TpbA (tyrosine phosphatase related to biofilm formation), that represses biofilm formation through the pel locus. The tpbA mutant displays pleiotropic phenotypes such as hyperbiofilm formation, enhanced EPS production, altered colony morphology, increased aggregation, elevated c-di-GMP, and abolished swarming. Loss of an uncharacterized GGDEF protein, PA1120 (TpbB), suppressed these phenotypes, indicating that TpbA controls c-di-GMP production through TpbB. Therefore, the mechanism for QS-control of biofilm formation has been extended to include a novel phosphatase (TpbA), a diguanylate cyclase (TpbB), and c-di-GMP; hence, the predicted additional level of control of the pel polysaccharide locus has been identified and involves c-di-GMP as controlled by a tyrosine phosphatase.
Previously, by screening 5850 transposon mutants for altered biofilm formation, we identified 137 transposon mutants of P. aeruginosa PA14 with over 3-fold enhanced biofilm formation [20]. Among these mutants, the tpbA (PA3885) mutant increased biofilm formation by 147-fold after 8 h in LB medium at 37°C (Fig. 1A). This significant increase in biofilm formation upon inactivating tpbA is partially due to enhanced attachment to the polystyrene surface because biofilm formation at the bottom of the plates (solid/liquid interface) increased gradually with the tpbA mutant while PA14 did not form biofilm on the bottom of the plate (Fig. 1B).
Motility often influences biofilm formation in P. aeruginosa; biofilm formation is inversely influenced by swarming motility [21], and swimming motility increases initial attachment to surfaces during biofilm development [22]. To examine the relationship between enhanced biofilm formation and motility in the tpbA mutant, we examined swimming and swarming motility for this mutant; rhlR [23] and flgK [22] mutants were used as negative controls for swarming and swimming motility, respectively. Although PA14 swarmed on the surface of plates at 24 h, the tpbA mutations abolished swarming like the rhlR mutation (Fig. 2A). The tpbA mutation also decreased swimming motility by 40% (Fig. 2B). Swarming is positively influenced by production of the biosurfactant putisolvins in P. putida [24] and rhamnolipids in P. aeruginosa [23]. However, no significant difference was found in the production of rhamnolipids between PA14 and the tpbA mutant (Fig. 2C). This shows the tpbA mutation abolishes swarming in a manner distinct from the production of rhamnolipids.
Congo-red is often used to observe colony morphology because it detects EPS production and this impacts biofilm formation; for example, the wspF mutant shows wrinkly colony morphology on Congo-red plates and has increased biofilm formation [25], while smooth colonies like the pelA mutant [3] form less biofilm. We found that the tpbA mutant formed a red, wrinkly colony when it was grown on Congo-red plates at 37°C, although PA14 and the pelA mutant formed white smooth colonies (Fig. 3A). When the bacteria were grown at 25°C, both PA14 and the tpbA mutant formed red wrinkly colonies, but the pelA mutant still formed a white smooth colony (Fig. 3A). These observations with the pelA mutant and wild-type PA14 are identical to the previous report that expression of the pel genes is induced at room temperature and repressed at 37°C [7]. Therefore, the red wrinkly colony formed by the tpbA mutant at 37°C implies increased production of EPS via Pel.
We also quantified the amount of EPS bound to cells of PA14 and the tpbA mutant at both 37°C and 25°C. As shown in Fig. 3B, the tpbA mutant produced 28-fold more EPS than PA14 at 37°C. The tpbA mutant also produced 4.3-fold more EPS than PA14 at room temperature. The pelA mutant (negative control) did not form EPS at both temperatures tested. We also found that the tpbA mutant formed a pronounced pellicle at 37°C after 1 day, but PA14 and the pelA mutant did not form a pellicle (data not shown). At 25°C, both the tpbA mutant and PA14 formed pellicles after 5 days. Taken together with the EPS production data, TpbA reduces pellicle formation by decreasing Pel activity.
To confirm the impact of the tpbA mutation on pel expression and to investigate its impact on the whole genome, a whole-transcriptome analysis was performed with biofilm cells of the tpbA mutant at 37°C at 7 h; planktonic cells were not assayed since we were primarily interested in how TpbA controls biofilm formation. Inactivation of tpbA altered diverse loci including genes related to EPS production (pelACDF induced approximately 4-fold), transport (PA2204 repressed approximately 5-fold, PA4142–PA4143 induced approximately 3-fold), type IV pili (PA4302 to PA4306 repressed approximately 4-fold), and a putative adhesin and its regulator (PA4624–PA4625 induced approximately 4-fold) (Tables S1 and S2). Expression of tpbA was induced as much as 120-fold in the tpbA mutant, suggesting that TpbA negatively regulates its transcription. The whole-transcriptome experiments were performed twice using independent cultures of PA14 and the tpbA mutant at 7 h, and most of the differentially regulated genes were consistently altered except pel genes which were induced the most in the samples containing an RNase inhibitor. A whole-transcriptome analysis was also conducted using biofilm cells at 4 h since the mode of growth switched from planktonic to biofilm for the tpbA mutant at this time (Fig. 1A). Similar to the 7 h results, several loci were induced including pelAEF (1.5- to 1.7-fold), tpbA (42-fold), PA1168–PA1169 (1.4- to 2.1-fold), PA3886 (3.5-fold), and PA4624–PA4625 (2- to 3.7-fold) (Table S1).
To verify induction of the pel locus, expression of pelA was determined by quantitative real time-PCR (qRT-PCR). Using two independent RNA samples extracted from biofilm cells at 7 h, pelA was induced 112±100-fold in the tpbA mutant vs. PA14. These results showed EPS production is induced significantly in the tpbA mutant due to overexpression of pel genes. qRT-PCR also confirmed induction of PA4625 (7±7-fold) as well as PA4139 (38±30-fold) that encodes a hypothetical protein.
Cell aggregative behavior is also related to biofilm formation so we investigated the role of TpbA on cell aggregation and found the tpbA mutant causes cell aggregation (Fig. 4A). Autoaggregation of the tpbA mutant was also observed in 96-well polystyrene plates during biofilm formation (data not shown). Our whole-transcriptome analysis showed that inactivating tpbA induced both PA4624 (encodes for a putative hemolysin activator) and PA4625 (encodes for an adhesin/hemagglutinin) by 2.1- to 4.9-fold. In E. coli, adhesin regulates cell aggregation as well as attachment [26]. To examine whether PA4624–PA4625 control adhesive activity in P. aeruginosa, we investigated biofilm formation with these mutants. Both mutants showed decreased initial biofilm formation; i.e., initial attachment, to polystyrene plates at 1 h and 2 h (Fig. 4B), and final biofilm formation at 24 h was also decreased for both the PA4624 and PA4625 mutants, which suggests that both gene products control attachment to the surface. Therefore, TpbA decreases cell aggregation probably by repressing the PA4624 and PA4625 genes.
To verify whether the phenotypes observed in the tpbA mutant were caused by loss of function of TpbA, we confirmed transposon insertion in tpbA by PCR at residue 25. Furthermore, biofilm formation for both PA14 and the tpbA mutant were examined with tpbA expressed in trans under the control of an arabinose-inducible promoter. tpbA expression reduced biofilm formation of the tpbA mutant by 33% (Fig. S1A) and abolished biofilm formation on the bottom of the plates (Fig. S1B). Similar results were found upon expressing tpbA in wild-type PA14 (OD540 value was 0.22±0.02 for PA14/pMQ70 and 0.02±0.01 for PA14/pMQ70-tpbA, Fig. S1A). In addition, the aggregative phenotype of the tpbA mutant was also complemented by expression of tpbA in trans (Fig. S1C). Taken together, TpbA functions as a negative regulator of biofilm formation and aggregation in PA14.
To investigate how TpbA regulates biofilm formation, EPS production, wrinkly colony morphology, and cell aggregation, genetic screening was conducted using Tn5-luxAB transposon mutagenesis to find suppressive loci for the phenotypes of the tpbA mutation. The double mutant library (tpbA plus random gene inactivations) was screened first for a reduction in aggregation; this step eliminated most cells with unaltered phenotypes by allowing them to aggregate and precipitate at the bottom of the tube. The cells remaining in the supernatant that failed to aggregate like the tpbA mutant were grown on Congo-red plates, incubated at 37°C for 3–4 days, and colonies displaying a white and smooth shape like the wild-type strain were chosen. After that, a third screen was performed by assaying biofilm formation using 96-well polystyrene plates to identify double mutants that had biofilm formation like the wild-type strain. Twenty-six mutants were identified that showed reduced aggregation, a white smooth colony, and reduced biofilm formation like the wild-type strain, and 19 of these mutations were in the pel locus (Fig. 5A, Table 1). Four of the other mutants have the Tn5-luxAB insertion in the TpbB gene (encodes a GGDEF-motif protein) and in the PA1121 gene (encodes a hypothetical protein). In addition, insertions were found in the PA1678 gene (encodes a putative DNA methylase) and in the promoter of the PA5132 gene (encodes a putative protease) (Table 1). Like the double mutants, all of the single mutants lacking the gene identified by genetic screening were tested for biofilm formation, and all of these mutants formed less biofilm as reported previously (Fig. 5) [3],[4].
Results of genetic screening and the whole-transcriptome analysis implied TpbA regulates c-di-GMP concentrations since loss of one of the GGDEF proteins (TpbB) masked the phenotypes of the tpbA mutant. tpbB encodes a functional GGDEF protein whose activity was confirmed by overexpressing this gene in P. aeruginosa [4]. We also confirmed that expression of tpbB increases cell aggregation and attachment to tubes so the tpbB mutation may be complemented (Fig. S2). In addition, we measured the cellular c-di-GMP concentrations of PA14 and the tpbA mutant using high performance liquid chromatography (HPLC) as reported previously [4]. The peaks corresponding to c-di-GMP were observed with the extracts of the tpbA mutant, but not with those of PA14, and the peak was confirmed by comparing the spectrum to purified c-di-GMP as well as by spiking the samples with purified c-di-GMP (Fig. S3). We estimated the cellular c-di-GMP concentration was 10±2 pmol/mg cells in the tpbA mutant. This is comparable to the c-di-GMP concentration found for a small colony variant that showed aggregation (around 2.0 pmol/mg cells) and a mutant with wrinkly colony morphology [27]. Overexpression of tpbB results in c-di-GMP concentrations of 134 pmol/mg cells in PA14 [4]. Therefore, TpbA reduces c-di-GMP concentrations in the cell and probably does so via TpbB.
tpbA encodes a 218 aa protein that has the conserved domain for a protein tyrosine phosphatase [28],[29] since it has the C(X)5R(S/T) motif beginning at aa 132 (CKHGNNRT). To confirm it is a tyrosine phosphatase, we purified TpbA by adding a polyhistidine tag at either the N-terminus (TpbA-nHis) or the C-terminus (TpbA-cHis) (note only the C-terminus fusion protein was active). Expression of recombinant TpbA was confirmed in E. coli by clear expression of a band at 24 kD (Fig. 6A). The purified TpbA protein had phosphatase activity with p-nitrophenyl phosphate (pNPP) that is often used as a general phosphatase substrate [29] (Fig. 6B). Further proof that TpbA is a tyrosine phosphatase was found using a tyrosine phosphatase specific inhibitor, trisodium orthovanadate [30], that completely inhibited the phosphatase activity of TpbA-cHis (Fig. 6B). The third and fourth lines of evidence that TpbA is a tyrosine phosphatase were found using tyrosine specific substrates; TpbA-cHis dephosphorylated both phosphotyrosine peptides, END(pY)INASL (peptide type I) and DADE(pY)LIPQQG (peptide type II) (Fig. 6C), and this activity was inhibited by trisodium orthovanadate. These results show conclusively that TpbA encodes a tyrosine phosphatase.
To see the effect of tyrosine phosphorylation on biofilm formation, biofilm formation was examined in PA14 with trisodium orthovanadate at 37°C for 4 h which should reduce dephosphorylation by TpbA. Trisodium orthovanadate increased PA14 biofilm formation 3.6-fold (Fig. S4), showing that cellular tyrosine phosphorylation increases biofilm formation.
The N-terminal region of TpbA protein has a putative signal peptide, predicted by pSORT [31], that appears necessary for secretion of this protein (28 aa, MHRSPLAWLRLLLAAVLGAFLLGGPLHA). This implied that processing of N-terminal region of TpbA protein may be essential for full phosphatase activity. To prove that TpbA has an active signal sequence, we expressed TpbA in E. coli and collected the proteins from cytosolic, periplasmic, and membrane fractions. All fractioned proteins were analyzed by SDS-PAGE, and we found that TpbA exclusively localized in the periplasm (data not shown). Hence, TpbA probably dephosphorylates its substrate in the periplasm which explains why phosphatase activity was seen only with the fusion protein with the His tag at the C-terminus.
Since TpbA is a tyrosine phosphatase that is found in the periplasm and since TpbB has three likely periplasmic tyrosines (Y48, Y62, and Y95) [32], we mutated the periplasmic tyrosine residues by converting them to phenylalanine and checked for TpbB activity in the tpbB mutant. Active TpbB, from overexpression of tpbB using the tpbB mutant and pMQ70-tpbB, always leads to aggregation whereas the empty plasmid does not cause aggregation (Fig. S5); hence, if a necessary tyrosine is mutated, there should be a reduction in aggregation. Aggregation was always observed with TpbB-Y95F in nine cultures; hence TpbB-Y95F remains active even though it lacks tyrosine 95 so this tyrosine is not phosphorylated/dephosphorylated. In contrast, the Y48F mutation of TpbB decreased aggregation for 43% of the cultures (20 of 46 cultures did not aggregate), and the Y62F mutation decreased aggregation for 24% of the cultures (9 of 37 cultures did not aggregation). Hence, both Y48 and Y62 are likely targets for tyrosine phosphorylation/dephosphorylation of TpbB with Y48 preferred. We confirmed that these mutations did not affect expression level of TpbB protein (data not shown).
Tyrosine phosphorylation and dephosphorylation have crucial roles in cellular signaling and are well-conserved among many organisms [33]. Some bacterial tyrosine phosphorylations have been identified and these regulatory systems control divergent cellular functions [34]. In order to predict whether TpbA function is conserved among other species, we conducted a BLASTP search and found the TpbA protein is highly conserved among P. aeruginosa (PAO1, PA14, C3719, and PA7 with an E value less than 3e-98) and is well-conserved among P. fluorescens Pf-5, P. fluorescens Pf0-1, P. mendocina, Burkholderia cepacia, Pelobacter carbinolicus, Desulfatibacillum alkenivorans, Bacteroides thetaiotaomicron, B. ovatus, B. caccae, Acinetobacter baumannii, Desulfococcus oleovorans, and Geobacter metallireducens (E values less than 3e-11). All of these conserved tyrosine phosphatases have the C(X)5R(S/T) signature and most are uncharacterized. Even though protein similarity is not very high, some eukaryotes, such as Homo sapiens and Arabidopsis thaliana, have TpbA homologs with a C(X)5R(S/T) signature. Therefore, TpbA and TpbA homologs may share important functions in procaryotes and eucaryotes.
QS regulates many genes in P. aeruginosa via a conserved cis-element in the promoter of each gene. N-(3-oxododecanoyl)-L-homoserine lactone binds to the LasR transcriptional regulator [35], and this complex interacts with the las-box, defined as CT-(N)12-AG sequence [36]. The Las-box is conserved among the promoters of the Las-QS regulated genes including lasB, rhlAB, and rhlI [36]. Another class of transcriptional regulation is governed by the lys-box, that is defined as a palindromic sequence, T-(N)11-A [37], and MvfR is a LysR-type transcription factor that binds to the lys-box [38]. We found that the promoter of tpbA (ptpbA) has a putative las-box 220 bp upstream of the start codon (CTCGCCTCGCTGAAAG) and a putative lys-box 90 bp upstream of the start codon (TGAAGCTGCCTCA). In order to examine if expression of tpbA is regulated by QS, we constructed a ptpbA::lacZ fusion plasmid (pLP-ptpbA) and transformed this into QS-related PA14 mutants (lasI, rhlI, and lasR rhlR). Expression of tpbA gene in biofilm cells was reduced by 42% in the lasI mutant, but not in the rhlI mutant (Fig. 7). Corroborating these results, inactivation of both lasR and rhlR also decreased expression of tpbA gene by 39% (Fig. 7). Similar results were obtained when the activity of ptpbA::lacZ was examined in planktonic cells (50% reduction in transcription for the lasI mutant and 37% reduction for the lasR rhlR mutant). Since loss of QS only affected expression of tpbA by 50%, other factors may also participate in the regulation of tpbA. These results suggest that Las-QS, rather than Rhl-QS, is an activator of tpbA expression with other unknown regulators.
We also investigated whether the tpbA mutation influences the regulation of Las- and Rhl-QS using plasR::lacZ and prhlR::lacZ plasmids. Expression of lasR was slightly increased (1.3-fold) with the tpbA mutation. This indicated that Las-QS has more impact on expression of tpbA than tpbA does on that of Las-QS. In addition, expression of rhlR was decreased by 2-fold in the tpbA mutant. Hence, LasR appears to enhance tpbA transcription and TpbA leads to increased rhl transcription.
In this study, we demonstrate that TpbA is a tyrosine phosphatase that regulates diverse phenotypes in P. aeruginosa including the concentration of cellular c-di-GMP. As a second messenger, c-di-GMP is a positive regulator of biofilm formation [4], EPS production [6], and pellicle formation [4], and a negative regulator of swarming motility [39]. The lines of evidence that show TpbA represses c-di-GMP production in P. aeruginosa that we found are (i) inactivating tpbA increases c-di-GMP (Fig. S3); (ii) inactivating tpbA increases biofilm formation (Fig. 1), EPS production (Fig. 3B), and pellicle formation, and c-di-GMP stimulates biofilm formation [4], EPS production [6], and pellicle formation [4]; (iii) inactivating tpbA increases expression of the pel locus (seen via the whole-transcriptome analysis and RT-PCR), and c-di-GMP activates expression of pelA [6]; (iv) inactivating tpbA increases aggregation (Fig. 4) and expression of adhesins (PA4625), and c-di-GMP stimulates adhesion [40]; (v) inactivating tpbA decreases motility (abolishing swarming and decreasing swimming in the tpbA mutant, Fig. 2AB), and c-di-GMP decreases swarming [40]; (vi) inactivating tpbB (encodes a GGDEF-motif protein that produces c-di-GMP [4]) suppresses the phenotypes observed in the tpbA mutant, and (vii) expression of tpbA and tpbB in trans complements aggregation/biofilm formation and aggregation, respectively. Thus, TpbA represses these phenotypes by decreasing c-di-GMP. A proposed regulatory mechanism for biofilm formation by TpbA is shown in Fig. 8.
EPS production in P. aeruginosa PA14 is regulated by PelA [3]. Transcription of pelA is higher at temperatures lower than 37°C, and PA14 forms more biofilm at lower temperatures [7]. However, the tpbA mutation seems to constitutively enhance pel expression independently from this temperature regulation as seen in the enhanced EPS production at 37°C (Fig. 3) and the whole-transcriptome analysis that was conducted at 37°C (Table S2). In addition to increased expression of pelA, additional activation of Pel proteins might be caused by the increased c-di-GMP concentration by the tpbA mutation since c-di-GMP binds PelD and increases EPS production [6]. In addition to the enhanced EPS production (Fig. 3B) and increased pel expression (Fig. 3A, Table S1, and qRT-PCR) seen in the tpbA mutant, another reason why inactivating tpbA increased biofilm formation is the elevated adhesin activity as seen via enhanced biofilm formation on the bottom of polystyrene plates (Fig. 1B). Cell surface adhesins affect bacterial adhesive activity [41], and we have discovered a novel adhesin (PA4625) that is related to TpbA (Table S1) and to initial biofilm formation (Fig. 4B). Since expression of adhesion factors is also positively regulated by c-di-GMP [40], elevated c-di-GMP level enhances adhesion of the tpbA mutant.
c-di-GMP seems to control the switch of motility-sessility of the tpbA mutant since inactivation of TpbA abolished swarming motility (Fig. 2A) and decreased swimming motility by 40% (Fig. 2B), although regulation of swarming motility is very complex as its activity is controlled by QS, flagellar synthesis, and production of rhamnolipids [42]. In addition, our whole-transcriptome results showed weak repression of some of flagellar biosynthesis genes (flg, fle, and fli loci) due to the elevated c-di-GMP, and activity of FleQ, a transcriptional activator of flagellar biosynthesis, is repressed upon binding c-di-GMP [43]. Hence, the increased c-di-GMP concentrations may repress motility of the tpbA mutant via the FleQ pathway that affects expression of flagellar synthesis genes.
Many genes are expected to be differentially regulated by changing c-di-GMP concentrations since it plays a role as a second messenger in P. aeruginosa. Similar regulation of gene expression was observed between the tpbA mutant and the other strains related to c-di-GMP production. For example, production of PA1107, TpbB, and PA3702 proteins that have a GGDEF-domain leads to activation of pelA expression [6]. Mutation in wspF, encoding a CheB-like methylesterase, increases both biofilm formation and c-di-GMP production [25]. wspF mutation altered expression of genes such as pelABCDEFG, PA4624, PA4625, PA2440, and PA2441 whose expression are induced in the tpbA mutant (Table S2). Common regulation of these genes may be partially controlled by the elevated cellular c-di-GMP concentrations. In contrast, expression of pelA was not induced in the bifA mutant that produces more c-di-GMP [44]. This may be because regulation of c-di-GMP signaling is complex in that the P. aeruginosa genome encodes 17 diguanylate cyclases, 5 phosphodiesterases, and 16 diguanylate cyclase-phosphodiesterase proteins [4].
Relevance of c-di-GMP to regulation of diverse cellular functions is now an emerging topic in bacteriology. This second messenger is an activator of cellulose synthase in Acetobactor xylinum [5] and controls many phenotypes in P. aeruginosa [4]. Several GGDEF proteins for synthesis and EAL proteins for degradation of c-di-GMP have been identified in P. aeruginosa [6],[39],[44], and increased production of c-di-GMP enhances biofilm formation and decreases swarming motility [4],[6],[39],[44]. Similarly, enhanced biofilm formation and/or abolished swarming motility were observed in the tpbA mutant via increased production of cellular c-di-GMP. Since TpbA does not possess GGDEF and EAL domains, this protein indirectly influences cellular c-di-GMP concentrations via its phosphatase activity as shown by activity with both pNPP, a broad substrate for phosphatases, and two phosphotyrosine-specific peptides (Fig. 6). We also found that processing the N-terminal signal sequence may be necessary for TpbA activity in the periplasm. Hence, our results reveal a novel regulatory mechanism for cellular c-di-GMP concentration by tyrosine phosphorylation in the periplasm of P. aeruginosa; control of c-di-GMP by tyrosine phosphorylation has not been shown previously.
There is little known about the regulation of GGDEF and EAL proteins in regard to regulation of c-di-GMP level. A chemosensory system, encoded by wspABCDEFR in PAO1, regulates c-di-GMP production via a His-Asp phosphorylation relay [25]. For the tpbA mutant, tpbB was found to reverse the phenotype of tpbA, suggesting that overproduction of c-di-GMP is clearly related to the phenotypes of the tpbA mutant as overexpression of this gene caused pronounced aggregation (Fig. S2). Probably, TpbB, or another GGDEF protein, might participate in c-di-GMP synthesis in the tpbA mutant. A comprehensive analysis of all of the P. aeruginosa GGDEF proteins has been completed, and those GGDEF proteins that abolished or decreased biofilm formation are PA0169, PA1107, TpbB, PA1181, PA1433, PA1727, PA3702, PA4959, and PA5487 [4]. Among these GGDEF proteins, only PA1107, TpbB, PA3702, and PA5487 increased biofilm formation when their genes were overexpressed [4]. Because TpbA is a periplasmic protein, its target GGDEF protein should have periplasmic regions. By a bioinformatics evaluation, of those four GGDEF proteins that increased biofilm formation, only PA1107 and TpbB have transmembrane regions. Taken together with the results of genetic screening, TpbB is the most likely target protein for TpbA. Also, our results imply the periplasmic Y48 and Y62 residues of TpbB are the likely targets for tyrosine phosphorylation. We are now investigating whether TpbA regulates the activity of GGDEF proteins to control cellular c-di-GMP concentrations in P. aeruginosa.
The relationship between tyrosine phosphorylation and biofilm formation is not well established. We found that trisodium orthovanadate treatment increased biofilm formation of PA14 (Fig. S4), indicating that tyrosine phosphorylation increases biofilm formation in P. aeruginosa. Recently, Ltp1, a low molecular weight tyrosine phosphatase in non-motile, Gram-negative P. gingivalis, was identified as a negative regulator of EPS production and biofilm formation [45]. A sequence similarity search shows TpbA is not a homolog of Ltp1, because TpbA has a signal sequence in its N-terminal region, and TpbA is translocated into the periplasm. Other differences were found in the position of the motif for the tyrosine phosphatase, since TpbA has the motif at the position 132 and Lpt1 has it at position 9. It appears the P. aeruginosa genome encodes another tyrosine phosphatase, annotated as ptpA, that has a tyrosine phosphatase motif at the position 7 and does not have a signal sequence at the N-terminus. The function of PtpA is unknown but it is essential [46].
In contrast to poorly-investigated Tyr phosphorylation, regulation of biofilm formation by phosphorylation has been identified for several systems; for example, for the His kinase/Asp response regulator phosphorylation systems RocS1/RocA1/RocR of P. aeruginosa PAK [47] and the PAO1 PA1611/PA1976/PA2824/RetS/HptB system of P. aeruginosa [48]. In addition, in the B. subtilis PrkC/PrpC system [18], loss of a membrane-anchored Thr kinase and its phosphatase reduces biofilm formation. Our results indicate that TpbA acts as a negative regulator of cellular c-di-GMP formation and loss of TpbA results in increased c-di-GMP concentrations that enhance biofilm formation and inhibit motility. These results show clearly that posttranslational modification through phosphatase activity is related to bacterial biofilm formation as well as to the regulation of the synthesis of cellular second messengers. In addition, by showing tpbA transcription is increased by LasR (Fig. 7) and by finding AHL-binding motifs, we have now linked quorum sensing to c-di-GMP concentrations and biofilm formation in P. aeruginosa. Similarly, Vibrio cholerae QS was found recently to reduce cellular c-di-GMP concentrations via a c-di-GMP-specific phosphodiesterase which leads to lower biofilm formation [49]. A common element in both studies is that QS seems to be a negative regulator of c-di-GMP. The tpbA mutation caused a hyper-aggregative phenotype (Fig. 4), and this would lead to flat biofilms since the wspF mutant, which accumulates increased c-di-GMP, formed flat biofilms [25]. Formation of flat and undifferentiated biofilms is also observed by loss of LasI function [12] that can activate tpbA expression. Hence, TpbA might participate in developing biofilm structure. These results are important in that the regulatory networks that control c-di-GMP concentrations are now linked to the environment and cell populations.
Strains used in this study are listed in Table 2. P. aeruginosa PA14 wild-type and its isogenic mutants were obtained from the Harvard Medical School [46]. Transposon insertion of the tpbA mutant was verified as described previously with a minor modification [50]. Briefly, the tpbA gene was amplified from chromosomal DNA using primers PA14_13660-VF and PA14_13660-VR (Table S3) which did not amplify chromosomal DNA from the tpbA mutant. In addition, the DNA fragment corresponding to the end of the transposon and tpbA gene was amplified with tpbA chromosomal DNA using primers PA14_13660-VF and GB-3a (Table S3) and PA14_13660-VR and R1 (Table S3) but these pairs of primers did not amplify PA14 wild-type chromosomal DNA. P. aeruginosa and E. coli were routinely grown in Luria-Bertani (LB) medium at 37°C unless noted. Gentamicin (15 µg/mL) and tetracycline (75 µg/mL) were used for growth of the P. aeruginosa transposon mutants, carbenicillin (300 µg/mL) was used to maintain P. aeruginosa plasmids, and kanamycin (50 µg/mL) and chloramphenicol (50 µg/mL) were used to maintain E. coli plasmids (Table 2).
For complementation of the tpbA and tpbB mutations, tpbA and tpbB were expressed under the control of the pBAD promoter in pMQ70 [51]. tpbA and tpbB were amplified using a Pfu DNA polymerase with primers PA14_13660-F1-NheI and PA14_13660-R-cHis-HindIII and PA14_49890-F1-NheI and PA14_49890-R-cHis-HindIII, respectively (Table S3). PCR products were cloned into the NheI and HindIII sites of pMQ70. The resulting plasmids, pMQ70-tpbA and pMQ70-tpbB, were transformed into PA14 and the mutants by conjugation. Briefly, 1 mL of overnight culture of the recipient strain (PA14 or the mutant), helper strain (HB101/pRK2013), and donor strain (TG1/pMQ70, TG1/pMQ70-tpbA, or pMQ70-tpbB) was washed with 1 mL of fresh LB medium. The mixture of three strains was incubated on LB plates at 37°C overnight. PA14 strains with pMQ70-based plasmid were selected on LB plates with 100 µg/mL rifampicin (to kill the donor and helper), 300 µg/mL carbenicillin (to kill P. aeruginosa without pMQ70-based plasmids), and 15 µg/mL gentamicin (if a recipient was a PA14 mutant constructed using a transposon insertion with the GmR gene). If indicated, 0.05% arabinose was added to induce gene expression.
Biofilm formation was examined in 96-well polystyrene plates using crystal violet staining [52]. Overnight cultures of P. aeruginosa were diluted to a turbidity of 0.05 at 600 nm with fresh LB medium, and then 150 µL of diluted bacterial culture was incubated in 96-well polystyrene plates for 2, 4, 8, 24, and 50 h. Ten wells were used for each strain and three independent cultures were used for each experiment. Trisodium orthovanadate, a tyrosine phosphatase-specific inhibitor, was added to LB medium at 10 mM.
To observe colony morphology, overnight cultures were diluted to a turbidity of 0.005 at 600 nm with T-broth (10 g/L tryptone), and 2 µL of diluted cultures were spotted on Congo-red plates (10 g/L tryptone, 40 µg/mL Congo-red, and 20 µg/mL Coomassie brilliant blue) [3]. Plates were incubated at 37°C or room temperature for 3 to 7 days.
Swimming motility was examined with cells grown to a turbidity of 1 at 600 nm using 0.3% agar plates with 1% tryptone and 0.25% NaCl [53] and swarming motility was examined with BM-2 plates (62 mM potassium phosphate, 2 mM MgSO4, 10 µM FeSO4, 0.1% casamino acid, 0.4% glucose, and 0.5% Bacto agar) [54]. Motility was measured after 24 h. Five plates were tested for each culture, and two independent cultures were used. The flgK [22] and rhlR [23] mutants were used as negative controls for swimming and swarming, respectively.
Aggregation was examined by diluting overnight cultures with fresh LB medium in 5 mL screw-capped tubes from 0% (no added fresh LB medium) to 100% (pure fresh LB medium). Cells were inverted gently several times and placed at room temperature for 15 min.
Overnight cultures of PA14, the tpbA mutant, and the pelA mutant were diluted to a turbidity of 0.005 at 600 nm in 4 mL T-broth, and the bacterial cultures were placed in a polycarbonate glass tube at 37°C or room temperature [3].
Pel-dependent EPS production was quantified as described previously [6] based on the amount of Congo red that binds to the EPS. Briefly, 1 mL of overnight culture was washed with 1 mL T-broth. Due to aggregative phenotype of the tpbA mutant, cell pellets of the tpbA mutant, wild-type, and pelA mutant (negative control) were sonicated three times at 3W for 10 sec. Bacterial suspensions in T-broth (500 µL) were incubated with 40 µg/mL Congo-red at 37°C or room temperature with vigorous shaking. After 2 h, the absorbance of the supernatants of the each suspension was measured at 490 nm using a spectrophotometer. T-broth with 40 µg/mL Congo-red was used as a blank.
Production of rhamnolipids was determined as described previously [55]. Overnight cultures were diluted to a turbidity of 0.05 at 600 nm in 25 mL LB medium and were re-grown at 250 rpm for 24 h to eliminate the effect of antibiotics. The supernatants of the bacterial cultures were used to determine the relative concentrations of rhamnolipids using orcinol/sulfuric acid. Rhamnose (Fisher Scientific, Pittsburgh, PA) was used as a standard.
The P. aeruginosa genome array (Affymetrix, P/N 510596) was used to investigate differential gene expression in biofilm cells between PA14 and the tpbA mutant. Biofilm cells were harvested from 10 g of glass wool [56] after incubation for 4 h and 7 h in LB with shaking at 250 rpm, and RNA was extracted with a RNeasy Mini Kit (Qiagen) [57]; note the RNase inhibitor RNAlater (Applied Biosystems, Austin, TX) was used for the 4 h and second 7 h set of microarrays. Global scaling was applied so the average signal intensity was 500. The probe array images were inspected for any image artifact. Background values, noise values, and scaling factors of both arrays were examined and were comparable. The intensities of polyadenosine RNA controls were used to monitor the labeling process. If the gene with the larger transcription rate did not have a consistent transcription rate based on the 13 probe pairs (p-value less than 0.05), these genes were discarded. A gene was considered differentially expressed when the p-value for comparing two chips was lower than 0.05 (to assure that the change in gene expression was statistically significant and that false positives arise less than 5%) and when the expression ratio was higher than the standard deviation for the whole microarrays [58], 1.4 for 4 h, 1.7 for the first 7 h replicate, and 2.2 for second 7 h replicate. All three sets of whole-transcriptome data were deposited at the Gene Expression Omnibus (GSE13871).
qRT-PCR was performed using the StepOnePlus™ Real-Time PCR System (Applied Biosystems, Foster City, CA). Expression of pelA, the PA4625 gene, and the PA4139 gene was determined using total RNA isolated from two independent biofilm cultures of PA14 and the tpbA mutant. The biofilm cells were grown and total RNA were isolated in the same manner as described above for the whole-transcriptome analysis. The primers for qRT-PCR are listed in Table S3. The housekeeping gene rplU [44] was used to normalize the gene expression data.
To isolate the suppressive loci for TpbA functions, a double mutant library was generated using the Tn5-luxAB transposon with the background of the tpbA mutation as described previously [59]. Briefly, 1 mL of overnight culture of the P. aeruginosa tpbA mutant and E. coli S17-1 (λpir) with Tn5-luxAB were grown on LB plates together overnight. Cells were harvested from the plate and resuspended in 10 mL of LB medium. Screening of cells with mutations in addition to tpbA was performed in three steps. Suppression of the highly-aggregative phenotype of the tpbA mutant was used first; the cell mixture (P. aeruginosa single and double mutants along with E. coli S17-1) was placed at room temperature for 15 min and the supernatant was used for secondary screening (cells with the tpbA mutant aggregative phenotype were therefore discarded). Supernatant cells were spread on Congo-red plates with 50 µg/mL gentamicin (to kill E. coli), and 75 µg/mL tetracycline (to kill the tpbA single mutant), and incubated for 3–4 days. P. aeruginosa double mutants with smooth surfaces were picked (the tpbA mutant was red and wrinkled). The crystal violet biofilm assay was used for the third screening, and mutants showing decreased biofilm formation in comparison to that of the tpbA mutant were chosen as phenotype reversal mutants. The insertion position of Tn5-luxAB transposon was determined by two-step PCR as described previously [59] with primers LuxAB inside and Arb1 for the first round of PCR and LuxAB outside and Arb2 for the second round of PCR (Table S3). The PCR product was ligated into pGEM-T easy (Promega, Madison, MI) and sequenced using a BigDye Terminator Cycle Sequencing Kit (Applied Biosystems, Foster City, CA).
c-di-GMP was isolated as described previously [60]. P. aeruginosa was grown in 1 L of LB medium for 16 h at 250 rpm, and formaldehyde (final concentration of 0.18%) was added to inactivate degradation of c-di-GMP. Cells were harvested by centrifugation at 8,000 g for 10 min at 4°C. Nucleotide extract was prepared as described previously [60]. Cell pellets were washed with 40 mL of phosphate buffered saline (pH 7) [61] with 0.18% formaldehyde and centrifuged at 8,000 g for 10 min at 4°C. The cell pellets were dissolved in H2O and boiled for 10 min. After cooling the samples on ice for 10 min, nucleotides were extracted in 65% ethanol. Supernatants were transferred, and the extraction was repeated. Pooled supernatants were lyophilized, and pellets were dissolved in 1 mL of 0.15 M triethyl ammonium acetate (TEAA, pH 5.0). The samples were filtered using a PVDF filter (0.22 µm), and 20 µL of each sample was fractionated using HPLC (Waters 515 with photodiode array detector, Milford, MA) with a reverse-phase column (Nova-Pak® C18 column; Waters, 150×3.9 cm, 4 µm). Separations were conducted in 0.15 M TEAA at a 1 mL/min flow rate using gradient elution with acetonitrile (0% to 15% concentration). Synthetic c-di-GMP (BIOLOG Life Science Institute, Bremen, Germany) was used as a standard. The peak corresponding to c-di-GMP from the extract of the tpbA mutant was verified by co-elution with standard c-di-GMP. E. coli AG1/pCA24N-yddV that has an elevated c-di-GMP concentration [62] was also used as a positive control.
To determine if TpbA is a phosphatase, tpbA was amplified with a Pfu DNA polymerase using primers PA14_13660-F-XbaI and PA14_13660-R-XhoI (Table S3). The PCR product was digested with XbaI and XhoI and was ligated in-frame to the polyhistidine tag sequence of the pET28b vector. The resulting plasmid, pET28b-13660c has the tpbA gene fused to a 6× His tag at the C-terminus (TpbA-cHis) and under control of the T7 promoter. The pET28b-13660c plasmid was confirmed by DNA sequencing with the T7 promoter and T7 terminator primers (Table S3). Production of TpbA-cHis was induced in E. coli BL21(DE3) cells with 1 mM IPTG at room temperature overnight. TpbA-cHis was purified using a Ni-NTA resin (Qiagen, Valencia, CA) as described in a manufacturer's protocol. Purified TpbA-cHis was dialyzed against buffer (50 mM Tris-HCl, 100 mM NaCl, 10% glycerol, 0.01% Triton X-100, pH 7.5) at 4°C overnight.
The p-nitrophenyl phosphate assay (pNPP) was used to examine TpbA-cHis phosphatase activity [63]. Purified TpbA-cHis protein was incubated in 100 µL of reaction buffer (50 mM Tris-acetate, 10 mM MgCl2, 10 mM pNPP, 5 mM DTT, pH 5.5) at 37°C for 1 h. The reaction was quenched by adding 900 µL of 1 M NaOH. Trisodium orthovanadate, a specific inhibitor for tyrosine phosphatase [30], was used at 10 mM. p-nitrophenol was measured at an absorbance of 405 nm. An extinction coefficient of 1.78×104 M−1 cm−1 was used to calculate the concentration of p-nitrophenol.
To examine if TpbA is a tyrosine specific phosphatase, a tyrosine phosphatase assay was performed using the Tyrosine Phosphatase Assay System (Qiagen). Eight micrograms of TpbA-cHis were incubated with either 50 µM phosphotyrosine peptide type I (END(pY)INASL) or peptide type II (DADE(pY)LIPQQG) in a reaction buffer (50 mM Tris-Acetate, 10 mM MgCl2, pH 5.5) at 37°C for 3 h. The reaction was quenched using a molybdate dye solution and incubated for 30 min at room temperature. Released phosphate was quantified by measuring the absorbance at 630 nm.
TpbA-cHis protein was expressed in BL21(DE3) cells with 1 mM IPTG for 4 h at 37°C. Periplasmic proteins were purified using a PeriPreps Periplasting Kit (Epicentre Technologies, Madison, WI) as well as cytoplasmic and membrane proteins. Escherichia coli CpdB [64] was used as a control of periplasmic protein and the E. coli OxyR [65] was used for the cytoplasmic control. Fractionated proteins as well as TpbB were analyzed by 12% SDS-PAGE.
Site-directed mutagenesis of the predicted periplasmic tyrosine residues of TpbB was performed to convert them to phenylalanine (Y48F, Y62F, and Y95F); it was reasoned that phenylalanine would provide a similar bulky side chain but remove the hydroxyl moiety needed for phosphorylation [66]. The mutations were introduced into pMQ70-tpbB using Pfu DNA polymerase and QuikChange Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA), and the primers are listed in Table S3. The resulting plasmids, pMQ70-tpbB-Y48F, pMQ70-tpbB-Y62F, and pMQ70-tpbB-Y95F were transformed into the tpbB mutant by conjugation and aggregation was assayed. DNA sequencing was used to confirm the tyrosine mutations and that no other mutations were introduced into the promoter or protein-coding sequences.
The promoter region of tpbA (ptpbA), including 399 bp upstream of the start codon and 31 bp of the open reading frame, was amplified using Pfu DNA polymerase with primers pPA14_13660-F-HindIII and pPA14_13660-R-BamHI (Table S3). The PCR product (430 bp) was cloned into the HindIII/BamHI sites of pLP170 to produce pLP-ptpbA, and it was conjugated into PA14 and the QS-related mutants using helper strain HB101/pRK2013 [67],[68]. Transformants were grown overnight in LB medium with 300 µg/mL carbenicillin, reinoculated at a turbidity of 0.05 at 600 nm, and grown for another 6 h. Biofilm cells were harvested from 4 g of glass wool after incubation for 6 h in LB at 37°C with shaking at 250 rpm. β-galactosidase activity was measured using suspension cells and biofilm cells as described previously [69]. Similarly, β-galactosidase activity of plasR::lacZ (pPCS1001) and prhlR::lacZ (pPCS1002) was examined in PA14 and the tpbA mutant.
|
10.1371/journal.pcbi.1001051 | Theoretical Analysis of the Stress Induced B-Z Transition in Superhelical DNA | We present a method to calculate the propensities of regions within a DNA molecule to transition from B-form to Z-form under negative superhelical stresses. We use statistical mechanics to analyze the competition that occurs among all susceptible Z-forming regions at thermodynamic equilibrium in a superhelically stressed DNA of specified sequence. This method, which we call SIBZ, is similar to the SIDD algorithm that was previously developed to analyze superhelical duplex destabilization. A state of the system is determined by assigning to each base pair either the B- or the Z-conformation, accounting for the dinucleotide repeat unit of Z-DNA. The free energy of a state is comprised of the nucleation energy, the sequence-dependent B-Z transition energy, and the energy associated with the residual superhelicity remaining after the change of twist due to transition. Using this information, SIBZ calculates the equilibrium B-Z transition probability of each base pair in the sequence. This can be done at any physiologically reasonable level of negative superhelicity. We use SIBZ to analyze a variety of representative genomic DNA sequences. We show that the dominant Z-DNA forming regions in a sequence can compete in highly complex ways as the superhelicity level changes. Despite having no tunable parameters, the predictions of SIBZ agree precisely with experimental results, both for the onset of transition in plasmids containing introduced Z-forming sequences and for the locations of Z-forming regions in genomic sequences. We calculate the transition profiles of 5 kb regions taken from each of 12,841 mouse genes and centered on the transcription start site (TSS). We find a substantial increase in the frequency of Z-forming regions immediately upstream from the TSS. The approach developed here has the potential to illuminate the occurrence of Z-form regions in vivo, and the possible roles this transition may play in biological processes.
| We present the SIBZ algorithm that calculates the equilibrium properties of the transition from right-handed B-form to left-handed Z-form in a DNA sequence that is subjected to imposed stresses. SIBZ calculates the probability of transition of each base pair in a user-defined sequence. By examining illustrative examples, we show that the transition behaviors of all Z-susceptible regions in a sequence are coupled together by the imposed stresses. We show that the results produced by SIBZ agree closely with experimental observations of both the onset of transitions and the locations of Z-form sites in molecules of specified sequence. By analyzing 12,841 mouse genes, we show that sites susceptible to the B-Z transition cluster upstream from gene start sites. As this is where stresses generated by transcription accumulate, these sites may actually experience this transition when the genes involved are being expressed. This suggests that these transitions may serve regulatory functions.
| DNA often occurs in an underwound, negatively superhelical topological state in vivo. In bacteria, gyrase enzymes act to generate negative supercoils, while topoisomerases dissipate them. The dynamic balance between these two processes determines a basal level of superhelicity that can change according to the environmental or nutritional state of the organism [1]. In addition, RNA polymerase translocation leaves a wake of negative supercoils and generates a bow wave of positive supercoils [2]–[4]. Together these effects induce substantial amounts of superhelicity in the topological domains into which bacterial genomes are subdivided. A variety of regulatory processes in prokaryotes, including the initiation of transcription from specific genes, are known to vary with the level of superhelicity experienced by the DNA involved [5].
It has long been thought that unconstrained superhelicity was not a factor in eukaryotic genomic regulation. Eukaryotes do not commonly have negatively supercoiling gyrases while they do have relaxing topoisomerases. Also nucleosomal winding both stabilizes supercoils and could inhibit the transmission of unconstrained superhelicity. However, it is now known that substantial amounts of transcriptionally induced negative superhelicity occur upstream (i.e. 5′) of RNA polymerases in the human genome [6], [7]. A superhelix density of is achieved there by a single transcriptional initiation event, while divergently oriented transcription can produce superhelix densities of in the region between the polymerase complexes. This superhelicity extends over at least kilobase distances, hence must be transmitted either through or around nucleosomes. Kinetically, this transcription driven superhelicity is generated faster than topoisomerases act to relieve it, so it abides long enough to be able to affect subsequent regulatory processes.
The levels of negative superhelicity achieved in both prokaryotes and eukaryotes are sufficient to drive in vivo structural transitions to alternative DNA conformations [7], [8]. The most studied DNA transition is superhelically induced duplex destabilization (SIDD), which facilitates or creates local sites of strand separation. SIDD has been implicated in a wide variety of regulatory processes, including the initiation of transcription from specific promoters in both prokaryotes and eukaryotes [9]–[17].
Here we focus on the transition from B-form to Z-form, a left-handed double helix. When the discovery of Z-DNA was announced this transition was predicted to occur at physiologically attained levels of negative superhelicity [18]–[20]. Z-DNA has been experimentally detected at inserted Z-susceptible sites in bacterial genomic DNA both in vitro and in vivo [21]–[26].
The study of alternate DNA structures in eukaryotes is more challenging, in part because DNA superhelicity in these organisms seems not to be stable, but rather is a transient state driven by transcriptional activity. However, there is substantial indirect evidence that Z-DNA also can occur in vivo in eukaryotes. Z-DNA has been implicated in a variety of regulatory events relating to replication, transcription, recombination, and other biological processes [27]. For example, it has been shown that the negative torsional stress induced by polymerase translocation during transcription can stabilize Z-DNA near transcription start sites [28]. The amount of Z-DNA found in these experiments was directly related to transcriptional activity, and thus to the level of transcription-driven superhelicity. Another set of experiments studied the formation of Z-DNA in the 5′ flank of the human c-myc gene [29], [30]. Three Z-susceptible regions were identified near the promoters of this gene. These experimental results suggest that the regions involved transform to Z-form during c-myc transcription, but revert to B-form when transcription is inhibited. These experiments indicate that transcriptionally driven superhelical stresses can drive B-Z transitions in mammalian cells.
Many attempts have been made to identify proteins that bind selectively to Z-DNA. A powerful method developed by Herbert [31] led to the isolation of double-stranded RNA adenosine deaminase (ADAR1) [32], a Z-DNA binding enzyme, as well as other Z-binding proteins. It has been shown that E3L, a Z-DNA binding protein found in poxviruses, inhibits the host cell's ability to perform transcription or mount an anti-viral response when it is bound to Z-DNA near transcription start sites [33]. On this basis it was suggested that an inhibitor of E3L binding might protect against poxviral infection. Although there are some indications that Z-binding proteins may be involved in gene regulation, this remains an active area of research [27].
The Z-form helix has dinucleotide repeat units, one of which must be in the syn- and the other in the anti-conformation, with helicity of −12 base pairs per turn [34]. (The minus sign indicates the left-handedness of the helix.) The free energy required for the B-Z transition under low salt conditions has been determined for each of the ten dinucleotides [21], [35]–[39]. The Z-form is energetically most accessible for certain alternating purine-pyrimidine sequences, the most favored being , with guanine in the and cytosine in the conformations. Z-formation has also been observed in sequences, although transitions there are almost twice as costly as at GC runs. The remaining alternating purine/pyrimidine sequence, , has a very high transition energy and is not normally found in Z-form. Perturbations which break the purine/pyrimidine alternation, although energetically costly, have also been observed in Z-DNA, as will be discussed below. The substantial nucleation energy for initiating a run of Z-DNA, which may be regarded as the cost of generating two junctions between B-form and Z-form, also has been determined [21], [40].
Soon after the discovery of Z-DNA several simple theoretical analyses of superhelical B-Z transitions were developed. These all assumed the simplest conditions of a single, uniformly Z-susceptible site embedded in an entirely Z-resistant background. The first such analysis simply predicted that physiological levels of negative superhelicity could drive B-Z transitions [18]. This approach was subsequently used to investigate the basic properties of these transitions, and to assess how the B-Z transition might compete with others in simple paradigm cases [19], [36], [41]–[43]. Finally, these simple theoretical approaches were applied to determine the energy parameters of the transition from experiments in which a single uniform insert (commonly ) placed within a superhelical plasmid was observed to undergo transition [21], [36], [40].
In this paper we present the first method to analyze the superhelical B-Z transition in its full complexity. This method, which we call SIBZ, can calculate the B-Z transition behavior of multi-kilobase length genomic DNA sequences under superhelical stress. It specifically includes the competition for transition among all sites within the sequence. SIBZ analyzes the states available to the entire sequence, where each base can be found in either the B-conformation or as a part of a Z-form dinucleotide pair. It then uses statistical mechanics to determine the equilibrium distribution among these states. Specifically, it calculates the probability of B-Z transition for each base pair in the sequence under the given conditions. In this way it identifies the Z-susceptible regions within the sequence, and assesses how they compete at any given level of superhelicity.
SIBZ was developed by modifying the SIDD algorithm to treat the B-Z transition, as described in the following section. Several other theoretical strategies have been developed or proposed for analyzing superhelical DNA transitions, which also might have been modified for this purpose. Although a formally exact method has been suggested based on recursion relations, it was found to be too computationally inefficient to warrant development [43], [44]. So an approximate algorithm was presented in the same paper that could make base pair-specific calculations. This method has not been made available for public use or evaluation. An alternative exact algorithmic strategy also has been developed and presented [45]. Although this approach could compute transition profiles (i.e. transition probabilities for each base pair), it too was found to be too computationally cumbersome to be practical. So a more efficient approximate method based on its approach was also presented. To create SIBZ we chose to modify the SIDD approach because it has been extensively developed, optimized and implemented in this group, and it features an attractive combination of high accuracy and computational efficiency.
There have been three previous theoretical methods implemented that analyze DNA sequences to identify potential Z-DNA forming regions [35], [46]–[48]. The first method, developed by the Jovin group, seeks to identify Z-susceptible sites based solely on their sequence characteristics [46]. The energetics of transition were not considered in this approach. Another method, called Z-Catcher, performs a mechanical calculation, but does not consider the thermodynamic equilibrium of the system [47]. Z-Hunt [35], [48] uses statistical mechanics, but only calculates the propensity of each fixed region within the sequence to form a Z-helix in isolation. Since the superhelical stresses that drive B-Z transitions couple together the transition behaviors of all base pairs that experience them, these approaches do not give information about how these competitive transitions behave in situ.
A DNA molecule in a topological domain is constrained by the constancy of its linking number , the number of times each DNA strand links through the loop formed by the other strand. The linking number of a relaxed domain is denoted . DNA domains in vivo are often found in a negatively superhelical state, in which . The resulting linking difference , also called the superhelicity, acts to deform the molecule, in particular imposing untwisting torsional stresses. These torsional stresses can be partially or fully relieved by local secondary structural transitions to conformations that are less twisted in the right-handed sense than the B-form. This absorbs some of the linking difference as the change of twist at the transition site, which allows the balance of the domain to relax a corresponding amount. A transition becomes favored when the decrease in stress energy it provides exceeds its cost.
To perform a rigorous statistical mechanical analysis of this phenomenon one must know four things about the transition involved. First, the sequence dependence of the free energy of transition is required. Second, one needs the nucleation energy of the transition. Third, the geometry of the alternate structure determines the amount of relaxation that the transition provides. For the Z-form this is a left-handed helix with 12 base pairs per turn. Fourth, one must know the relative flexibility of the alternate structure because, if it is more flexible than the B-form, its torsional deformation may be able to relieve additional stress. This is true for strand separation because single strands of DNA are quite flexible. However, since Z-DNA is a rigid structure it is not relevant for B-Z transitions. With this information one can perform a statistical mechanical analysis of the transition in a superhelical domain having any base sequence, as described below.
An equilibrium statistical mechanical model of the superhelical strand separation transition has already been developed [45], [49], [50]. This approach, known as SIDD, has been shown to accurately predict the locations and relative amounts of destabilization of the DNA duplex experienced in defined, kilobase scale sequences at specified superhelicities [9]–[12], [15], [17], [45], [51]. In this paper, we modify the algorithmic strategy used in the SIDD method to treat stress-induced B-Z transitions (SIBZ).
There are possible states available to a sequence of base pairs that is subject to a monomeric two-state transition (that is, one in which any individual base pair can either be in the B-form or in the alternate state). This number does not depend on whether the sequence is linear or circular. This is the situation for the strand separation transition, in which the repeat units of both states are monomeric. However, it does not hold for the B-Z transition because the repeat unit of Z-DNA is two base pairs (dimeric), while that of the B-form is a single base pair (monomeric). We will first derive an expression for the number of Z-form states available to a linear molecule of base pairs, and then use this result to determine the number of states of a circular molecule having the same length.
Let denote the number of states available to a linear molecule comprised of base pairs experiencing the B-Z transition. In any given state each base pair in the sequence is either a monomer (i.e. in B-form) or part of a dimeric pair with one of its neighbors (i.e. in Z-form). There are two possible arrangements for the first base pair in the sequence. It can be a monomer (i.e. in B-form), in which case there are ways that the rest of the sequence can be arranged. Otherwise, it can be in a dimeric unit with the second base pair (i.e. in Z-form). In this case the disposition of the first two base pairs is determined, so there are ways to arrange the rest of the sequence. Because these two alternatives are mutually exclusive and exhaust the possibilities, it follows that(1)Using the fact that and , one can calculate for any length from Eq. (1). This recursion relation together with these initial conditions show that , the st Fibonacci number. An excellent approximation to this number for all is given by(2)This shows that the number of possible state in the linear B-Z transition grows exponentially with sequence length , but with base equal to the golden ratio , rather than base 2 as holds for transitions in which both states have monomer units.
All the states found above for a linear molecule also are available to a circular molecule of the same length. However, in this case it is also possible that base pairs 1 and can form a dimeric unit, provided neither is already dimerized with its other neighbor. The number of states of the linear molecule in which neither base pair 1 nor base pair are dimerized is . So this is the number of states of the circular molecule in which base pair 1 dimerizes with base pair . It follows that the number of states available to the circular molecule is(3)
In principle all states available to the molecule compete for occupancy. Once the free energy associated to each state has been evaluated, the partition function may be calculated as(4)where the sum is over all states, and , where is the Boltzmann constant and is the temperature.
At thermodynamic equilibrium the available states are weighted according to the Boltzmann distribution. That is, in the equilibrium distribution each state occurs with relative frequency(5)This means that the occupancy of states decreases exponentially as their free energies increase, so only the relatively low energy states are significantly occupied. At equilibrium the ensemble average value of a parameter , that has value in state , is given by(6)The equilibrium probability of transition for base pair is found by averaging the parameter according to the above equation, where in any state where base pair is transformed, and in all other states. The transition profile is the graph of vs . It shows the probability of transition for each base pair in the sequence under the assumed conditions. As will be shown below, this profile can change significantly as the imposed superhelix density changes.
We consider a DNA molecule containing base pairs of defined sequence, on which a superhelical density is imposed. Here , where bp/turn is the helical twist rate for the B-form. A state of this molecule assigns to each base pair one of two conformations, either B-form or Z- form. This is done in a manner consistent with the dinucleotide repeat unit of Z-DNA, as described below. The residual superhelicity in that state is the linking difference remaining to stress the molecule after the change of twist consequent on transition. This includes the untwisting of the transformed base pairs from the right-handed B-form to the left-handed Z-form, together with a small amount of untwisting of the two stands at each B-Z junction [21]. Thus, in a state where bases pairs are in the Z-form the residual superhelical linking difference is given by(7)Here bp/turn is the twist rate for the B-helix, bp/turn is the twist rate of the Z-helix, and is the number of runs of Z-form DNA. (A run of transformed base pairs is defined as a maximal segment in which all base pairs are in the non-B structure.) The twist at a B-to-Z junction has been measured to be turns [21]. In our current applications the superhelicity is regarded as remaining constant.
Next we determine the free energy of each state of this molecule. This is comprised of the energy cost of the transition, plus the energy of accommodating the residual superhelicity. The quadratic free energy associated to residual superhelicity [52] is given by(8)Here [53], [54], is the gas constant, and is the absolute temperature. The energy cost of the transition includes two factors. First, a nucleation energy for each run of Z-DNA is required to form the junctions between the B-form and the Z-form. This has been measured to be approximately 5.0 kcal/mol/junction, so the nucleation energy for each Z-DNA run is kcal/mol [21], [36], [40]. Second, the energy to transform the specified base pairs into the Z-DNA conformation is also needed. The published values of the B-Z transition free energies for all 10 dinucleotide pairs are given in the second and third columns of Table 1 [35]. The bases in each dinucleotide pair must alternate, one in anti and the other in syn conformation. As the values in the table show, the transition energy of a particular dinucleotide depends strongly on whether it is AS (5′ anti 3′ syn) or SA (5′ syn 3′ anti). A Z-Z junction occurs when adjacent dinucleotides have different anti-syn alternations, either (AS)(SA) or (SA)(AS). This violation is energetically costly, as shown in the last column of the Table.
Some of the energy values shown in Table 1 are calculated estimates, while others were measured experimentally [21], [35], [37]–[39]. The energies shown can be evenly divided between the two base pairs involved, whether in a dinucleotide or in a Z-Z junction, to determine the transition free energy associated with each base pair.
It is energetically most favorable for purines and pyrimidines in the Z-form to be in the syn and the anti states, respectively. There is a substantial energy cost for a base pair to have the opposite conformation. Although in principle every dimeric sequence can be driven into Z-form, four (CG, GC, CA = TG, and AC = GT) have substantially lower transition energies than do the others. The equilibrium distribution will be dominated either by the untransformed state or by states in which transition occurs at sequences composed of the energetically most favored dinucleotides.
The total free energy of a state that has specified base pairs comprising dinucleotide repeat units in runs of Z-form is given by(9)The precise manner in which the base pair transition energies are determined from Table 1 is described below.
We evaluate the equilibrium B-Z transition in a negatively supercoiled DNA molecule by appropriately modifying the previously developed SIDD method, which has been described elsewhere [45], [49], [51], [55]. Briefly, one first finds the lowest energy state of the system, whose free energy is denoted by . Then one sets an energy threshold , and finds all states whose free energy does not exceed . This is done by an exhaustive enumeration procedure [45], [49]. From these states one calculates an approximate partition function and the equilibrium ensemble average values of all quantities of interest.
Since the population of a state at equilibrium decreases exponentially as its free energy increases, the states that are neglected because their energies exceed the threshold are occupied with very low frequencies. If, for example, the threshold is set at kcal/mol, as is used below, the neglected states are occupied no more than times the frequency of the lowest energy state at K. A careful density of states calculation was performed to assess the aggregate influence of the neglected high energy states, and to approximately modify the calculated ensemble averages to account for them [49], [51]. Although this correction could not be performed on the transition probabilities of individual base pairs, this approach showed that all other ensemble averages were accurate to between four and five significant figures. Subsequently, we developed an exact, but very slow algorithm that performs these calculations [50]. By comparing its results with those of SIDD at different energy thresholds it was determined that, when thresholds in the range 10 to 12 kcal/mol were used, the accuracy of all calculated parameters, including the transition probabilities of individual base pairs, exceeds four significant digits [45].
Since there is a substantial nucleation energy kcal/mol associated with opening each run [21], [36], [40], states with numerous Z-runs will be correspondingly less populated. This allows us to impose a cutoff on the number of runs that may occur. In the SIDD analysis of strand separation it was found that three run states were the most that were encountered at any reasonable energy threshold and superhelix density . However, sequences that are energetically most susceptible to the B-Z transition tend to be shorter than the A+T-rich regions that favor denaturation. As shown by the energies in Table 1, there are many very costly types of “imperfections” which may hinder the extension of a Z-DNA run. So in the B-Z transition it can be energetically less expensive to initiate a new run at another favorable site than to extend an existing run into an energetically unfavorable region. Extensive sample calculations performed in tuning the SIBZ algorithm have shown that limiting consideration to states with a maximum of four runs is sufficient for high accuracy at the superhelical densities and sequence lengths of interest in this paper.
One important step in the SIBZ algorithm is the assignment of transition energies to the dinucleotide repeat units in each Z-run. This is complicated because the transition energy associated to each unit can have any of three significantly different values depending on the anti and syn characters of the base pairs in that unit and in its neighbors. Briefly, within each unit one base pair must be in syn and the other in anti. Also there is a significant energy penalty assessed in cases where there are Z-Z junctions with one or both neighbors.
In the SIBZ algorithm we use the following procedure to assign the energetically most favorable anti or syn conformations to all base pairs in a potentially Z-forming run. First, since the unit cell of Z-DNA is a dinucleotide, we allow only an even number of base pairs in any Z-run. We assign all purines in the run to be syn and all pyrimidines to be anti, as the dinucleotides with the lowest transition energies have this character. (See Table 1.) Since in most cases forming a Z-Z junction is more costly than flipping a base pair into its non-favorable conformation, we next change the conformation of a base pair if it has the same anti or syn character as both of its nearest neighbors. This procedure eliminates any repetition of the same conformation (anti or syn) longer than two base pairs. Then we search for quartets of the form AASS or SSAA. In these we flip the central two bases so that an alternating syn-anti character is obtained. Finally, when two bases within a dinucleotide unit have the same conformation, which is not permitted in Z-DNA, we flip the base pair which yields the minimum Z-Z junction energy with its neighbor. There is an ambiguity in this procedure for internal runs of even length at least four, such as ASSSSA, due to the order in which they are flipped. However, the states involved are always high energy because they will have at least one Z-Z junction as well as at least two unfavorable dinucleotides.
To determine the B-Z transition energies of a region, we scan its sequence by dinucleotide units, adding energies of syn-anti or anti-syn pairs according to Table 1. Whenever the alternating anti-syn character is disturbed we add the appropriate Z-Z junction energy. Also, we set the minimum length of a Z-run to be eight base pairs, as shorter regions have not been seen experimentally to form Z-DNA [21]. In this way we assign a B-Z transition energy to each segment in the DNA sequence of even length between 8 and 250 base pairs, which is a reasonable maximum cutoff for a Z-run length.
The algorithmic strategies for finding the lowest energy state, identifying all states that satisfy the threshold condition, and analyzing them to determine equilibrium values of parameters of interest are the same as those developed previously in the SIDD algorithm [45], [49]. We treat circular sequences as described there. A linear sequence is circularized by joining its end to its start with 50 T bases, and the resulting sequence is treated using the same method. In this case we do not report the information for the augmenting segment. These issues and methods have been described previously [45], [49], [51].
To assess its performance characteristics, we used SIBZ to analyze the pBR322 plasmid ( bp) using an energy threshold of kcal/mol. At superhelical density this analysis took 0.12 minutes to run on a MacBook Pro with dual Intel processors. On average there were 23.9 Z-form base pairs and 2.3 runs of transition. A total of 15,829,349 states were found to satisfy the energy cutoff condition . At superhelical density this analysis took 2.25 minutes to run on the same machine. In this case there were averages of 44.2 Z-form base pairs and 3.9 runs of transition, and 1,047,067,293 states satisfied the energy cutoff condition. We find that the execution time is almost constant at superhelix densities , and increases quadratically thereafter. The algorithm scales approximately linearly with sequence length. These performance characteristics suggest that the SIBZ analysis of the complete human genome at would take approximately 12 hours on a 100 CPU cluster of slightly faster (viz. Opteron) processors, if the sequence was partitioned into 5 kb segments that were analyzed individually. A similar analysis at would take approximately ten days.
We note that there is substantial variability of execution speed depending on the attributes of the sequence being analyzed. SIBZ executes quickly on sequences that have one dominant Z-susceptible site. However, the analysis under identical conditions of a sequence in which there is substantial competition among numerous sites can take up to ten times longer.
Z-Hunt was the first algorithm to predict Z-forming regions in DNA sequences based on energy considerations [35]. Although the original version only accepted sequences shorter than 1 Mbp, recently Z-Hunt was implemented to identify potential Z-forming regions in longer sequences, and specifically in the human genome [48]. In both versions of Z-Hunt a series of fixed length segments within a sequence are separately tested for their Z-forming potential. This is done by inserting the segment in a standard background, which is a circular plasmid in which the inserted segment is the only site that can undergo a structural transition. Z-Hunt then calculates the propensity of the segment to form Z-DNA under these standardized conditions. A Z-score is assigned to each segment by comparing its ability to adopt Z-form with those of a collection of randomly generated sequences. Unlike in SIBZ where we assign a superhelical density, Z-Hunt bases its Z-score on the superhelix density at which onset of transition occurs in this standard background. So there is no direct relationship between a segment's Z-score and its probability of transition at a specific superhelix density. Z-Hunt also provides no information about the competition among multiple Z-susceptible regions within the sequence.
Z-Catcher uses a different approach to identify sites with Z-forming potential [47]. This algorithm includes a superhelix density as one of its inputs [47]. It treats the B-Z transition as a simple binary, “on-off” process at a single site. A critical threshold superhelix density is calculated for each individual segment of the sequence being analyzed, at which the energy required by the B-Z transition of that site exactly balances the stress energy released from this transition when it occurs alone in a standard background. If the input superhelix density is more negative than this critical , the region is said to be Z-forming. Its output is a list of predicted Z-forming sites, with no weight or probability assigned to them. Z-Catcher analyzes individual sites as though complete transition at that site is the only possibility. No consideration is given to how each site competes with all other sites having Z-forming potential within the rest of the sequence. This algorithm is purely mechanical; it does not analyze the equilibrium behavior of the sequence.
SIBZ is the only method developed to date that analyzes the fully competitive B-Z transition behavior of DNA sequences in situ at thermodynamic equilibrium under any level of negative superhelicity. It is the only approach that calculates the equilibrium probabilities of transition for each base pair under the given conditions. This provides a more realistic and rigorous analysis, and enables more direct comparisons to be made between its predictions and experimental results than are possible with either Z-Hunt or Z-Catcher.
We have applied the SIBZ algorithm to analyze a variety of DNA sequences for their Z-forming properties. These include artificial sequences designed to investigate attributes of the transition, as well as sequences from bacteria and from eukaryotes. We use SIBZ to calculate ensemble average numbers of Z-form base pairs and runs of Z-DNA, as well as the transition profile, which is the graph of the equilibrium probability for transition of each base pair in the sequence. These calculations illustrate how the transition properties vary with superhelical density, and allow comparisons with experimental results and with the results from other methods. Finally, we calculate the propensity for Z-forming regions to be found near the transcription start sites of 12,841 mouse genes.
The B-Z transition behavior of susceptible regions within a DNA sequence can vary in complicated and highly interdependent ways. This complexity arises because superhelical stresses globally couple together the transition behaviors of all base pairs that experience them. When one region undergoes transition, the change of twist involved fractionally relaxes the level of stress experienced by all other base pairs in the domain. This can be seen from Eq. (7), where a change in the number of transformed base pairs causes a corresponding change in the residual superhelicity experienced by the entire domain. In consequence, the transition behavior of each base pair is affected by the transformation of any other base pair. This global coupling is the primary reason why superhelical transitions cannot be understood by studying individual sites in isolation, but must be considered in their actual context.
This competition between different Z-forming regions within a sequence can lead to a rich repertoire of complex, interactive behaviors. We illustrate this with sample calculations on a designed sequence containing two regions susceptible to Z-DNA formation. This sequence consists of 5000 T base pairs, into which we insert two Z-susceptible regions at distant locations. The thymidine background is chosen to insure that only our inserted segments are susceptible to Z-formation. The first insert is a segment while the second is a segment that contains six Z-Z junctions. The segment is less costly to transform, because the Z-Z junctions in the segment are energetically expensive. However because it is shorter, transition at the segment also relieves less superhelicity. We used SIBZ to calculate the probability of transition of each of these regions over a range of negative superhelix densities. Plots of these probabilities as a function of are shown in Fig. 1.
Just beyond the onset of transition where is still small, the superhelical free energy of the untransformed state also is relatively small. Under these circumstance the energy relief afforded by transition is less than it is at more extreme superhelicities. So in this regime the magnitude of the transition energy is the dominant factor in determining which regions transform. This is why the shorter but energetically less costly Insert 1 is the first to transform, as shown in the figure. As increases the superhelical free energy becomes quadratically larger. Now transitions at longer sequences become more desirable because they relieve more superhelical stress energy. Under these circumstances the difference in transition free energy due to the ZZ-junctions becomes less important than the benefit afforded by transforming a substantially longer segment. For this reason a coupled transition-reversion event occurs around , in which transition of Insert 2 is coupled to the reversion of Insert 1 back to B-form. In the range it is energetically too costly for both segments to transform to Z-DNA simultaneously, so such states occur infrequently at equilibrium. Transition of the long Insert 2 has caused substantial relaxation,which decreases the residual superhelicity felt by Insert 1 below the value that would drive it to transform. So at these stress levels the probability of transformation of Insert 1 drops to near zero. As increases beyond the point where Insert 2 has a high probability of being entirely in Z-form, the additional stress accumulates as negative residual superhelicity. When this reaches a sufficient level Insert 1 again transforms to Z-DNA. Beyond both inserts have high probabilities of simultaneously being in Z-form. One sees that there are specific intervals of superhelicity within which 1) neither insert transforms, 2) the first transforms but not the second, 3) the second transforms but not the first, or 4) both inserts transform simultaneously. The transition in this example experiences every logical possibility.
When Z-Hunt is applied to this sequence it assigns Z-score of to Insert 1, and to Insert 2. The above analysis shows that when the competition between sites is considered there are circumstances when a region with a lower Z-score may transform while one with a higher Z-score does not. The analysis of individual sites in isolation simply does not capture the complexity of behavior that can occur in stress-driven transitions.
We have analyzed the B-Z transition properties of three circular DNAs - the pBR322 plasmid, and the X174 bacteriophage and Bdellovibrio phage MH2K genomes. Fig. 2 shows the B-Z transition probability profiles of these sequences calculated at superhelical density . In each case the B-Z transition is substantially confined to a small number of sites where it is energetically most favorable, although there are several additional locations that have smaller, but still significant, transition probabilities. All sites with transforming potential are seen to be relatively short. The longest Z-forming regions found in any of these three sequences at contains 14 base pairs. In each of these sequences there is at least one predominant region whose probability of forming Z-DNA exceeds 70%. In pBR322 the largest peak has probability near 70%, and there are three other sites whose probabilities exceed 25%. Phage X174 contains a single region, slightly longer than that in pBR322, whose transition probability is close to unity. Because this region is so dominant at this superhelix density, other portions of this sequence have only low probabilities of Z-formation. This dominance is a consequence of this site having a highly favorable transition energy over a sufficiently long region.
We compared the performance of SIBZ with those of Z-Hunt and Z-Catcher when run on these sequences [35], [47], [48]. The energies from Table 1 were used in all three programs. When Z-Hunt was applied to the pBR322 sequence it found one 15 bp long segment at location 1448 with a Z-score of 2444, and a 17 bp long segment at position 1407 with a Z-score of 1845. As shown in Fig. 2a, SIBZ also finds these two peaks to be the most dominant, with the segment at location 1448 having the higher transition probability. This agrees with the relative Z-score rankings provided by Z-Hunt. However, the relative probabilities of these two regions are not proportional to their Z-score. Moreover, SIBZ documents several other regions that also have significant transition probabilities that Z-Hunt does not identify. One sees that, although the sites found by Z-Hunt agree with the major sites found by SIBZ, the latter provides more information regarding other Z-susceptible regions. The results from SIBZ, because they are expressed as transition probabilities of the fully competitive transition at the assumed superhelix density, are both more precise and more easily interpretable than are Z-scores.
Z-Catcher does not identify any Z-susceptible regions in pBR322 at superhelix density . At superhelix density it finds the two dominant segments at locations 1407 and 1448. It also finds two other Z-segments at positions where no Z-forming potential is seen by the other algorithms. The segment at position 1448 is found to comprise 28 base pairs, which is significantly longer than predicted either by Z-Hunt or by SIBZ. Thus, the predictions of Z-Catcher seem to differ considerably from those of either Z-Hunt or SIBZ.
The behavior of a B-Z transition varies significantly as the negative superhelical density is modified. In general, as is increased, larger numbers of transformed base pairs are required to relieve torsional stresses. If the most energetically susceptible region is sufficiently long, the most favored way to do this would be by extending the transition to encompass increasing amounts of this region. However, the most Z-susceptible regions in natural DNA sequences are usually relatively short, as is seen in these three sequences. In that case what commonly happens is that, as (and hence also the level of imposed stress) increases, successively more energetically costly distinct Z-forming regions are transformed (possibly coupled to reversion of other regions as shown above). In either of these strategies the average number of base pairs adopting the Z-form increases with negative superhelical density. This is demonstrated in Fig. 3, which shows the average number of Z-DNA base pairs as a function of for pBR322 and for X174. One sees that in both sequences there is a threshold for the onset of transition around , and the expected number of transformed base pairs increases approximately linearly thereafter. (We note that the onset of transition in Fig. 1 occurs at a slightly less extreme superhelix density than is seen in Fig. 3. This occurs because according to Table 1 the shorter Z-susceptible insert in the constructed sequence has the lowest possible transition energy. This is not true of the most Z-susceptible site in either X174 or pBR322.)
Z-DNA forming regions have been observed with two-dimensional gel electrophoresis in plasmids engineered to contain Z-susceptible inserts [37], [56]. The susceptibility of a region to be driven into Z-form was found to depend on its base sequence and on the level of supercoiling of the plasmid. In one experiment Peck et al. inserted a sequence (here called Sequence 1) into the BamHI site of pBR322 [56]. This is the most energetically susceptible sequence to Z-DNA formation, as shown in Table 1. In another set of experiments pBR322 derivatives were analyzed, each of which contains an insert whose sequence includes “imperfections” relative to the optimal Z-forming segments [37]. Here we focus on two of these engineered plasmids, one with (Sequence 2) inserted into the BamHI site of pBR322 and the other with (Sequence 3) inserted into Pvu II site of pBR322.
The Z-forming insert sequences 1 and 2 both have the same length, each containing 16 dinucleotides (i.e. 32 base pairs). This means they have the same ability to relieve superhelical strain. However, Sequence 2 contains a Z-Z junction consisting of GG bases, which breaks the alternating syn-anti pattern and costs an extra 4 kcal/mol to form (see Table 1). So the onset of transition in Sequence 2 is expected to occur at a more extreme superhelix density because it has a higher total B-Z transition energy. Transition at Sequence 3 has two disadvantages relative to the others. It is shorter in length, containing 13 Z-forming dinucleotides instead of 16, and it has two energetically costly “imperfections”. The GA and TC nucleotides in the anti-syn conformation cost 3.4 kcal/mol each, resulting in an additional transition cost of 6.8 kcal/mol relative to a perfect alternating CG sequence. Therefore, the critical superhelix density for complete Z-formation of this region is expected to be substantially higher than for either Sequence 1 or 2.
Fig. 4 shows the probabilities predicted by SIBZ of the inserted sequences transforming to Z-DNA in each of these three pBR322 derivatives, plotted as a function of superhelical density . The curves labeled Sequence 1, Sequence 2 and Sequence 3 refer to the plasmids with those as their inserts.
Experimentally measurements have been made of the critical superhelical density required to flip the entire inserted sequence into Z-DNA [37], [56]. These analyses regarded the transition as two-state, analogous to an “on-off” switch, and determined the critical superhelicity at which these inserted segments were switched on. These critical densities are shown in column 3 of Table 2.
It is difficult to make exact comparisons between our theoretical predictions and these experimental results because they do not involve entirely comparable quantities. The experiments measure a critical for completion of an “on-off”, two-state B-Z transition, while SIBZ calculates the equilibrium probability of the plasmid experiencing any amount of transition. However, the trend observed in the SIBZ results of Fig. 4 closely agrees with experimental data. Transition occurs at the least extreme superhelix density in the plasmid with the Sequence 1 insert, later in the plasmid with Sequence 2 insert, and lastly in the plasmid with the Sequence 3 insert. Moreover, the horizontal displacement between these curves is about 0.0025 between Sequence 1 and Sequence 2, and about 0.008 between Sequence 2 and Sequence 3. These agree closely with the experimentally measured differences.
In order to directly compare the experimental finding with the SIBZ results we must decide what level of transition corresponds to the “on-off” switch having been thrown. We define this to occur when the probability of Z-form is 80%. Column 4 in Table 2 shows the superhelix densities at which this occurs for each of the three inserts, as determined from the curves in Fig. 4. At this level SIBZ results are seen to agree closely with the experimentally measured values.
We next compared the predictions of SIBZ with the results of antibody binding experiments that identified Z-forming regions in the human c-myc gene [29], [30]. Here the formation of Z-DNA is driven by the negative superhelicity that is generated when the gene is transcribed. This was confirmed by the observation that Z-formation was suppressed when transcription was inhibited.
The first experiment used anti-Z antibodies to isolate regions of Z-DNA formation [29]. They found three restriction fragments from the c-myc gene region that showed antibody reactivity when the gene was being transcribed. These three fragments are all located in the upstream region of the gene in proximity to its three promoters. Fig. 5 shows the SIBZ transition profile of a 5kb region around the c-myc gene that spans these locations. This profile was calculated at superhelical density , a reasonable value for transcriptionally driven superhelicity [6], [7]. The three large, gray horizontal bars labeled Z1, Z2, and Z3 identify the segments that were found in this experiment to contain Z-forming regions [29]. Five of the six sites identified by SIBZ as having the highest Z-forming probabilities occur within these three segments. The peak that is not within any of the three experimentally identified fragments is located around position 2100 in Fig. 5. Perhaps this region was missed because of its proximity to the Z2 fragment.
A second experiment identified the exact locations of the Z-forming sites within these three segments Z1, Z2, and Z3 [30]. This was done by isolating the fragments and inserting each individually into the circular pDPL6 plasmid [22]. Regions of Z-DNA were detected by measuring diethyl pyrocarbonate reactivity within these negatively superhelical constructs. One Z-form region was found experimentally in each of the Z1 and Z3 segments, and two such regions were found in Z2. The experimentally determined locations of these regions are shown in Fig. 5 by small, blue horizontal bars located below each larger, labeled segment. These regions coincide precisely with the strongest Z-forming sites predicted by SIBZ. In each segment all the sites that are predicted to be most Z-susceptible are found experimentally to actually be in Z-form. SIBZ correctly identifies all sites within the three segments that occur in Z-form, with no false positives. This shows that our theoretical model produces results that agree precisely with those obtained by experiments.
The experiments described above on the c-myc gene suggest that Z-forming regions may occur in proximity to transcription start sites. The accuracy of SIBZ in identifying these regions allows us to use it to address this question. To this end we analyzed 5 kb regions around the transcription start sites (TSSs) of 12,841 mouse genes. We oriented all genes to read left to right, and aligned them so their TSSs were at position 2500. We calculated the probability of B-Z transition at each of the 5,000 positions in each sequence. We then averaged the transition probabilities at each position that were calculated for all the sequences. This analysis was performed at two superhelix densities, and . The results are shown in Fig. 6.
A substantial enrichment of predicted Z-form sites is observed immediately upstream of the TSS at both superhelix densities, with the number of these sites increasing with . The number of predicted sites immediately downstream from the TSS is approximately half the maximum number immediately upstream. The density of predicted Z-form regions in the far upstream, inferentially intergenic portions of the sequences approximately equals that in the far downstream, inferentially transcribed regions, and is approximately 30% of the maximum at both superhelix densities.
To analyze this data further we define a region within an individual sequence as Z-forming if its probability of B-Z transition exceeds 80%. At superhelix density a total of 2519 of the 12841 genes (19.6%) are found to have one or more Z-forming regions within 1,000 bp upstream of (i.e. 5′ to) their TSS. At the more extreme superhelix density of this number increases to 4269 (33.2%). There is a clear enrichment of Z-forming regions directly upstream of TSSs relative to downstream. At there are 1083 genes that have one or more predicted Z-forming regions within 50 base pairs upstream of the TSS, and 572 genes with such a site within 50 base pairs downstream, a nearly two-fold change. At the corresponding numbers are 2077 genes and 1304 genes, respectively. We also find that the majority of these 12,841 mouse genes contain Z-susceptible regions somewhere within their 5 kb regions. At we observe that 32.7% of these genes do not contain a Z-forming region (at the 80% level) anywhere in their sequence. This percentage drops to 12.7% when .
A similar analysis was performed by Droege using Z-Catcher to analyze a large collection of human genes [47]. That work documented a similar enrichment of predicted Z-susceptible regions in the 5′ flanks of genes. Our results confirm and reinforce theirs, although the methods used in the two analyzes are not equivalent.
In this paper we have developed the first statistical mechanical method to analyze the competitive B-Z transition within long superhelical DNA domains of specified base sequence. The output of this method is the probability of transition calculated for each base pair in the sequence, rather than simply a list of sites or a Z-score. We have demonstrated the essentially competitive character of these transitions, and shown how the transition behavior can change in complicated ways with the level of imposed superhelicity. We find that there are regions with clear Z-forming potential in genomic DNA sequences. Our results agree in detail with experimental measurements of both the onset of transitions and the locations of Z-forming regions.
Our analysis of 12,841 mouse genes documents a substantial increase in the occurrence of potential Z-forming regions immediately upstream from transcription start sites. At a superhelical density of we find that 33.2% of these genes have one or more Z-forming regions within 1,000 bp upstream of their TSS. Approximately half of these have such a site within 50 bp 5′ of the TSS. We note that in eukaryotes this superhelical density is attained in these regions through transcription-driven superhelicity [7]. This suggests that Z-DNA could play roles in transcriptional regulation. However, the fact that less than half the genes have Z-forming regions in their immediate 5′ flanks (i.e. within 1 kb of the TSS) suggests that there may be distinct classes of genes, some of which use such sites for regulatory purposes while other do not. Also, since Z-forming regions are relatively short, around 14 base pairs, there is room for them to be interspersed with other motifs, including A+T rich regions. These issues will be addressed more fully in a subsequent paper.
When we compare the results of SIBZ with those of Z-Hunt and Z-Catcher we find that Z-Hunt and SIBZ agree in identifying the most dominant sites. However, SIBZ also identifies several sites where the probability of transition remains significant, that neither of the other methods find. Z-Catcher seems to be less sensitive than either of the others, only identifying the most susceptible sites when its input superhelix density is relatively large. Neither Z-Hunt nor Z-Catcher analyze the actual B-Z transition behavior of a superhelical DNA sequence, which is competitive and can vary in highly complex ways with the superhelix density. Instead they seek only to identify those individual sites within the sequence that have the greatest potential to form Z-DNA. They say nothing about how these sites compete, and provide no information about transition probabilities that can be directly compared with experimental results.
We note, however, that Z-Hunt and Z-Catcher still may prove useful for specific purposes. In particular, in our experience each is substantially faster than SIBZ. It is difficult to compare their relative speeds, especially when they are only accessible through websites. However, both Z-Hunt and Z-Catcher appear to return results on 5 kb sequences almost instantaneously, whereas SIBZ takes from ten seconds to three minutes depending on conditions and sequence characteristics. Since it seems to identify the major sites reasonably well, Z-Hunt in particular may serve as an initial screen for such sites, with SIBZ used to perform full analyses as needed.
The methods presented here can be applied to any two-state transition, provided the geometry, deformability, and transition energetics of the states are known. In this regard the best characterized DNA transition is strand separation. Both enthalpies and entropies of denaturation have been measured for every base pair and every choice of its nearest neighbors. The dependence of the free energy on ionic strength also is known. Together these allow one to predict how this transition behavior will vary with changing ionic conditions and temperature, as well as with superhelicity.
At present the energetics of the B-Z transition are not so well characterized. It is known that specific alternating purine-pyrimidine sequences are substantially favored for Z-formation, and their transition free energies have been measured. But no quantitative information is currently available regarding how these free energies partition between entropy and enthalpy, nor about their ionic strength dependences. However, it has been reported that the B-Z transition shows little temperature dependence in the range between C and C [57]. This suggests that entropy changes are much smaller for B-Z transitions than for denaturation, where the DNA is disordered and its interactions with the solvent are thereby substantially altered. The extremely close accord documented here between the predictions of SIBZ and experimental results suggests that the transition energetics we use are accurate. We note that this accuracy is achieved without having any tunable parameters in our model.
Although SIBZ is effective at analyzing the B-Z transition behavior of a supercoiled DNA molecule, it still focuses on only part of the complete picture. It is well known that local strand separation at A-T rich regions also occurs in negatively supercoiled molecules. To enable the accurate analysis of the full transition behavior of superhelical DNA one must include both types of transitions in a unified model. This would permit all sites susceptible to SIDD and/or to B-Z transitions to compete. We are working to develop such a model.
A website is available (http://benham.genomecenter.ucdavis.edu) where the members of the scientific community may submit sequences of interest to them for analysis by the SIBZ algorithm. The sequence must be either in FASTA format or in a file that contains sequence characters exclusively. Sequences of any length up to 10 kb may be submitted, although sequences of length around 5 kb are preferred. This site may also be used for SIDD analysis of the same sequences.
In the near future we hope to analyze the SIBZ characteristics of a large number of genomic sequences, up to and including complete genomes of model organisms. The results for each sequence will be posted in a database on the same Web site as they are completed.
|
10.1371/journal.pntd.0007184 | Application of a targeted-enrichment methodology for full-genome sequencing of Dengue 1-4, Chikungunya and Zika viruses directly from patient samples | The frequency of epidemics caused by Dengue viruses 1–4, Zika virus and Chikungunya viruses have been on an upward trend in recent years driven primarily by uncontrolled urbanization, mobility of human populations and geographical spread of their shared vectors, Aedes aegypti and Aedes albopictus. Infections by these viruses present with similar clinical manifestations making them challenging to diagnose; this is especially difficult in regions of the world hyperendemic for these viruses. In this study, we present a targeted-enrichment methodology to simultaneously sequence the complete viral genomes for each of these viruses directly from clinical samples. Additionally, we have also developed a customized computational tool (BaitMaker) to design these enrichment baits. This methodology is robust in its ability to capture diverse sequences and is amenable to large-scale epidemiological studies. We have applied this methodology to two large cohorts: a febrile study based in Colombo, Sri Lanka taken during the 2009–2015 dengue epidemic (n = 170) and another taken during the 2016 outbreak of Zika virus in Singapore (n = 162). Results from these studies indicate that we were able to cover an average of 97.04% ± 0.67% of the full viral genome from samples in these cohorts. We also show detection of one DENV3/ZIKV co-infected patient where we recovered full genomes for both viruses.
| Dengue viruses 1–4 (DENV1-4), Zika virus (ZIKV) and Chikungunya virus (CHIKV) are tropical and subtropical viruses that share a common arthropod vector, and have very similar clinical presentations that are difficult to distinguish. With the recent outbreaks of DENV, ZIKV and CHIKV globally, a single methodology able to simultaneously distinguish these viruses and provide full-genome information would greatly increase our capacity to rapidly characterize outbreaks. As a proof of principle, we have applied this methodology to two large cohorts in Sri Lanka and Singapore taken during recent dengue and Zika outbreaks, respectively. Herein, we present the results of this application to these cohorts and provide the tools to replicate these methodologies for other cohorts.
| Dengue viruses 1–4 (DENV1-4), Zika virus (ZIKV) and Chikungunya virus (CHIKV) are viruses spread by the Aedes aegypti and Aedes albopictus and are among the foremost arboviral threats to humans today [1]. DENV and ZIKV are flaviviruses with positive-sense, single-stranded RNA genomes of ~11 kb that encode for a single polyprotein, which is then post-translationally cleaved into three structural proteins (C, prM and E) and seven non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) [2,3]. There are four different DENV serotypes (DENV1-4) that share 60–75% identity at the amino acid level. Collectively, DENV1-4 are responsible for an estimated 390 million infections yearly [4]. Evidence suggests that the geographic distribution of DENV is increasing [5] and poses a threat to travelers visiting affected regions [6].There is a single serotype of ZIKV that can be further broken down into two primary lineages, the African and the Asian/American. At the time of writing, ZIKV had spread to 86 countries [7], and is a documented cause of microcephaly in infants when transmitted vertically from an infected mother during pregnancy [8]. CHIKV is an alphavirus with three major lineages, all of which comprise a single serotype [9]. The genome of CHIKV is a positive-sense, single-stranded RNA genome of ~11.6 kb that is post-translationally cleaved into four nonstructural proteins (nsP1, nsP2, nsP3 and nsP4) and five structural proteins (C, E3, E2, 6k and E1) [9]. CHIKV has seen resurgence in the tropical and subtropical world with several notable epidemics in recent years [10].
Clinical presentation of infection with any of these six viruses is often similar; undifferentiated fever, headache, nausea/vomiting, persistent myalgia/arthralgia, and rash [11]. Although PCR assays that can discriminate between these viruses exist, they are often not used in conjunction and are reliant on short primer sequences designed to target relatively conserved regions of the viral genomes [12–15]. As these viruses all share a common replication strategy dependent on error-prone RNA-dependent RNA polymerases, rapid mutation necessitates the constant optimization of molecular diagnostic protocols for their accurate detection [12]. Further, these methodologies are limited in that they provide no information on the particular strain infecting the individual [12–15]. Recent studies have shown that polymorphisms in the viral genome can have profound effects on the pathogenicity and epidemic potential of the virus [16–22]. In response to these shortcomings, next-generation sequencing (NGS) is increasingly being used as a tool to obtain the full genome sequence of viruses in clinical samples [23]. However, the principle drawback to this methodology is the often-overwhelming amount of host material present in a clinical sample relative to the virus. In order to produce sufficient data for full viral genome assembly from these clinical samples, only a small number of samples can typically be run per lane of sequencing making this approach prohibitively expensive for many laboratories. To overcome the inherent limitations of this direct sequencing approach, a targeted enrichment approach to increase the sensitivity and efficiency of whole genome sequencing has been described for several pathogens of clinical importance [24–26]. One limitation of this approach is the high upfront cost associated with the enrichment baits. In the current study, we present a novel computation method, BaitMaker, to design baits that target the conserved and variable regions of a viral genome. The delineation of the conserved and variable regions is made possible by employing a computationally efficient k-mer based clustering approach on the available genetic information for DENV1-4, ZIKV and CHIKV in the NCBI database. We have then applied our methodology to two large cohorts: a collection of blood samples from the 2012–2015 DENV epidemic in Sri Lanka (n = 170) and samples collected during the 2016 outbreak of ZIKV in Singapore (n = 162) [27,28].
Targeted viral genome enrichment followed by high-throughput sequencing is an approach to enrich viral genomes present in meager quantities in clinical samples. The targeted-enrichment methodology uses biotinylated DNA baits, 120 nucleotides (nt) in length that are complementary to the viral genome. Commonly used algorithms to design baits generate tiled, overlapping baits across conserved genomic regions selected by multiple sequence alignment [24,29–32]. While effective, these methodologies can generate redundant and overlapping baits, which serves to increase the cost of the methodology in practice. In order to minimize the number of baits necessary to capture a target viral genome, we developed a new computational method called BaitMaker. BaitMaker generates non-overlapping baits at an interval of 500 nt in the viral genome. This interval was chosen under the assumption that an average deep sequencing library size ~300 nt and one bait can pull down two overlapping 300 nt DNA fragments. Hence, our approach differs from similar methodologies where placements of baits are tiled across the target genome [24,29–32] and allows for far fewer baits to be designed for each virus. In order to ameliorate the potential impact of reducing the number of baits available for capturing a targeted viral genome, BaitMaker incorporates a k-mer based pattern search and clustering strategy against a viral strain database (e.g. NCBI) to identify both conserved and diverse regions in the viral genome. BaitMaker then utilizes this information in one of two modes to design baits: (i) In ‘Conserved mode’, BaitMaker designs baits targeted to the species-level conserved regions whereas in (ii) ‘Exhaustive mode’, BaitMaker designs baits for both the conserved regions as well as regions with strain level variations. (Fig 1 and Methods). The source code for BaitMaker method is freely made available at GitHub: https://umasangumathi.github.io/BaitMaker/
For DENV serotypes 1, 2, 3 and 4, there were more than 11,000 genome sequences available in the NCBI database (S1 Table). Using BaitMaker-Conserved mode we generated conserved baits for each DENV serotype (65 baits total). As these baits are not designed to cover the genomic regions that are more variable, we designed an additional 22 baits that were specific to the Asian strains of DENV1-4 we were working with at the time (S2 Table). For ZIKV and CHIKV, there were fewer available sequences in the NCBI database at the time of design (238 and 1260, respectively) (S1 Table) hence, we developed the BaitMaker Exhaustive mode and utilized it to account for potentially under-represented variation in these viruses. The resulting panels for these viruses were 67 baits for ZIKV and 53 baits for CHIKV covering both conserved and variable regions of the respective genomes (S2 Table).
To create a virus capture panel targeting DENV1-4, CHIKV and ZIKV simultaneously, we pooled the 87 baits (65 targeting the conserved regions, 22 targeting the variable regions) specific to DENV1-4, the 67 baits specific to ZIKV and the 53 baits specific to CHIKV for a total of 207 baits. To assess whether we could effectively capture the targeted viral genomes in the context of high host background, we infected HuH7 cells with DENV1, 2, 3 or 4, Vero cells with ZIKV and BHK-21 cells with CHIKV. To simulate a higher level of host contamination that would potentially be found in clinical samples, the supernatant was discarded from these cultures and total RNA was extracted from the infected cell monolayers and Illumina libraries were constructed from the total RNA. These libraries were then divided in half where the first half was sequenced directly, and the second half was enriched with our bait panel prior to sequencing. The resulting sequencing reads from both conditions were then mapped against the reference genome for each respective strain. Overall, we observed an average 2-log increase in percentage of sequencing reads mapped to the viral genome following enrichment in the DENV1-4, CHIKV and ZIKV samples (Fig 2, S1 Fig and S3 Table).
In order to assess the sensitivity of the assay, a 1:4 serial dilution of total RNA extracted from ZIKV infected Vero cells was prepared. In order to keep the level of host RNA constant, RNA extracted from uninfected Vero cells was used as the diluent. An approximate increase of two Ct values for each successive dilution with a range from 29.79 to 41.08 was observed for threshold detection. Libraries from these dilutions were then enriched and sequenced (S3 Table). Our results indicate that as the Ct value increases, the depth of coverage decreases with respect to the breadth of genome coverage (Fig 3). In the sample with most virus (Ct value 29.79), 90% of the genome was covered with >4500x reads per genomic position. In the next dilution (31.57 Ct), this decreases to an average coverage of ~1000x reads per genomic position. Importantly, genome coverage at these depths is generally sufficient to assess inter/intra-host diversity in the virus population [33]. We also observe that the number of low frequency variants detected decreases as the dilution increases (S3 Table). At further dilutions (Ct values 33.45 and above), the average depth of coverage which covers 90% of the genome drops below ~100x. Although this level of coverage is generally considered to be insufficient for the estimation of viral subpopulations [33,34], consensus genomes can still be constructed and used for phylogenetic analyses.
To measure the efficacy of the bait panel in clinical samples, we first tested the protocol on clinical samples representative of each of the different serotypes of DENV. These four samples were collected from DENV-infected patients who were enrolled in the early dengue infection and outcome (EDEN) study in Singapore [35,36] and from the DengueTools study in Sri Lanka [28]. The proportion of DENV reads relative to the host genome for different samples varied from 0.10% to 90.7%. After enrichment, the proportion of DENV-specific reads was between 94% and 99.6% for all serotypes tested. [Table 1 and S3 Table]. The most successful enrichment occurred in the DENV4 sample, where only 0.10% DENV4 specific reads were present in the unenriched library. Following enrichment, the number of DENV4 specific reads was increased almost 3000-fold to 94% DENV4 specific sequences.
We then applied the bait panel to two clinical cohorts; one from Sri Lanka and another from Singapore. The first cohort was collected as a laboratory-based enhanced sentinel surveillance system in the Colombo District of Sri Lanka from 2012–2014 [28]. This collection period coincided with a severe dengue epidemic predominantly (~80%) driven by DENV1 with a smaller number (~20%) of cases due to DENV4 [28,37,38]. From this collection, we utilized our enrichment platform to obtain full or nearly full genome sequences for 143 DENV1 and 27 DENV4 isolates (S2 Fig). The second cohort of 162 samples was collected from August to September 2016 as part of a national response effort to the first outbreak of ZIKV in Singapore (S2 Fig) [27,39]. During the course of our investigation, we were also able to detect one dual infection in the Singaporean cohort where we were able to recover both DENV3 and ZIKV from the same patient (Fig 4).
To further examine the assumptions made by our method BaitMaker to generate baits, we investigated the baits’ features such as GC content, melting temperature, Gibbs free energy and sequence identity of the bait with the target genome and its effects on bait and target genome hybridization. We tested whether the baits’ features affected the baits’ pull-down efficiency, which is measured by the distribution of sequencing reads mapped to the genome around the bait. Based on the principal component analysis on the baits’ features and pull-down efficiency (S3 Fig) and effects of sequence identity on pull-down efficiency (S4 Fig), we observed that only sequence identity had an influence on pull-down efficiency and this efficiency decreases as the sequence identity between bait and target genome decreases. We next tested the effect of library size on the genome enrichment of the samples in both DENV and ZIKV cohorts (S5 Fig). We found that when the library size exceeds 300 nt, there is no further increase in coverage and this is expected as the baits were designed for library size of 300 nt. This further suggests that for larger library size, we can increase the bait design interval and thus only a smaller number of baits would be required to capture complete genomes. Thus, BaitMaker can efficiently design a set of minimal number of baits which is required to capture a complete genome.
DENV, ZIKV and CHIKV together represent the most significant arboviral threats to humans today with hundreds of millions of infections annually. These viruses are predominantly transmitted to humans by Aedes aegypti and Aedes albopictus species of mosquitoes whose global distribution has exploded in recent years, placing an estimated two-thirds of the world’s population at risk from contracting these diseases [4]. Calculating the true burden of disease for each of these viruses is complicated by not only overlapping clinical presentations but importantly by the fact that they co-circulate [11]. Given the commonality of the arthropod vectors these viruses employ for their transmission, the contribution of co-infection with these viruses to morbidity and mortality is poorly understood.
Full-genome sequencing of DENV1-4, ZIKV and CHIKV directly from clinical samples is not a routine practice in clinical, or even research, settings due to the costs associated with labor, reagents and time. Additionally, given the low success rate of current methodologies, loss of precious samples is a strong deterrent. Other groups have utilized a tiling approach wherein baits are designed to cover the entire length of a limited number of target genomes [24,29–32]. Although effective, we show here that this strategy represents an over-allocation of baits and is unnecessary to obtain full genome coverage of the target virus. Our BaitMaker program designs a minimal number of baits for viral capture that are non-overlapping and has the ability to capture the known variation in viral strains. Our approach significantly reduces the overall number of baits necessary to capture the diversity in these viruses which in turn minimizes the cost incurred and should assist in the broader implementation of this methodology.
Although we have currently only tested our methodology with Illumina sequencing, we foresee extending this methodology onto other sequencing platforms. Indeed, this methodology may have an even larger impact for the so called ‘third generation’ sequencing platforms (Pacbio and Nanopore) where average library sizes are far longer than Illumina but do not yield as many reads per run. Increasing the average library size would presumably translate to an even smaller number of baits required to capture full or near full genome coverage of the target virus as our results have shown strong positive correlation between library size and breadth of genome coverage (S5 Fig).
One particularly important benefit of utilizing a targeted enrichment approach over a Sanger sequencing or an unbiased metagenomic approach, is the number of reads specific to the viral genome are substantially increased and thus allows for assessment of intra-host genetic diversity. These data are particularly important for the analysis of low-frequency variants that may represent precursors to a change in overall pathogenicity [16,17]. As vaccines for these viruses are now in various stages of development, in-depth analysis of viral sub-populations will be critical to monitor vaccine escape mutants as they develop.
In this study, we have applied this methodology to a febrile cohort (n = 170) collected during a severe dengue outbreak in Sri Lanka and to a cohort (n = 162) collected during the 2016 Zika outbreak in Singapore. We were able to obtain full or nearly full genomes for the target viruses down to the limit of detection by qPCR. Importantly, we have combined bait panels for all four DENVs, ZIKV and CHIKV into a single assay and have detected a co-infection in a patient sample. Given the relatively small number of samples we have tested here and the overlapping clinical presentation of these viruses, the detection of a DENV/ZIKV co-infection indicates that the global prevalence of ZIKV and CHIKV could be higher than current estimates. Clinical management for DENV, ZIKV and CHIKV is largely supportive in nature with the vast majority of cases treated as outpatients and left to convalesce outside a clinical setting. Whether co-infection with these viruses is a predicator of adverse clinical outcome is largely understudied but it is a significant question that could potentially change clinical management and outcome for some of these patients [40]. In the one sample where we identified co-infection, the amount of DENV3 was much greater than ZIKV. Whether this result is due to differential viral kinetics, viral interference or temporal differences in the acquisition of each virus is an interesting question and is a subject of ongoing work.
Finally, we believe that application of the approach employed here for bait selection would potentially improve upon the large, pan-viral enrichment panels such as those recently published by Briese et al, 2017 and Wylie et al, 2015 [41,42]; with fewer baits required to achieve full genome coverage, more baits could be allocated to capture the known diversity in the targeted viral families. Increasing the amount of diversity in these panels would in turn increase the likelihood of capturing novel viruses in clinical and environmental samples.
Investigations described for the Zika samples were conducted as part of outbreak response operations by the Ministry of Health, Singapore to control the spread of Zika. Samples were taken opportunistically with verbal informed consent from subjects. Approval for the collection of dengue samples was obtained from the Ethics Review Committee, Faculty of Medicine, University of Colombo, Sri Lanka and informed consent obtained from participants in written format. Parental consent was obtained for study participants below 18 years of age. Data from both studies was anonymized prior to analysis and all methods were performed in accordance with the relevant guidelines and regulations.
For targeted-enrichment method, we designed reverse complementary DNA baits of 120 nt in length, targeting the viral genome of interest. As the hybridization takes place at 65°C, we designed baits such that they had a melting temperature greater than this. The baits in our panel had melting temperature ranging from 69 to 87°C and GC content from 31 to 66%. We developed two modes to design baits (i) Conserved mode, to design baits at the species-level conserved regions and (ii) Exhaustive baits, to design baits for both conserved regions as well as regions with strain level variations.
DENV1-4 and ZIKV were cultured in 2x105 HuH7 and Vero cells, respectively, for 48 h or until cytopathic effects were observed. CHIKV was cultured in 2x105 BHK21 cells for 20 h. After incubation, supernatant was removed and cell layers were scrapped into 250 μl of sterile PBS. RNA from serum and urine samples were extracted directly. All viral cultures and clinical samples were handled in a Class II-A2 biosafety cabinet under BSL-2 conditions according to national regulations pertaining to the handling of infectious agents.
RNA extraction was done using TRIzol Reagent (Life Technologies) according to the manufacturer’s instructions. Briefly, 250 μL of sample was added to 750 μL of TRIzol Reagent in a Phase Lock Gel Heavy 2 mL tube (5 PRIME) and incubated for 10 min at room temperature. Following incubation, 200 μL of chloroform (Sigma-Aldrich) was added and the mix was incubated for a subsequent 10 min. The sample was centrifuged for 5 min at 13,000 rpm. The aqueous phase was decanted into a new tube and 2 μL of Glycoblue Coprecipitant (ThermoFisher Scientific) and 500 μL of 2-propanol (Merck) was added to precipitate the RNA. The sample was centrifuged at 13,000 rpm for 15 min to obtain an RNA pellet. The pellet was washed with 500 μL of 75% ethanol (Merck), centrifuged for 3 min at 13,000 rpm, air-dried and resuspended in nuclease free H2O.
Quantitative PCR (qPCR) was performed according to established methodologies [13]. Briefly, we used the QuantiTect Probe RT-PCR Kit (Qiagen) reagents and the CFX96 Real-Time System (Bio-Rad) where each 25 μL PCR reaction contained 12.5 μL 2X QuantiTect PCR mastermix, 1 μL of each 10 mM primer, 0.5 μL 0.2 mM probe, 0.5 μL reverse transcriptase, 3 μL RNA template and 6.5 μL H2O. Every PCR was performed as follows: reverse transcription at 50°C for 30 min, initial PCR activation at 95°C for 5 min and 45 amplification cycles consisting of a 95°C denaturation for 10 sec and a 60°C annealing/extension for 30 sec. Sequences of primers and probes are as follows; ZIKV-F: 5'- TGG TCA TGA TAC TGC TGA TTG C -3', ZIKV-R; 5'- CCT TCC ACA AAG TCC CTA TTG C -3', ZIKV-probe5'- /56-FAM/CGG CAT ACA GCA TCA GGT GCA TAG GAG /3BHQ_1/ -3', Vero (African green monkey) GAPDH-F:5′- GGG TGT GAA CCA TGA GAA GTA T-3′, GAPDH-R; 5′- GAG TCC TTC CAC GAT ACC AAA G-3′ and GAPDH-probe: 5'- /5HEX/AC AAC AGC CTC AAG ATC GTC AGC A/3BHQ_1/ -3'. The relative amount of viral transcript to GAPDH was calculated using the 2-ΔΔCT method. Data were expressed as fold change RNA compared to the control.
Illumina libraries were constructed from total RNA using NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs) in conjunction with NEBNext Multiplex Oligos for Illumina (New England Biolabs) according to the manufacturer’s instructions with minor modifications. Briefly, 5 μL of total RNA was added to first strand synthesis buffer and random primers before incubating at 94°C for 2 min in order to generate RNA fragments larger than 500 nt. Following first strand and second strand cDNA synthesis, double-stranded cDNA was purified using Mag-Bind RxnPure Plus beads (Omega Bio-Tek) and eluted in 60 μL nuclease-free water. In order to obtain a library size between 400–600 nt, size selection of the libraries was performed using Mag-Bind RxnPure Plus beads (Omega Bio-Tek) in a two-step selection, by adding 35 μL, then subsequently 15 μL of beads to the reaction. The library was eluted in 20 μL nuclease-free water and amplified by PCR. Libraries were purified using the MinElute PCR Purification Kit (Qiagen), eluted in 25 μL nuclease-free H2O and visualized on a 1.5% agarose gel and quantified using a Bioanalyzer High Sensitivity DNA Assay (Agilent).
Targeted viral enrichment was achieved using custom designed biotinylated, 120mer xGen Lockdown baits (Integrated DNA Technologies). Prior to capture of viral sequences, 1 μL each of xGen universal blocking oligo TS-p5 and TS-p7 (Integrated DNA Technologies), matched accordingly to the library index was added to 20 μL of library DNA and 0.5 μL of 5 μg Cot-1 DNA (Invitrogen) to block binding of baits to non-viral regions of library fragments. Blocked libraries were ethanol precipitated and resuspended in 2.5 μL H2O, 3 μL Nimblegen hybridization solution and 7.5 μL Niblegen 2X hybridization buffer (Roche). Following a 10 min incubation at room temperature, resuspended libraries were denatured at 95°C for 10 min and cooled on ice before the addition of the DENV, CHIKV and ZIKV bait pool. A total amount of 3 pmol of baits were added and hybridized to the libraries for 4 h at 65°C. To capture virus specific library fragments, 100 μL magnetic M-270 streptavidin Dynabeads (Life Technologies) were added to the hybridization reaction and the mix was incubated for a further 45 min at 65°C, with shaking at 2000 rpm in a ThermoMixer C (Eppendorf). Streptavidin beads were washed to remove unbound DNA using SeqCap EZ hybridization and wash kit (Roche) according to the manufacturer’s instructions. A post-capture PCR amplification of 20 cycles with P1 and P2 primers (Illumina) was performed and the enriched library was purified using the MinElute PCR Purification Kit (Qiagen). The purified, enriched library was eluted in 25 μL nuclease-free H2O, visualized on a 1.5% agarose gel and quantified using a Bioanalyzer High Sensitivity DNA Assay (Agilent). For the complete protocol, please see S1 File.
Enriched and unenriched libraries were constructed and sequenced on an Illumina MiSeq (Duke-NUS Genome Biology Facility, Singapore) and Illumina HiSeq 4000 (Genome Institute of Singapore). FastQC [45] was used to confirm the quality of the reads generated, and Trim Galore [46] was used to trim and filter the reads with a minimum quality cutoff of 20 and a minimum read length of 35 nt. As the viral species and strain is unknown in most of the cases, it is necessary to identify the nearest species and the strain present in the sample. Therefore, Blast toolkit [47] was used to search the nearest hit in the NCBI nucleotide database for every read using the megablast option. A metagenomic analysis software MEGAN [48] was used to cluster reads at the species level to visualize. As Blast analysis is time-consuming, only a portion of the reads were used to identify the species and strain. The species cluster with the maximum number of reads assigned was selected as the initial reference strain and used to generate a consensus genome. The consensus genome was generated by using bam2cons_iter.sh script from the ViPR pipeline [49]. The bam2cons_iter.sh uses BWA [50] to perform iterative mapping of the reads to the reference genome and a consensus is generated based on the maximum frequency of a nucleotide at a given position. From the obtained consensus genome, the nearest NCBI hit is found and used as a reference genome to rerun the bam2cons_iter.sh script with default parameters. This iterative consensus genome generation approach enables generation of a full genome consensus for the virus present in the sample. For final mapping with BWA mem v0.7.5 aligner was used to map the reads to the consensus reference genome and picard tools v1.95 [51] were used to mark PCR duplicates. Base calibration and indel realignment was done by GATK v3.3 [52]. Single nucleotide variants for each sample were detected using LoFreq2 software [33], which incorporates base-call quality scores as error probabilities into its model to distinguish SNVs from the average sequencing error rate, and assigns a p-value to each position (Bonferroni-corrected p-value > 0.05). LoFreq has previously been applied to DENV datasets, and its SNV predictions on these datasets have been experimentally validated down to 0.5% allele frequency [33], hence we filtered the SNPs with a threshold of coverage (>1000) and allele frequency (>0.5%). Finally, the genome coverage graph along with the baits positions and SNP positions were plotted using Circos [53].
The Pearson product-moment correlation analysis between the mean library size and one-standard deviation of Gaussian distribution was performed in R v3.3. The mean library size of the sample was computed using Picard-tools. The average one-standard deviation of Gaussian distribution per bait (>95% identity), was calculated by fitting a Gaussian distribution to the genome coverage in a window of 480nt around the bait. Quickfold from the mfold [54] package was used to find the Gibbs free energy for DNA bait secondary structure formation at temperature 65°C, 1 mM Na and 0 mM Mg. The principle component analysis between the GC content, melting temperature, identity and mean coverage at the region where the bait hybridizes with the genome was carried out in R.
|
10.1371/journal.ppat.1005074 | A Novel Virus Causes Scale Drop Disease in Lates calcarifer | From 1992 onwards, outbreaks of a previously unknown illness have been reported in Asian seabass (Lates calcarifer) kept in maricultures in Southeast Asia. The most striking symptom of this emerging disease is the loss of scales. It was referred to as scale drop syndrome, but the etiology remained enigmatic. By using a next-generation virus discovery technique, VIDISCA-454, sequences of an unknown virus were detected in serum of diseased fish. The near complete genome sequence of the virus was determined, which shows a unique genome organization, and low levels of identity to known members of the Iridoviridae. Based on homology of a series of putatively encoded proteins, the virus is a novel member of the Megalocytivirus genus of the Iridoviridae family. The virus was isolated and propagated in cell culture, where it caused a cytopathogenic effect in infected Asian seabass kidney and brain cells. Electron microscopy revealed icosahedral virions of about 140 nm, characteristic for the Iridoviridae. In vitro cultured virus induced scale drop syndrome in Asian seabass in vivo and the virus could be reisolated from these infected fish. These findings show that the virus is the causative agent for the scale drop syndrome, as each of Koch’s postulates is fulfilled. We have named the virus Scale Drop Disease Virus. Vaccines prepared from BEI- and formalin inactivated virus, as well as from E. coli produced major capsid protein provide efficacious protection against scale drop disease.
| Asian seabass or Lates calcarifer is a large, valuable fish kept in maricultures. Scale drop syndrome is an emerging disease in this species that currently results in significant economic losses for affected farms. Mortality rates can become as high as 50%, both in juvenile and adult fish. With the increasing mariculture of L. calcarifer, it is expected that economic losses due to the syndrome will increase as well. We provide conclusive evidence that the single causative agent for the syndrome is a newly identified member of the genus Megalocytivirus that has been designated scale drop disease virus.
| Scale drop syndrome in Lates calcarifer, Asian seabass, was first reported in 1992 in Penang, Malaysia, and since then outbreaks have been seen in Indonesia and in the Strait of Malacca. The phenotypic symptoms and pathology of the syndrome were described in detail by Gibson-Kueh et al. [1]. Typically, affected fish are characterized by darkened bodies, scale loss, tail and fin erosion, pallor of gills, and sometimes exophthalmia. Furthermore, fish often show lethargic behavior and severely affected fish stop schooling, sometimes show spiral swimming and a large proportion of the fish eventually die. The pathological symptoms comprise vasculitis in all major organs, including the skin and the brain, tissue degeneration, hemorrhages and necrosis. The cumulative mortality is estimated around 40–50%. The syndrome is so far described to affect Asian seabass only, both juvenile and adult fish, and seems to follow a seasonal pattern: the south-west monsoon/inter-monsoon season starting around September may be a trigger. It is an illness with unknown etiology of which the incidence is on the rise in commercial fish farms. As Asian seabass is a large, valuable fish kept in maricultures, scale drop syndrome currently results in significant economic losses for affected farms. With the increasing aquaculture of L. calcarifer, from 11,000 tonnes in 1990 to 75,000 tonnes in 2012 [2], the syndrome is expected to occur with increased frequency. The identification of the cause of the disease, and if possible, the development of a vaccine, are therefore highly desired.
Scale drop syndrome spreads between cages, indicating that an infectious agent is involved. Initially, it was believed that scale drop syndrome was caused by infections with Tenacibaculum maritimum, but so far this or other known microorganisms could not be linked to scale drop syndrome [1]. Since antibiotic treatment does not work to prevent disease, we hypothesized that a yet unknown viral agent is involved.
Detection of an unknown virus requires specialized techniques. Nowadays, next generation sequencing platforms combined with viral purification and library preparations provide and excellent method to identify viruses of which the genome composition is unknown. The VIDISCA method (Virus discovery cDNA-AFLP) is one of the library preparation methods which has been successfully used to identify several novel viruses [3–7]. Our aim was to detect a possible viral pathogen using this approach.
In this study the VIDISCA library of sera of fish affected by scale drop syndrome was sequenced via Roche-454 next generation sequencing, enabling identification and characterization of a novel virus, which we named Scale Drop Disease Virus (SDDV). Three candidate vaccines were designed and all Koch's postulates were fulfilled.
Four serum samples of scale drop syndrome-affected fish from Singapore were analyzed with VIDISCA-454. A total of 42,918 sequence reads were obtained of which 15 showed limited identity to known viruses in GenBank, indicating that the samples may contain an unknown virus. At the nucleotide level, these sequence reads showed identity to conserved parts of the megalocytiviruses from the family Iridoviridae.
A real time qPCR was developed on a VIDISCA fragment which appeared to be part of the gene encoding the putative DNA-dependent RNA polymerase of the virus. Three out of four serum samples of the affected fish were PCR positive. Organ material from spleen, heart and kidney of diseased fish was also PCR positive. All serum, spleen, and kidney control samples from healthy fish remained negative in the qPCR. Testing of 30 more sera from early and late stage scale drop syndrome affected fish collected at a mariculture farm in Indonesia revealed another 25 positive fish, whereas none of 6 healthy control fish samples were positive (S1 Table).
The near complete genome sequence of the novel virus was obtained via genome walking. The genome length is at least 124,244 bp, consists of dsDNA, and it contains no less than 129 ORFs. A Blastn search showed that the closest relatives are among the megalocytiviruses, however, the identity was low (depending on the gene, at most 60%). Dot plot analyses confirmed that there are large differences with all known members of the Iridoviridae, including the megalocytiviruses (S1 Fig). In addition, an unusual low % G+C is present in the viral genome: only 37% whereas this value ranges between 53% and 55% for the other full length sequenced megalocytiviruses [8].
At this point it was not clear whether the novel virus was the causative agent or an innocent bystander of scale drop syndrome, but as a working hypothesis we called this virus Scale Drop Disease Virus (SDDV). In literature, a set of 26 conserved Iridoviridae genes is used for genome comparison [9] and all of these 26 are present in the near complete sequence of SDDV. The location of each gene was determined and compared to the locations in other known members of the Iridoviridae family for which the full genome has been determined. S2 Table shows that the position of the conserved genes is unique and distinctively different from the other viruses. Phylogenetic analysis based on the deduced amino acid sequences of the 26 conserved proteins revealed that the virus clusters with the megalocytiviruses, yet it does not cluster as tightly as the other members (Fig 1). Full genome sequences are however not available for all known megalocytiviruses, therefore an additional phylogenetic analysis was performed using the deduced amino acid sequences of the major capsid protein (MCP), which allowed inclusion of all (sequenced) members of the Iridoviridae family. S2 Fig shows that there are no close relatives, none of the known viruses clusters close to SDDV.
PCR-positive sera were pooled and used to inoculate seabass kidney (SK) SK21 cells. A negative control was composed of the pooled sera from three PCR-negative fish without clinical signs of scale drop syndrome. Enlarged cells were already observed two days after inoculation of the SK21 cell cultures, and after six days, a high level of cytopathogenic effect (CPE) was observed. Cultures were harvested by applying three freeze-thaw cycles, and replication of the virus was confirmed by qPCR on the harvest. Subsequently, the harvest was used for a second passage on SK21 cells using 10 DNA copies per cell as inoculum, which corresponds to a multiplicity of infection (MOI) of 0.01 TCID50/cell (determined by back calculation). CPE was clearly visible in the second passage on day four and it was complete (100%) on day five. A third passage was started using the day four harvest of the second passage, again using 10 DNA copies per cell (MOI 0.01 TCID50/cell) as inoculum. The CPE was evident in the infected cultures, defined initially by distinct morphological changes in the cells, and at later times this was coupled with an increase in detachment from the culture vessel (shown in Fig 2B and 2D). Viral replication was monitored by determining the number of viral genome copies on day 1 to 4 and the infectious titer on day 2, 3 and 4. The genome copy number increased from 2 x 106 to 3 x 1010 genome copies/mL. The virus titer (TCID50), increased from 3.2 10log TCID50/mL to 8.1 10log TCID50/mL on day 4 (Fig 2E). Similar results were obtained when the virus was inoculated on Asian seabass brain cells (SBB, established at MSD AH, S. Koumans, D. Remmers, S3 Fig). Negative control inoculation did not induce CPE and remained qPCR-negative during three subsequent passages on SK21 and SBB cells.
The possibility that CPE was caused by dilution of a toxic factor instead of a virus was excluded by conducting serial passages of the SK21 culture harvest. A total of 5 successive passages at MOI 0.01 TCID50/cell were conducted on SK21 cells in T25 flasks. Cells were harvested at the time of near-complete CPE, usually day 3–4 after inoculation. Harvests of the cultures were titrated and showed to be between 7.6–8.2 10Log TCID50/mL culture harvest. Contamination by viral or bacterial contaminants like mycoplasmas was excluded as well by using the tests as described in the European Pharmacopeia [10]. Furthermore VIDISCA-454 was performed on the culture supernatant of passage 3. Besides the novel virus, no sequences with significant identity to any virus were detected.
To confirm correlation between presence of the virus and the induction of CPE, a qPCR analysis of DNA samples isolated from CPE-positive and -negative wells in a representative plate of the titration assay was performed. In all wells that were CPE-positive a high concentration of SDDV DNA was detected (> 108 copies/mL, shown in S3 Table).
Members of the Iridoviridae have an outer protein capsid composed of capsomers, covering an inner lipid membrane bilayer that envelopes the genome. The lipid membrane adopts an icosahedral morphology that roughly follows the contour defined by the outer layer of capsomers [11]. Electron microscopy of SDDV shows virions with a diameter of about 140 nm (Fig 3). In the concentrated virus suspension two types of particles can be discerned: particles that have the outer membrane and capsid, where the core appears dark and particles that have lost the outer membrane and appear with a lighter-colored core (Fig 3A). Most particles had no or had lost the outer membrane. The icosahedral symmetry, the inner core and capsid are typical for members of the Iridoviridae. An internal membrane is not visible (Fig 3).
As Iridoviridae family members have been reported to be present as enveloped and non-enveloped viruses, we investigated if a lipid envelope was essential for infectivity of the virus in SK21 cells by applying chloroform treatment. Titrations of the virus harvest incubated with 0%, 10%, and 50% (v/v) chloroform yielded different TCID50 values. The 10log TCID50/mL of the 0% (v/v) control was 7.5, compared to 5.5 of the sample treated with 10% (v/v) chloroform and 5.4 10log TCID50/mL of the 50% (v/v) chloroform-treated sample S4 Fig). This indicates that at least a subset of virions remained infectious after chloroform treatment. The 100-fold decrease in TCID50 could be the result of loss of the enveloped subset of virions that are sensitive to chloroform, but this could also be an effect of chloroform on naked virions. A lipid envelope, if present, does not seem essential for infectivity of the virus.
To examine if the newly discovered virus was indeed the causative agent of scale drop syndrome, L. calcarifer were infected with virus culture harvest. Virus culture harvests (passage 3) were applied via different routes of infection: intraperitoneal (IP; 0.1 mL or 5.5 x 106 TCID50/fish), intramuscular (IM; 0.01 mL or 5.5 x 105 TCID50/fish), and a combination (IP 0.1 mL + IM 0.01 mL). A fourth group of fish was infected IP with 0.1 mL of a 1:10 dilution of culture harvest in virus dilution buffer (PBS with increased salt concentration (1.5% NaCL); IP 1:10) to match the infectivity dose in the IM infection. The fifth group contained control fish. Records of fish mortalities in the observation tanks between day 0 and day 28 are shown in Fig 4A (15 fish/group). Clinical symptoms characteristic of scale drop syndrome, including tail and fin erosion, broken fins, scale loss and exophthalmia, were clearly visible on day 7–14 after infection in the IP-challenged group (Fig 4C; day 10) and the IP- + IM-challenged group. In the IP-challenged fish, mortality was observed from 5 days post-infection. Cumulative mortality reached 60% for the IP group and 47% for the IP + IM group. Fish that received an IP 1:10-dose only or an IM dose showed less severe clinical symptoms and a later onset of mortality and lower cumulative mortality: 20% (IP 1:10) and 13% (IM). No mortality was observed in control fish. Pooled serum samples collected from animals at 1, 3, 7, 10 and 14 days after infection were analyzed by qPCR for the presence of viral DNA. Viral DNA was detected in all groups on day 3, 7, 10 and 14, and not in the control group (Fig 4D). The amount of viral DNA copies peaked around 10 days post-infection.
Further fulfillment of Koch’s postulates [12] was achieved by confirming the presence of the infectious agent in the experimentally infected fish. Pooled sera of day 7 and 10 of the abovementioned infections were used to inoculate SK21 cells. All sera were tested at 1:100 and 1:1000 dilution. Sera from all infected groups gave CPE on day 3 with the exception of the 1:1000 IP 1:10 dose day 10 serum, which gave CPE on day 7 (S4 Table). Control sera remained CPE negative in this assay. The presence of SDDV in the CPE positive cultures was confirmed by PCR and sequencing of the PCR products.
A vaccination-challenge study was performed to investigate whether a vaccine could protect the fish from scale drop syndrome. The optimal SDDV challenge dose for IP injection in 67 g fish was determined to define the best virus concentration to infect fish following vaccination. Groups of 15 fish were injected with 2.0 x 108 TCID50/fish, 2.0 x 107 TCID50/fish, and 2.0 x 106 TCID50/fish using a 0.1 mL IP injection. Mortality reached 100% (15/15) in the 2.0 x 107 TCID50/fish and 2.0 x 106 TCID50/fish groups, but unexpectedly not in the group that received the highest viral dose (2.0 x 108 TCID50/fish: 11/15 or 73%) at day 27 post inoculation. Based on these results a dose of 2.0 x 107 TCID50/fish was chosen to challenge the fish after vaccination.
A formalin-inactivated virus vaccine, a BEI (binary ethyleneimine)-inactivated virus vaccine, and a recombinant MCP protein produced in E.coli (recMCP) vaccine were tested. An oil-adjuvanted vaccine made with dilution buffer only was used as control. Vaccines were administered as 0.1 mL IP dose at day 0 of the experiment when the fish averaged a weight of 60 g. Fish were challenged with an IP challenge of 2.0 x 107 TCID50/fish on day 28 post vaccination, when the average weight had reached 83 g. Survival after challenge was monitored for 28 days (Fig 5). Mortality in the control group was high with only 8% survival after 28 days (2/25), whereas all three prototype vaccines provided protection with relative protection percentages of 74% (formalin-inactivated), 70% (BEI-inactivated) and 91% (recMCP). The virus concentrations of the formalin-inactivated vaccine and the BEI-inactivated vaccine differed 10-fold (Table 1 and Fig 5). To examine whether the formalin-inactivated vaccine could provide better protection at a higher dose, a formalin-inactivated vaccine was also made from the same virus stock (108 TCID50/mL) that was used to make the BEI-inactivated vaccine. The relative protection percentage of this vaccine was however equal (70%).
The presence of viral DNA was examined in sera of 5 surviving vaccinated-challenged fish on day 28 post infection, of each vaccine group, and the two surviving fish of the placebo vaccination. The two sera of the surviving placebo vaccinated fish were both positive for SDDV DNA. Also 3 of the 5 sera of the fish receiving the BEI-inactivated SDDV vaccine were positive for SDDV DNA, indicating that residual replication of the virus occurred. The sera of the formalin-inactivated and recMCP vaccinated fish were all negative for SDDV DNA (S5 Table).
The Kaplan Meier survival curve of fish vaccinated with placebo was significantly different from those of the vaccinated fish (p<0.001, Tarone-Ware test), thus all three vaccines provide protection against death caused by SDDV infection. The recMCP had the highest relative protection percentage (91%), which may suggest that this vaccine provides a better protection than inactivated virus, however statistical tests showed that the difference between the protection by recMCP and the inactivated virus vaccines was not significant. The commercial MSD vaccine Aquavac IridoV against red seabream iridovirus (RSIV) showed no cross protection against SDDV (47% survival in placebo-vaccinated and 47% survival in Aquavac IridoV-vaccinated fish).
This study describes the characterization of a virus which infects Lates calcarifer and causes scale drop syndrome. This previously unknown virus belongs to the Iridoviridae family and has only limited similarity with the known members. The pathogen was detected exclusively in sera and organs from fish with scale drop syndrome. To establish its role in scale drop syndrome, the virus was cultured and the culture harvests were used to infect Lates calcarifer, which resulted in the development of scale drop syndrome. The virus could subsequently be reisolated from the affected fish; hence all of Koch’s postulates were fulfilled. We propose to name the virus Scale Drop Disease virus (SDDV), as we provide conclusive evidence that the virus is the single causative agent for the syndrome that can now be classified as scale drop disease.
SDDV is a member of the Iridoviridae, a family of viruses that consists of five genera. The three genera that infect vertebrates (fish, amphibians and reptiles) are Ranavirus, Lymphocystivirus and Megalocytivirus. The two genera that infect invertebrates (insects, crustaceans and possibly mollusks) are Chloriridovirus and Iridovirus. Analyses of putatively encoded proteins of conserved genes of SDDV revealed most phylogenetic relatedness with the megalocytiviruses. SDDV clusters with the megalocytiviruses but forms a separate branch within this genus. The megalocytiviruses are well known pathogens in tropical fish in South East Asia and Japan. One of the most extensively investigated megalocytiviruses is RSIV. RSIV is the causative agent of a disease with high mortalities that was first found in red sea bream (Pagrus major), which is a species of major importance for the tropical fish industry [13]. Later, two other megalocytiviruses that cause systemic diseases in fish, infectious spleen and kidney necrosis virus (ISKNV) and turbot reddish body iridovirus (TRBIV) were discovered [14]. Based on sequence analysis and serological studies, it has been concluded by the ICTV that the group of ISKNV/TRBIV/RSIV are strains of the same viral species. Because the full genome of ISKNV was the first to be determined, ISKNV is the type species of this group [15]. Only threespine stickleback virus and SDDV cluster apart from ISKNV (see S2 Fig).
The genome size and genome type of SDDV are characteristic for the Iridoviridae. Their genome is a single molecule of linear, double stranded DNA between 105 and 212 kbp, which is terminally redundant and circularly permuted [9,16]. Also based on its morphology, the classification of SDDV as a member of the Iridoviridae is justified (Fig 3). Irodovirids display icosahedral symmetry and are characterized by a central DNA-protein complex, an outer proteinaceous capsid and an intermediate lipid membrane associated with polypeptides that covers and protects the genetic material. The diameter of SDDV, 140 nm, is in the range of 120–200 nm that has been reported for other the Iridoviridae family [14]. The virions can be both enveloped and non-enveloped, depending on the mode of exit from the cells, i.e. through lysis or budding. The EM pictures revealed that only some virus particles contained an envelope (Fig 3), but one should keep in mind that loss of an envelope can also be caused by the experimental procedures used for EM. Nevertheless, the remaining infectivity after chloroform treatment shows that an envelope is not essential for SDDV infectivity. The EM pictures did not reveal whether SDDV has an internal lipid membrane. Advanced cryo-electron microscopy will be necessary to elucidate if such membrane is present or absent.
An efficacious protection against SDDV would be very beneficial for the industry. Formalin-inactivated and BEI-inactivated SDDV whole virus vaccines established in this study show promising protection, although these vaccines have to be further developed. Previous studies on RSIV have shown that inactivated whole virus vaccines offer good protection against disease [17,18]. An efficacious commercial formalin-inactivated vaccine against RSIV is available from MSD (Aquavac IridoV, an oil-adjuvanted vaccine). We tested if the Aquavac IridoV vaccine provides cross-protection against SDDV, but such cross-protection could not be shown. Most likely, SDDV and RSIV do not have sufficient antigenic epitopes in common, which is not unexpected based on the fact that the genetic differences and biological mechanisms of replication differ considerably between the viruses. The RSIV vaccine from Biken [19], which is a formalin-inactivated RSIV culture supernatant of GF cells, was not tested.
In addition to the inactivated whole virus vaccine, the vaccine based on a recMCP protein produced in E.coli showed highly efficacious protection against SDDV. It has already been suggested for other members of the Iridoviridae that MCP and other viral surface proteins are candidates for vaccine development, but so far no highly efficacious subunit vaccine has been described. Caipaing et al. [20] showed that only fish vaccinated with inactivated intact RSIV virus and not fish vaccinated with protein components such as MCP survived RSIV challenge. Fu et al. [21] showed that recombinant ISKNV MCP from prokaryotic origin emulsified with ISA 763 oil at a dose of 50 μg/fish gave a relative percent survival (RPS) of 64.3%. Our vaccination challenge study with 27 μg/mL (2.7 μg/fish) SDDV recMCP from prokaryotic origin in ISA 763A VG oil provided a RPS of 91%, and thus the SDDV-MCP protein vaccine is a very promising candidate for vaccine development against SDDV. Further studies are necessary to investigate what factors determine the immune response against recombinant viral proteins, in particular MCP, in the different megalocytiviruses.
Apart from the clear significance to all stakeholders involved in Asian mariculture, we provide data that contribute to the complete picture of the pathogenicity of the Iridoviridae. This is not only valuable to all researchers that investigate this family of viruses, but also the broader community, e.g. ecologists, oceanographers etc, that may encounter unsuspected and/or emerging diseases in fish.
L. calcarifer samples from Singapore were collected at a mariculture farm in December 2010 and July 2011 from fish with typical signs of scale drop syndrome. Serum was collected from blood that was allowed to clot for 2 hours at room temperature and subsequently stored overnight at 4°C. Blood was centrifuged and serum was transferred to tubes and stored at < -20°C. Various organs from diseased fish were collected (heart, spleen, kidney). From healthy fish, two spleens, two anterior kidneys, two dorsal kidneys and two serum samples were obtained. Three groups of fish from an Indonesian farm were collected in June and November 2012: six control fish from a cage with no symptoms of scale drop syndrome, twenty-five fish from cages with early stage scale drop syndrome and five fish from a cage at the peak mortality stage of scale drop syndrome. The total list of samples is supplied in the S1 Table.
All animal procedures were carried out in Singapore, and performed in strict accordance with the specific regulations that govern animal research in Singapore, following the Guidelines set forth by the National Advisory Committee for Laboratory Animal Research (NACLAR) on the Proper Care and Use of Animals for Scientific Purposes (2004). Research on animals is regulated by the Agri-Food and Veterinary Authority of Singapore under the Animal and Birds Act, Animals and Birds (Care and Use of Animals for Scientific Purposes) Rules. The site carrying out the research is audited annually and licensed to perform animal research (License No. VR001). The MSD AH Innovation Ltd IACUC reviewed and approved the animal care and use protocol (license number EXT-EXP/05082011).
Four serum samples of affected fish and two serum samples of healthy appearing fish from Singapore were analyzed by VIDISCA-454. The VIDISCA-454 was performed as described by de Vries et al. [22]. In short, serum was centrifuged for 10 minutes at 10,000 x g and the supernatant was treated with TURBO DNase (2U/μl, Ambion). Subsequently, nucleic acids were extracted by the Boom extraction method [23]. A reverse transcription reaction with Superscript II (Invitrogen) was performed using non-ribosomal random hexamers [24]. Subsequently, second strand DNA synthesis was performed with 5 U of Klenow fragment (New England Biolabs). Double-stranded DNA was purified by phenol/chloroform extraction and ethanol precipitation and digested with Mse I restriction enzyme (New England Biolabs). Adaptors with different Multiplex Identifier sequences (MIDs) were ligated to the digested fragments of the different samples. Before PCR amplification, the fragments were purified with AMPure XP beads (Agencourt AMPure XP PCR, Beckman Coulter). Next, a 28 cycles PCR with adaptor-annealing primers was performed. The program of the PCR-reaction was: 5 min 95°C, and cycles of 1 min 95°C, 1 min 55°C, and 2 min 72°C, followed by 10 min 72°C and 10 min 4°C. After purification with AMPure XP beads, the purified DNA was quantified with the Quant-it dsDNA HS Qubit kit (Invitrogen) and diluted to 107 copies/μl. Samples were pooled and Kapa PCR (Kapa Biosystems) was performed to determine the quantity of amplifiable DNA in each pool. Subsequently, the Bioanalyser (hsDNA chip, Agencourt) was used to determine the average nucleotide length of the libraries. The pools were diluted until 106 copies/μl, titrated with beads (DNA:beads ratio of 0.5:1, 1:1, 2:1 and 4:1) and used in an emulsion PCR according to the supplier’s protocol (LIB-A SV emPCR kit). Sequencing was done on a 2 region GS FLX Titanium PicoTiterPlate (70x75) with the GS FLX Titanium XLR 70 Sequencing kit (Roche). Sequence reads were analyzed using the blastn and blastp algorithms (National Center for Biotechnology Information).
Tissue samples (spleen, kidney and heart) were homogenized to a 10% (w/v) homogenate in PBS using glass beads. DNA was isolated from homogenized tissue samples, serum and tissue culture virus harvest using the Qiagen DNeasy Blood & Tissue kit according to the manufacturer’s instructions with some adjustments: Fifty μL of tissue homogenate was digested by Proteinase K (20 μl, 600 mAU/ml, Qiagen), mixed with 130 μL solution ATL (Qiagen), and incubated for 60 minutes at room temperature. Fifty μL of serum was mixed with 20 μl Proteinase K and 150 μL PBS, or 200 μl cell culture harvests was added to 20 μL Proteinase K. The mixtures were subsequently incubated with 20 μL RNase A (20 mg/mL) left for 2 minutes at room temperature and from this step on the manufacturer’s instructions were followed.
Sequences from the putative DNA dependent RNA polymerase gene (beta subunit) were used to design primers for PCR analysis. A primer set for red sea bream iridovirus (RSIV) was used as a control (S6 Table). A quantitative PCR was set up based on a 164 bp amplicon of the putative DNA dependent RNA polymerase. The qPCR reactions were performed on a Bio-Rad CFX thermocycler and contained 1 U SuperTaq (HT Biotechnology Ltd.), 1x qPCR buffer (0.5 M KCl, 0.1 M Tris-HCl), 300 nM dNTPs (HT Biotechnology Ltd.), 200 nM forward primer, 200 nM reverse primer, 300 nM probe, 3.5 mM MgCl2 and 2 μl template DNA in a total volume of 25 μl. See S6 Table for detailed primer and probe information. Cycling conditions were 95°C for 4 min, followed by 35 cycles at 95°C for 30 sec, 50°C for 30 sec and 72°C for 30 sec. The rampspeed was set to 1.5°C/sec from 95°C to 50°C, and from 50°C to 72°C. Data were analyzed using Bio-Rad CFX Manager 2.0 software. A duplicate measurement of a dilution series of a cloned PCR product in pCR4-TOPO (Invitrogen) functioned as a standard curve. Positive or negative classification of the samples was based on the threshold cycle, as compared to the standard curve. The specificity of the qPCR was checked by gel electrophoresis of the amplified PCR product. Calibration curves with slope and y intercept were calculated by the CFX software, and PCR efficiency calculated from the slope was between 95% and 105%. The r2 of the calibration curve was >0.99. The lower detection limit of the qPCR is 50 copies/μL.
The viral sequences identified after performing VIDISCA-454 were used as template for primer design to perform gap-filling PCRs on DNA isolated from serum. Furthermore, DNA libraries digested with either Csp6I, CviAII or AseI were used to sequence fragments that overlap with sequences obtained with VIDISCA. The gap-filling PCR products and the overlapping PCR products were sequenced using BigDye terminator chemistry (BigDye Terminator v1.1 Cycle Sequencing Kit, Applied Biosystems). Sequences were analyzed using Codoncode Aligner Software (version 4.0.4). Open reading frames (ORFs) were identified via ORF finder [25]. Only ORFs larger than 300 nt were scored, with the exception of the ORF putatively encoding the transcription elongation factor TFIIS which is smaller than 300 nt. The near full length sequence is deposited in GENBANK under accession number KR139659.
Blastn (“somewhat familiar sequences”) searches were performed to identify the closest relative of SDDV, and the similarity in genome organization via dot plot analysis. The following parameters were used: match/mismatch scores: 1,-1 and gap costs: existence 2, extension 1 [26]. Nucleotide and protein sequence alignments were generated using the multiple sequence alignment tool ClustalW. Phylogenetic trees were created with MEGA5 software using the neighbor-joining method, with partial deletion in case of gaps or insertions [27]. Only DNA sequences encoding continuous ORFs were included in an alignment. A bootstrap analysis of 500 replicates was performed to provide confidences to the clustering.
The Seabass kidney SK21 cell line was established from Lates calcarifer kidney by the Schweitzer Biotech Company (Taiwan). The cell line was cloned at MSD Animal Health in order to obtain a line that supports virus replication optimally (S. Koumans, MSD Animal Health). SK21 cells were cultured in 899 mL Leibovitz's L15 medium (Life Technologies) supplemented with 100 mL (10% (v/v)) FCS (Biochrom AG) and 1 mL (1% (v/v)) of a Neomycin Polymyxin antibiotics solution (1000 x stock) at 28°C in a humidified incubator at ambient CO2 levels. For virus culture, cells were seeded at 3.0 x 104 cells/cm2 and cultured for 24 hours prior to inoculation. The monolayer had reached a cell density of 3.5–4.0 x 104 cells/cm2 at the time of infection.
Serum of early scale drop syndrome fish from Indonesia was used for virus culture. The initial inoculum consisted of a 1:10 (v/v) dilution of this pooled serum in culture medium. Culture medium was removed from the flask and the monolayer (3.5–4.0 x 104 cells/cm2) was covered with inoculum at 28°C and ambient CO2 levels for a minimum of 30 min. The inoculum was removed, fresh culture medium was added and cells were cultured until CPE was observed using an inverted light microscope (Olympus CKX41). Virus was harvested by freeze-thawing (-70°C to 4°C; once to three times), and subsequently the harvest was cleared from cell debris by centrifugation at 1000 x g for 5 minutes at 4°C. The virus was passaged by inoculating subconfluent monolayers of cells (3.0 x 104 cells/cm2), as described above, with a freeze-thawed and cleared harvest of a previous passage, which was diluted in culture medium to a concentration of 0.01 TCID50/cell.
An SK21 cell suspension (6.0 x 104 cells/mL in cold (4°C) 50% culture medium + 50% Leibovitz's L15 medium (Life Technologies)) was seeded at 100 μL/well on a 96 wells tissue culture microtiter plate. The plates were incubated for 24 hours at 28°C in a humid atmosphere at ambient CO2 levels, where cells reached a density of 3.5–4.0 x 104 cells/cm2 at the time of inoculation. Ten-fold serial dilutions of cell culture harvest or serum were prepared up to 10−9 dilution and 100 μl per well was added to the SK21 cells. Per dilution, 10 wells were inoculated and negative controls were included in each experiment. The plates were incubated at 28°C for 9–10 days and screened for CPE with an inverted light microscope. TCID50 values were determined according to the method and calculations described by Reed and Muench [28].
A passage 3 virus harvest obtained 4 days after inoculation (12 mL) was concentrated (60-fold) in a Beckmann-Coulter ultracentrifuge at 30,000 x g for 16 hours at 4°C, suspended in 200μL L15 medium (Life Technologies), and used for negative staining and cryo transmission electron microscopy (EM). For negative staining EM, 10 μl of a 1:2 mixture of virus suspension and 3% (w/v) ammonium molybdate, pH 6.8, was applied to a 100 mesh copper grid with carbon film. After 1 minute incubation excess was removed by blotting. Specimen were observed and photographed in a JEOL JEM2100 transmission electron microscope (JEOL Ltd, Japan) equipped with a Gatan US4000 camera. For electron tomography of negative stained virus particles, 10 nm gold particles were added as fiducials to the virus mixture to facilitate alignment. A series of (two times binned) images were recorded from virus particles at magnifications of 30.000 to 50.000 times by tilting the specimen from -65 to +65 with increments of 1. The images were processed using a fiducial-bead based alignment procedure and back-projection algorithm, as implemented in the IMOD reconstruction package, to convert the information in the series of tilted projection images into a 3D tomogram [29]. For cryo-EM, 4 μl of the virus suspension was applied to a holey carbon grid. The specimen was then blotted and vitrified in liquid ethane using a Vitrobot (FEI Company). Frozen specimen were observed at -180°C using a Gatan CT3500 cryoholder. Images were taken at 2–4 μm underfocus.
Chloroform treatment of cultured virus harvest was applied to destroy a potential lipid membrane. A virus harvest of 7.5 10log TCID50/mL was mixed and incubated with 0% (vol/vol), 10% (vol/vol) and 50% (vol/vol) chloroform for one hour at 4°C, and subsequently spun at 1000 x g for 10 minutes at 4°C. The water phase, and the 10−1 to 10−3 dilutions thereof, were titrated on SK21 cells as described above, and TCID50/mL was determined.
An 83 kDa fusion protein construct with an N-terminal Glutathione-S-transferase and an internal histidine-tag for immobilized metal affinity chromatography purification was generated in E. coli. The major capsid protein (MCP) encoding sequence was synthesized (Genscript) and obtained in a plasmid backbone. An EcoRI-HindIII fragment was cloned in pET41A+ (Novagen). The resulting plasmid (pET41a+ GST-his-SDDV-MCP) was transformed to E.coli Bl21star (DE3) using standard transformation protocols and subsequently cultured on plates and in liquid culture at 37°C in Animal Component-free Luria Bertani medium (LBACF) containing antibiotics. A 500 mL baffled plastic Erlenmeyer flask with 100 mL Terrific Broth-medium Kan50 was inoculated with 1mL of the freshly grown overnight culture of transformed E.coli. The culture was incubated at 37°C and 175 rpm until it reached O.D. 600 nm ≈ 0.500. At this point, IPTG was added to a final concentration of 1 mM and the culture was incubated for 2.5 hours. The culture was centrifuged at 5,000 rpm for 5 minutes at 4°C, and the pellet was stored at –20°C. The bacterial pellet was thawed on ice and resuspended in 2 mL PBS. Lysozyme (100μg), benzonase (1000U) and MgCl2 (until 10mM) were added to improve lysis and reduce viscosity. The bacteria were sonicated on ice until a homogeneous suspension was obtained, mixed gently and incubated at RT for 1.5 hours. Thirteen mL denaturing lysis buffer (50 mM TRIS-HCl, 300 mM KCl, 6 M Urea pH 8.0) was added. The lysate was incubated at 4°C for 1 hour, sonicated again, and centrifuged at 9000 x g for 10 minutes at 4°C. The supernatant was transferred to a 15 mL tube and centrifuged for 30 minutes at 9000 x g and 4°C. After passage through a 0.45 μm filter, protein was purified from the supernatant using IMAC cartridges (BioRAD laboratories cat. no. 732–4612). The default denaturing IMAC procedure was carried out on a BioRAD Profinia apparatus. The GST-his-SDDV-MCP molecules that bound to the IMAC cartridge were eluted from the column using 15 mL of denaturing elution buffer (50 mM TRIS-HCl, 300mM KCl, 6 M Urea, 250 mM imidazole, pH 8.0). Urea and imidazole were removed from the elution fraction by dialysis against 1 liter of PBS pH 7.4. The concentration of the dialyzed GST-his-SDDV-MCP solution was determined by comparison with a BSA-dilution series on an SDS-page gel (S5 Fig). The density of the signals was measured by GeneTools software from Syngene.
SK21 cells were seeded at 3.0 x 104/cm2 and incubated for 24 hours at 28°C. The overnight culture medium was removed from the flask. Subconfluent monolayers at a density of 3.5–4.0 x 104 cells/cm2 were infected with an MOI of 0.01 TCID50 per cell in a reduced volume (0.5 mL). The cells were incubated at 28°C and ambient CO2 for 30 min. The inoculum was removed, fresh culture medium was added and cells were cultured until >50% CPE was observed at day 3 after inoculation. The virus was harvested by one freeze-thaw cycle at -70°C and 4°C, and subsequently cleared from cell debris by centrifugation at 1000 x g for 5 minutes at 4°C. The harvest was stored at -70°C until infection.
The undiluted harvest was used for infection of fish (0.1 mL/fish of undiluted SDDV: 5.5 x 106 TCID50 per fish for intraperitoneal (IP) injection; 0.01 mL/fish of undiluted SDDV: 5.5 x 105 TCID50 per fish for intramuscular (IM) injection). A 1:10 dilution of culture harvest in dilution buffer (PBS + 1.5% NaCl) was used for IP infection of 5.5 x 105 TCID50 per fish to match the IM infection (group 1–4). Control fish (group 5) were injected with dilution buffer only. Fish were infected at an average weight of 21 g.
A total of 460 fish were available: four groups of 95 fish were kept in four separate tanks for infection, and 80 fish in a fifth tank served as controls. Tanks were filled with sea water (30 ppt) of 28°C ± 2°C. Fish were starved for at least 36 hours prior to IP or IM infection to empty the gastro-intestinal tract in order to reduce the risk of damage to internal organs during injection. Immediately before the infection, 20 fish from each group were weighed together to obtain the average body weight for each group. All fish were anaesthetized using AQUI-S (AQUI-S, New Zealand) prior to the inoculation procedure. Fish were fed ad libitum from the day after infection.
Infected fish (groups 1–4) were kept in four separate 250L tanks. A vertical net was installed in each 250L tank to create partitions of 1/3 and 2/3 of the tank. The 1/3 partition held 15 fish for mortality observation. The other 2/3 (80 fish) were used for harvesting of sera and organs over the time course of the experiment. Uninfected control fish (group 5) were housed in one 250L partition of a separate 500L tank to align water temperature with the infected fish. From each of the 5 groups, sera of 15 fish were sampled at time point day 1 (dissemination control), 3, 7, 10 and 14 post-infection (total 75 fish). Excess fish (5 for each group, if without mortalities) were culled at day 14 for collection of serum. The 15 fish in the observation tank were kept until day 21 to evaluate mortality of each infection method. At each sampling time point, the serum was pooled by group.
Pooled sera of experimentally infected fish (see above), harvested at day 7 and day 10 after infection, were screened for presence of infectious virus. Sera were diluted 1:100 (v/v) and 1:1000 (v/v) in culture medium for inoculation of SK21 cells, which was carried out as described above. If no CPE occurred in the first passage of the virus, a second passage was performed.
Virus for vaccine production was cultured as described above. Cultures were harvested with one freeze-thaw cycle and stored at -70°C until inactivation. Formalin inactivation was performed by diluting formalin (37% (w/w) formaldehyde) 10 times to 3.7% (w/w) by mixing with water. This diluted stock was diluted another 100 times with cool (4°C) virus culture harvest to a final concentration of 0.037% (w/w) formaldehyde. The mixture was continuously stirred during an inactivation period of 10 days at 4°C. Binary ethyleneimine (BEI) inactivation was performed by activating 1.09 M bromoethylamine hydrobromide (BEA) with 1.91 M NaOH 1:1 (v/v) and subsequent incubation of 1 hour at 37°C to achieve a BEI concentration of 0.55 M. The culture harvest was inactivated at a final concentration of 0.01 M BEI. The pH of the mixtures was checked 2 and 21 hours after BEI addition and adjusted to be within the range of 7.4–7.65 by adding 1.0 M NaOH or 1.0 M HCl. The total Incubation time was 45 hours at 37°C. BEI was neutralized by adding sodiumthiosulphate to the mixture, and the pH was adjusted to be within the range of 7.4–7.65 as described above. The formalin- and BEI-inactivated culture harvests were titrated on SK21 cells to confirm successful inactivation of the virus. Three serial passages of the inactivated material confirmed absence of live virus in the inactivated harvest. The inactivated virus preparations were stored at 4°C until vaccine formulation.
Vaccines were formulated in MONTANIDE ISA 763A VG oil (Seppic), a water-in-oil emulsion based on non-mineral oils. The preparation of the emulsion was performed by mixing the water phase into the oil phase at 1000 rpm (mixing velocity). The water phase consisted of inactivated virus harvest in culture medium, or purified recombinant protein in PBS. In Table 1 an overview of vaccines and antigen concentrations after formulation is provided.
Five groups of 25 Asian seabass (60 g) were randomly assigned to treatment groups (vaccines see Table 1). Fish were vaccinated with 0.1 mL of the prototype vaccines by intraperitoneal injection on day 0. Negative control fish were injected with 0.1 mL of placebo vaccine (vaccine dilution buffer PBS + 1.5% NaCl, formulated in ISA 763A VG oil as described above). Fish were starved for at least 36 hours prior to the vaccination. Immediately before the vaccination, fish from each group were weighed together to obtain the average body weight for each group. All fish were anaesthetized using AQUI-S (AQUI-S, New Zealand) prior to the vaccination procedure. Fish were fed ad libitum from the day after vaccination. Vaccinated fish were kept in 250 L partitions of 500 L tanks that were created by installing a vertical net in the tank. On day 28 post vaccination, all fish were challenged. The challenge dose was 2.0 x 107 TCID50/fish and the fish were subsequently kept in 125 L partitions of 250 L tanks that were created by installing a vertical net in the tank. Mortalities were recorded daily up to 28 days post challenge.
Kaplan Meier survival curves were constructed using SPSS v22 (IBM). Analysis of the similarity of the curves was performed using the Tarone-Ware test.
The relative protection percentage (RPS) in the vaccination-challenge study was calculated as follows. The ‘percentage protected control’ was calculated as the number of survivors in the control group, divided by the total number of control fish, multiplied by 100% (% protected control = [# survivors control] / [total # control fish] * 100% = x %). The ‘percentage protected in the vaccine group’ was calculated as the number of survivors in the vaccine group, divided by the total number of vaccinated fish, multiplied by 100% (% protected vaccine group = [# survivors vaccine] / [total # vaccinated fish] * 100% = y %). The RPS is formulated as RPS = 1-(% protection test / % protected control) or in a mathematical formulation RPS = 1-(x/y)* 100%.
A representative virus has been deposited with the Collection Nationale de Cultures de Microorganisms (CNCM), Institut Pasteur, 25 Rue du Docteur Roux, F-75724 Paris Cedex 15, France, under accession number CNCM I-4754)
|
10.1371/journal.ppat.1000138 | Superior Immunogenicity of Inactivated Whole Virus H5N1 Influenza Vaccine is Primarily Controlled by Toll-like Receptor Signalling | In the case of an influenza pandemic, the current global influenza vaccine production capacity will be unable to meet the demand for billions of vaccine doses. The ongoing threat of an H5N1 pandemic therefore urges the development of highly immunogenic, dose-sparing vaccine formulations. In unprimed individuals, inactivated whole virus (WIV) vaccines are more immunogenic and induce protective antibody responses at a lower antigen dose than other formulations like split virus (SV) or subunit (SU) vaccines. The reason for this discrepancy in immunogenicity is a long-standing enigma. Here, we show that stimulation of Toll-like receptors (TLRs) of the innate immune system, in particular stimulation of TLR7, by H5N1 WIV vaccine is the prime determinant of the greater magnitude and Th1 polarization of the WIV-induced immune response, as compared to SV- or SU-induced responses. This TLR dependency largely explains the relative loss of immunogenicity in SV and SU vaccines. The natural pathogen-associated molecular pattern (PAMP) recognized by TLR7 is viral genomic ssRNA. Processing of whole virus particles into SV or SU vaccines destroys the integrity of the viral particle and leaves the viral RNA prone to degradation or involves its active removal. Our results show for a classic vaccine that the acquired immune response evoked by vaccination can be enhanced and steered by the innate immune system, which is triggered by interaction of an intrinsic vaccine component with a pattern recognition receptor (PRR). The insights presented here may be used to further improve the immune-stimulatory and dose-sparing properties of classic influenza vaccine formulations such as WIV, and will facilitate the development of new, even more powerful vaccines to face the next influenza pandemic.
| The rise and spread of the highly pathogenic avian H5N1 influenza virus has seriously increased the risk of a new influenza pandemic. However, the number of vaccine doses that can be produced with today's production capacity will fall short of the demand in times of a pandemic. Use of inactivated whole virus (WIV) vaccines, which are more immunogenic than split virus or subunit vaccines in an unprimed population, could contribute to a dose-sparing strategy. Yet, the mechanisms underlying the superior immunogenicity of WIV vaccine formulations are unknown. Here, we demonstrate that the viral RNA present in inactivated virus particles is crucial for the improved immunogenic properties of WIV in mice. By triggering Toll-like receptor 7 (TLR7), the viral RNA activates innate immune mechanisms that augment and determine subsequent adaptive responses. Efficient TLR7 signalling is lost in split virus and subunit vaccines with the processing steps that lead to disruption of the integrity of the virus particle and exclusion of the RNA. Our results prove for the first time to our knowledge that the immune-potentiating mechanism of a classic vaccine is based on activation of the innate immune system by one of its structural components. These findings may reflect a general principle for viral vaccines and provide a rational basis for further improvement of influenza vaccines, which are urgently needed in the face of the current H5N1 pandemic threat.
| The first cases of human infection with highly pathogenic avian influenza (HPAI) H5N1 virus occurred in 1997 during an outbreak in Hong Kong [1]. Since then HPAI H5N1 has spread across Asia, Europe, Africa and the Pacific, and has caused a cumulative number of 338 laboratory confirmed human cases of infection, with a fatality rate of >60% [2]. Although no sustained human to human transmission has been observed yet, the threat of an imminent H5N1 pandemic requires maximum preparedness [3]. Vaccination is considered the cornerstone of protection against epidemic and pandemic influenza. However, an anticipated scarcity of the antigenic vaccine components and a narrowed time window between vaccine production and deployment puts special constraints on the vaccine formulation to be used in a pandemic situation [4],[5]. Consequently, pandemic vaccine formulations should ideally be dose sparing and uncomplicated to produce [6],[7].
Whole inactivated virus (WIV) vaccines consisting of formalin-inactivated whole virus particles were the first registered influenza vaccines licensed in 1945 in the United States [8]. However, the use of this vaccine formulation caused a relatively high incidence of adverse events, including local reactions at the site of injection and febrile illness, particularly among children [9],[10]. In the 1960 and 1970s, WIV vaccines were therefore largely replaced by less reactogenic split virus (SV) and subunit (SU) formulations [8]. SV and SU vaccines contain detergent- and/or ether-disrupted (split) virus particles or purified viral haemagglutinin (HA) and neuraminidase (NA) proteins, respectively. Apparently, disruption of whole inactivated influenza virus particles diminishes the reactogenicity of the vaccines.
In primed individuals, unadjuvanted WIV, SV, and SU vaccines in general induce similar immune responses in terms of haemagglutination inhibition (HI) titres (for a meta-analysis over 24 studies see [11]). However, in individuals that have not been exposed to the vaccine antigens before, WIV vaccines are more immunogenic than SV and SU vaccines [9],[11],[12]. Similarly, in naïve animals immunization with WIV raises stronger immune responses than immunization with SV or SU [13]–[15], especially after a single administration. In the case of an H5N1 pandemic, the majority of the population is expected to be immunologically naïve to the H5N1 subtype. In this scenario, use of WIV as basis for an optimized vaccine may be of advantage, for its immunogenic superiority seems to rely on the ability to activate unique mechanisms in the priming event of the immune response.
Thus, WIV seems to harbour an intrinsic immune-potentiating component that is lost during processing of inactivated virus particles to SV and SU vaccine formulations. In earlier experiments, we and others observed that immunization of mice with WIV vaccine results in a Th1-skewed immune response and strong antibody induction with high levels of IgG2a antibodies [14]–[16]. This response type was found irrespective of the murine genetic background or subtype of virus (either H1N1 or H3N2) and conferred protective immunity against challenge with homologous virus [15],[16]. By contrast, immunization with SU vaccine yielded responses of a Th2 phenotype with lower antibody levels mainly consisting of the IgG1 subtype, which did not lead to protection. “Empty” reconstituted viral envelopes (virosomes) resembling intact virus particles but devoid of the viral nucleocapsid elicited responses similar to those after vaccination with SU formulations [15]. This identifies the viral nucleocapsid which contains the viral genomic ssRNA as the immune-potentiating component of WIV.
In the past decade, it has become increasingly clear that the acquired immune response to microbial infection is regulated through recognition of pathogen-associated molecular patterns (PAMPs) by Toll-like receptors (TLRs) and other pattern recognition receptors of the innate immune system [17]–[20]. However, the importance of TLR signalling in immune responses to vaccines remains largely unclear. A recent study showed that TLR signalling is not important for the antibody-enhancing effect of classical vaccine adjuvants such as Complete Freund's adjuvant (CFA) [21]. Since CFA contains dried mycobacteria, and therefore mycobacterial PAMPs [17], this observation casts doubt on the importance of PAMPs and TLRs in augmenting immune responses to vaccination. Influenza viral genomic ssRNA is a natural PAMP recognized by TLR7 [22]. Here, we investigate whether PAMP recognition by TLRs, in particular recognition of viral ssRNA by TLR7, is responsible for the superior response to WIV vaccines compared to SV and SU influenza vaccine formulations.
To analyze the role of ssRNA and other PAMPs in the response to influenza vaccines in detail, we immunized wild-type C57BL/6 mice, TLR7 knock-out mice, and MyD88/TRIF double knock-out mice with different vaccine formulations. MyD88 (myeloid differentiation factor 88) is an adaptor molecule which functions downstream of all known TLRs and IL1R family members with the exception of TLR3, which instead recruits a MyD88-related adapter molecule, TRIF (TIR domain-containing adaptor protein inducing interferon β) [17]. Consequently, a deficiency of both MyD88 and TRIF excludes signalling by all TLRs. Mice were immunized intramuscularly with β-propiolactone-inactivated H5N1 (NIBRG-14) WIV, SV, or SU vaccine. Quantitative PCR using primers specific for segment 7 of the viral genome revealed that WIV contained per vaccine dose at least 5×108 copies of viral RNA, the natural ligand of TLR7. In SU or SV vaccine the amount of RNA was 500 and 5,000 times lower than in WIV, respectively. Four weeks after immunization, serum and spleen cells were collected for evaluation of humoral and cellular immune responses.
Serum HI titres in WIV-immunized TLR7−/− mice and MyD88−/−/TRIF−/− mice were found to be significantly lower than in WIV-immunized wild-type mice (Figure 1A; p = 0.021 and p = 0.001, respectively). Although sera from TLR7−/− mice immunized with WIV showed a higher geometric mean titre (GMT) than sera from WIV-immunized MyD88−/−/TRIF−/− mice, this difference was not significant (p = 0.053). Most of the HI titres of SV- and SU-immunized wild-type mice were below detection level, precluding evaluation of the effect of the knock-out mutations on the HI responses to these vaccines.
Similar to the HI titres, virus neutralization (VN) titres of pooled serum samples from mice immunized with WIV were lower in the knock-out groups than in the wild-type group (Table 1). These results clearly show that TLR signalling is critically involved in the response to WIV immunization. Yet, in the knock-out groups, VN titres obtained after immunization with WIV were still modestly higher than those obtained after vaccination of wild-type mice with the other vaccines. This points to TLR-independent pathways contributing to the superior antibody response to WIV vaccine.
Serum titres of H5N1-specific IgG were determined by ELISA. In accordance with the HI and VN results, IgG titres were significantly decreased in WIV-immunized TLR7−/− and MyD88−/−/TRIF−/− mice compared to wild-type mice (Figure 1B; p = 0.010 and p = 0.001, respectively). However, like the VN titres, the IgG titres in the WIV-immunized mutant mice were still significantly higher than those induced by SV (TLR7−/−: p = 0.001; MyD88−/−/TRIF−/−: p = 0.005) or SU (TLR7−/−: p = 0.005; MyD88−/−/TRIF−/−: p = 0.021) immunization again indicating involvement of TLR-independent pathways. The relative contributions of TLR-dependent and -independent mechanisms to the superior IgG response to WIV can be estimated by comparing the difference in geometric mean titre (GMT) between WIV-immunized wild-type and MyD88/TRIF-deficient mice with the difference between WIV-immunized wild-type mice and SV- or SU-immunized wild-type mice. Using this procedure the TLR-dependent contribution was calculated to be 73% and 83% for WIV versus SV and WIV vs SU, respectively (for calculation, see Text S1). The IgG responses to SV and SU vaccine in both TLR7−/− or MyD88−/−/TRIF−/− mice did not differ from those in wild-type mice, except for the IgG response to SU in TLR7−/− mice, which was slightly but significantly decreased (p = 0.038; Figure 1B). Together with the HI and VN results, these findings demonstrate that the superior antibody response to WIV is predominantly regulated by TLRs, TLR7 in particular, while TLRs do not seem to play a prominent role in SV and SU antibody responses.
We next investigated the role of TLRs in the Th1 polarization of the response characteristically found after WIV vaccination. We first assessed numbers of IFNγ- and IL4- producing T cells (Th1 and Th2 cells, respectively) in a cytokine-specific Elispot assay, after re-stimulation of spleen cells from immunized mice with H5N1 SU vaccine. Numbers of Th1 cells were significantly decreased in WIV-immunized knock-out mice compared to wild-type mice (p = 0.003 and p = 0.010 for TLR7−/− and MyD88−/−/TRIF−/− mice, respectively), and matched those found in SV- and SU-immunized wild-type mice (Figure 2). No difference was found between TLR7−/− and MyD88−/−/TRIF−/− mice. Numbers of influenza-specific IL4-producing cells were extremely low in all animals for all vaccine formulations without significant differences between knock-out and wild-type mice (not shown). These data indicate that stimulation of TLR7 by ssRNA is the predominant determinant of the strong Th1-type cellular response induced by WIV.
We further determined the subtype profiles of H5N1-specific serum IgG by ELISA (Figure 3). IFNγ is known to stimulate production of IgG2a subtype antibodies by activated B cells, while IL4 stimulates IgG1 secretion [23]. In C57BL/6 mice, however, the IgG2c subtype is produced instead of IgG2a [24],[25]. Hence, a predominance of IgG2c or IgG1 is indicative of a Th1- or Th2-type response, respectively. WIV immunization of TLR7−/− mice as well as MyD88−/−/TRIF−/− mice resulted in significantly reduced IgG2c levels as compared to wild-type mice (Figure 3; p = 0.001 for both types of knock-out mice), supporting a role for TLR7 in Th1 polarization. IgG1 was increased in WIV-immunized TLR7−/− mice (P = 0.050), adding to the preponderance towards a Th2-type response to WIV in these mice. The average of ratios of serum IgG2c and IgG1 concentrations (determined with appropriate IgG subtype protein standards) was 17.82 (SD 8.44) for the wild-type mice immunized with WIV, compared to 0.53 (SD 0.41) for TLR7−/− mice immunized with WIV. SV and SU vaccines induced predominantly IgG1 and low levels of IgG2c, consistent with a Th2-type response (Figure 3). For reasons unknown, SU vaccine induced lower IgG1 titres in both types of knock-out mice compared to the wild-type mice (TLR7−/−: p = 0.050; MyD88−/−/TRIF−/−: p = 0.014). Whether the presence of some residual RNA in SU vaccine might play a role remains to be shown.
The response characteristics of the different H5N1 vaccines in wild-type mice were well in line with those previously found for other influenza subtypes [15],[16]. This consistency is supportive of a general mechanism underlying the differences in responses to WIV, SV and SU vaccine, which operates irrespective of the virus subtype used to vaccinate.
The above results demonstrate that TLR signalling plays an important role in the magnitude and Th1 skewing of the response to WIV influenza vaccines. Yet, in TLR-ko mice, WIV remained more immunogenic than SV and SU vaccines, inducing significantly higher titres of total IgG (Figure 1B) and Th1-type antibody subtypes (IgG2b, IgG2c, IgG3; Figure 3; p<0.05 for all comparisons). Thus, next to TLR-dependent mechanisms, a (minor) TLR-independent factor seems to contribute to the superior magnitude and Th1-skewing of the immune response to WIV. Type I interferons, including IFNα, have been shown to stimulate antibody responses and isotype switching to IgG2a when added to influenza subunit vaccine or other protein antigens [26],[27], even without the need for additional TLR stimuli. We have previously shown for an H3N2 influenza virus strain that, unlike SU vaccine, WIV vaccine efficiently induced interferon α (IFNα) production in plasmacytoid dendritic cells (pDCs) in vitro [15]. We therefore evaluated the induction of IFNα by the H5N1 influenza vaccine formulations used in this study and its TLR7 dependency in vitro. In pDCs of wild-type mice cultured from bone marrow cells (Figure 4A, black bars) or enriched from splenocytes (Figure 4B, black bars) WIV but not SV or SU induced IFNα production. In bone marrow-derived pDCs from TLR7−/− mice, IFNα production upon incubation with WIV was strongly decreased as compared to wild-type DCs (Figure 4A), confirming the results of others [22]. However, spleen-derived pDCs from TLR7−/− mice exposed to WIV produced similar amounts of IFNα as compared to pDCs from wt mice (Figure 4B). Thus, while in pDCs cultured from bone marrow induction of IFNα production by WIV is strictly dependent on TLR7, in pDCs enriched directly from spleen cells it is independent of TLR7. This implies that bone marrow pDCs and spleen pDCs are not completely identical. In line with this notion, bone marrow pDCs and spleen pDCs were earlier found to respond differently to HSV virus infection with respect to the TLR9 dependency of the IFNα response [28]. Our results show that WIV is indeed able to induce IFNα in a TLR7-independent way. This may also be the case in the in vivo situation, where in accordance with its well-described adjuvant functions IFNα may lead to the production of Th1 type antibodies in TLR-deficient mice [26].
Possible TLR-independent pathways activated by WIV may involve the retinoic acid-inducible gene (RIG-I) [29]–[32]. RIG-I is a cytoplasmic RNA-helicase that recognizes influenza virus by binding viral ssRNA bearing 5′-triphosphates which leads to IFNα production [33],[34]. The inactivated virus particles in WIV vaccine retained their membrane-fusion property (Text S2) and part of the viral genomes could therefore have entered the target cell cytoplasm to be sensed by RIG-I.
Taken together our observations show that the superior immune response to WIV, relative to that to SV or SU vaccines, is driven primarily by TLR-dependent mechanisms. Herein the presence of the viral RNA in the vaccine seems to play a crucial role. In contrast to SV and SU vaccines WIV contains substantial amounts of viral RNA. Removal of ssRNA from WIV by detergent solubilization and ultracentrifugation followed by reconstitution of the viral membrane envelopes to virosomes abolishes the capacity of the vaccine to induce production of IFNα by pDCs in vitro (Text S3 and Figure S1A) and type 1 immune responses in vivo [15]. On the other hand, ssRNA purified from WIV and condensed with polyethylenimine (PEI) did induce IFNα production in vitro (Text S3 and Figure S1B). Obviously, exposure of the viral RNA to β-propiolactone in the course of virus inactivation leaves the RNA intact to trigger TLR7-mediated signaling pathways (Figure 4), which translates into a strong and Th1-skewed antibody response to WIV in wild-type mice. In addition, the viral RNA may contribute to the TLR-independent part of the response to WIV since TLR7-independent production of IFNα could only be induced in pDCs by WIV and not by formulations (SV, SU, or reconstituted viral envelopes) which lack viral RNA (Figure 4B) [15]. These lines of evidence point to the ssRNA in WIV as the key component that enhances and steers the adaptive immune response by involvement of innate immune mechanisms.
IFNα induction in pDCs clearly discriminates WIV from SV and SU vaccines but seems to occur independent of TLR7. The fact that the immune response to WIV is predominantly dependent on TLR7 then suggests that other TLR7-mediated mechanisms, possibly involving conventional DCs and B cells, critically contribute to the immune reaction. Recently, an in vitro study on B cells showed that TLR7 stimulation or CD40-CD40L binding by itself triggers IgG1 antibody production, but when simultaneously present induce proliferation and a switch to IgG2a production [25]. Additional stimulation of IFNα/β receptors on the same cells further drives the production of IgG2a at the expense of IgG1 antibodies [25]. Although this model might represent an over-simplification of the in vivo situation, it is in line with our data. The different scenarios encountered upon immunization of wild-type and mutant mice with WIV, SV, or SU are summarized in Table 2. WIV provides the ssRNA for direct triggering of TLR7 in B cells as well as the CD40 ligand for CD40 stimulation on B cells through strong T helper cell induction, which was shown also to depend on TLR7 signalling. Together with IFNα produced by TLR7-mediated and/or TLR7-independent mechanisms, these signals will lead to the enhanced and strongly polarized Th1-type antibody responses characteristic for WIV. In the absence of TLR7, WIV-induced IFNα can still stimulate moderate production of Th1 type antibodies and increase the total IgG. In contrast, SV and SU vaccines are poor inducers of T helper cells and IFNα, and cannot stimulate B cells directly via TLR7. Consequently, SV and SU vaccines induce lower and more Th2-polarized antibody responses.
Our data provide mechanisms which explain the superiority of WIV vaccine to prime HA-specific immune responses in mice. Whether similar mechanisms are operational in humans and contribute to the stronger immunogenicity of WIV compared to SV or SU in unprimed individuals remains to be elucidated. Despite the favourable immunogenic properties of WIV, recent clinical trials performed in the context of pandemic vaccine development show that even with WIV at least two immunizations with a substantial amount of antigen (15–30 µg) and/or the addition of adjuvants will probably be required to achieve immune responses that comply with the CPMP criteria. If TLRs are involved in the priming of humans with WIV, their role during recall responses may be less critical, given the fact that in general WIV, SU, and SV induce similar HI titres in primed populations [11]. Use of WIV derived from wild-type virus instead of recombinant vaccine strains resulted in good antibody titres even without the addition of adjuvants and might thus be an option to obtain satisfying immune responses [35]. Evaluation of adjuvants in combination with WIV in clinical trials is so far restricted to aluminium salts. However, where adjuvanted and non-adjuvanted WIV were compared side-by-side, effects of this Th2 adjuvant on vaccine efficacy were absent, poor, or inconsistent [36]. So, better adjuvants have to be found that work synergistically with WIV in order to exploit the full potential of intact inactivated virus particles as vaccines.
In conclusion, our data reveal, for the first time to our knowledge, that TLRs play an eminent role in the immune responses to a classic influenza vaccine. Of the three influenza vaccine formulations studied here, only WIV efficiently triggered TLR7-mediated mechanisms leading to superior immune responses. Processing of inactivated whole virus particles into SV or SU eliminates the immuno-potentiating effect of the viral ssRNA, the primary PAMP in WIV vaccine, and results in a loss of quantity and shift in the quality of the immune response. Thus, TLR-dependent mechanisms appear to form the basis for WIV's antigen-sparing quality and hence its recognized strong potential as a pandemic vaccine candidate [7],[12]. Optimizing TLR7-signalling by rational vaccine design may produce even more potent vaccines, which are urgently needed in the face of the current influenza pandemic threat.
H5N1 virus (NIBRG-14, a 2∶6 recombinant of A/Vietnam/1194/2004 [H5N1] and A/PR/8/34 [H1N1] virus produced by reverse genetics technology) was provided by the National Institute for Biological Standards and Controls (NIBSC; Potters Bar, UK), propagated on embryonated chicken eggs, inactivated with 0.1% β-propiolactone to obtain WIV, and processed into split virus vaccine or subunit vaccine according to standard procedures [37],[38]. The haemagglutinin protein concentration in the vaccines was determined by single radial immunodiffusion (SRID) [39]. Endotoxin levels in all vaccines met the requirements of the European Pharmacopoeia standard. (If, nevertheless, contamination of endotoxin [signalling via TLR4] would have played an important role we should have observed substantial differences in the response between TLR7-deficient mice [capable of signalling via TLR4)]and MyD88/TRIF-deficient mice [deficient in all TLR-derived signalling]. However, such differences were not found for any of the vaccines.) CpG DNA (ODN D19) was purchased from Eurogentec (Seraing, Belgium).
For immunization experiments, C57BL/6, TLR7−/− and MyD88−/−/TRIF−/− mice (generated from MyD88−/− mice [40] and TRIF−/− mice [41]) were bred at the University of Massachusetts Medical School (Worcester, MA). For in vitro studies, 10- to 12-week-old female C57BL/6 mice were purchased from Harlan Netherlands B.V. (Zeist, The Netherlands), and TLR7−/− mice (a gift from S. Akira and C. Reis e Sousa) were bred at the University Medical Center Groningen. All experiments were conducted with approval of the local Institutional Animal Care and Use Committees. Mouse groups were matched for sex and age. Groups (n = 6–8) of C57BL/6, TLR7−/−, and MyD88−/−/TRIF−/− mice were intramuscularly injected with 50 µl of PBS in each calf muscle containing a total of 5 µg haemagglutinin protein per mouse of either WIV, SV, or SU vaccine formulation or no vaccine as a control. At 28 days after immunization, sera and spleens were collected for evaluation.
Relative viral RNA content of the different vaccines was determined using a two-step real-time RT-PCR assay amplifying a 193-bp fragment within the M1 gene of influenza A viruses. For this purpose RNA was extracted from WIV, SV, or SU (5 µg HA) with the QIAamp viral RNA Mini Kit (QIAGEN, Venlo, The Netherlands), cDNA synthesis was performed on 5 µl of viral RNA (one-tenth of the final elution volume) using the Verso cDNA kit from ABgene (Westburg, Leusden, The Netherlands), and 1 µM UNI12 primer (5′-AGCAAAAGCAGG-3′, corresponding to viral noncoding nucleotides 1 to 12 [42]). Real-time PCR was performed with 200 nM M1-FOR primer (5′-CCTGGTATGTGCAACCTGTG-3′) and M1-REV primer (5′-AGCCTGACTAGCAACCTCCA-3′); purchased from Eurogentec, and the Absolute QPCR SYBR Green Mix (ABgene). Amplification was performed on a StepOne apparatus (Applied Biosystems), and consisted of 15 min initial activation at 95°C, followed by 40 thermal cycles of 15 sec at 95°C and 60 sec at 60°C. In each experiment, a standard curve (R2>0.99 within the range of 1×102 to 1×109 copies per reaction) was drawn to convert the respective cycle threshold (Ct) values into the number of viral genome copies. This standard consisted of a pCR2.1-TOPO plasmid construct in which was cloned a 473-bp sequence of influenza A/Puerto Rico/8/34 segment 7.
The HI assay was performed as described before [15]. Briefly, heat-inactivated mouse serum was absorbed to 3 volumes 25% kaolin/PBS (Sigma-Aldrich, Inc., St. Louis, MO), 20 min at room temperature (RT). After centrifugation, 50 µl of supernatant was serially diluted two-fold in a round-bottom microtitre plate (Costar, Corning Inc., Corning, NY), in duplicate. Subsequently, 50 µl PBS was added containing 2 HAU of H5N1 (NIBRG-14) virus and incubated for 40 min at RT. We used 2 HAU of virus instead of the standard 4 HAU to increase the sensitivity of the assay. Finally, 50 µl of 1% guinea pig erythrocytes (Harlan) in PBS was added to each well and HI titres were determined after 2 h incubation at room temperature. HI titres are given as the reciprocal of the highest serum dilution producing complete inhibition of haemagglutination.
The levels of virus-neutralizing (VN) serum antibodies were determined with a VN assay [15],[43]. The VN titre was defined as the reciprocal of the highest serum dilution capable of inhibiting 200 TCID50 of H5N1 vaccine strain virus (NIBRG-14) from infecting Madin-Darby canine kidney cell monolayers in a microtiter plate. Infection was measured by an ELISA on intracellularly produced viral NP protein. Inhibition of infection by simultaneous incubation with mouse serum was established if the ELISA absorbance value (A492) measured was below the cut-off value, determined by the equation: [(average A492 of the positive controls (infected cells) minus average A492 of the negative controls (non infected cells)) divided by 2] plus the average A492 of the negative controls. Serum samples were tested in quadruplicate.
Microtitre plates (Greiner, Alphen a/d Rijn, The Netherlands) were coated with 0.2 µg influenza H5N1 (NIBRG-14) subunit vaccine per well in 100 µl coating buffer, overnight. After blocking with 2% milk in coating buffer for 45 min, 100 µl of two-fold serial dilutions of serum samples in 0.05%Tween 20/PBS (PBS/T) were applied to the wells and incubated for 1.5 h, in duplicate. Subsequently, 100 µl of horseradish peroxidase-conjugated goat anti-mouse IgG-isotype antibody (Southern Biotech, Birmingham, Alabama) was applied for 1 h. All incubations were performed at 37°C. Staining was performed using o-phenylene-diamine (OPD) (Eastman Kodak Company) and absorbance was read at 492 nm (A492) with an ELISA reader (Bio-tek Instruments, Inc.). After subtraction of background levels, serum dilutions yielding an OD of 0.2 were calculated using linear regression, of which the reciprocal of the average of the duplicates represents the titre.
This assay was performed as described previously [15]. In short, erythrocyte-depleted splenocytes were seeded at a concentration of 5×105 cells in 100 µl medium per well, in triplicate in a microtitre plate (Greiner), which was pre-coated with anti-IFNγ or anti-IL4 capture antibody (Pharmingen, San Diego, CA) and blocked with 4% BSA/PBS (Sigma-Aldrich). Cells were stimulated with 1 µg H5N1 (NIBRG-14) subunit vaccine per well, overnight in a humidified CO2 incubator at 37°C. Cells were lysed with 100 µl of H2O per well and plates were washed extensively, after which 100 µl of biotinylated anti-IFNγ or anti-IL4 (Pharmingen) in 2% BSA/PBS was added 1 h at 37°C. Subsequently, the plates were incubated with 100 µl of alkaline phosphatase conjugated streptavidin (Pharmingen) in 2% BSA/PBS for 1 h at 37°C, spots were visualized with 5-bromo-4-chloro-3-indolylphosphate (Sigma-Aldrich) substrate immobilized in solidified agarose. Plates were scanned and spots were counted manually.
Plasmacytoid DCs were generated from bone marrow cells of C57BL/6 or TLR7−/− mice by seeding 1–2×106 bone marrow cells per well of a 24-well plate and culturing the cells for one week in Iscove's Modified Dulbecoo's Medium (IMDM) with 10% FCS and 100 ng/ml FLT3L (R&D Systems, Abingdon, UK) [22].
Single splenocyte suspensions were produced by collagenase D (Roche Diagnostics GmbH, Germany) treatment of the spleens, and spleen cell populations enriched for plasmacytoid DCs (pDCs) were obtained after magnetically labelling of pDCs with anti-mPDCA-1 antibody conjugated MicroBeads (Miltenyi Biotech GmbH, Germany) and separation over a MACS Column (Miltenyi), according to the manufacturers protocol. Percentages of pDCs in the positively selected population were determined by FACS analysis using anti-mPDCA-1-PE antibody (Miltenyi) and anti-CD11c-FITC (GeneTec Inc., Canada). Cell suspensions containing 1–2×105 pDCs in 100 µl were seeded in a microtitre plate and stimulated in triplicate with an equal volume containing 1.0 µg HA of either WIV, SV, or SU vaccine, or 1.0 nmol CpG DNA. After 20 h of incubation in a humidified CO2 incubator at 37°C, supernatants were collected and subjected to the IFNα ELISA.
IFNα detection in cell-culture supernatants was performed using a sandwich ELISA as described previously [15]. IFNα concentrations were calculated from a recombinant IFNα (HyCult, Biotechnology, Uden, The Netherlands) standard curve performed in quadruplicate using linear regression, and expressed in units per ml.
Statistical analysis on HI titres, antibody titres, and Elispot counts was performed with SPSS (SPSS 1202 Inc., Chicago, IL) using the Mann-Whitney U test with a CI of 95%. All p values are two-tailed. Statistical significance was defined as p<0.05.
|
10.1371/journal.pgen.1004843 | Recurrent Loss of Specific Introns during Angiosperm Evolution | Numerous instances of presence/absence variations for introns have been documented in eukaryotes, and some cases of recurrent loss of the same intron have been suggested. However, there has been no comprehensive or phylogenetically deep analysis of recurrent intron loss. Of 883 cases of intron presence/absence variation that we detected in five sequenced grass genomes, 93 were confirmed as recurrent losses and the rest could be explained by single losses (652) or single gains (118). No case of recurrent intron gain was observed. Deep phylogenetic analysis often indicated that apparent intron gains were actually numerous independent losses of the same intron. Recurrent loss exhibited extreme non-randomness, in that some introns were removed independently in many lineages. The two larger genomes, maize and sorghum, were found to have a higher rate of both recurrent loss and overall loss and/or gain than foxtail millet, rice or Brachypodium. Adjacent introns and small introns were found to be preferentially lost. Intron loss genes exhibited a high frequency of germ line or early embryogenesis expression. In addition, flanking exon A+T-richness and intron TG/CG ratios were higher in retained introns. This last result suggests that epigenetic status, as evidenced by a loss of methylated CG dinucleotides, may play a role in the process of intron loss. This study provides the first comprehensive analysis of recurrent intron loss, makes a series of novel findings on the patterns of recurrent intron loss during the evolution of the grass family, and provides insight into the molecular mechanism(s) underlying intron loss.
| The spliceosomal introns are nucleotide sequences that interrupt coding regions of eukaryotic genes and are removed by RNA splicing after transcription. Recent studies have reported several examples of possible recurrent intron loss or gain, i.e., introns that are independently removed from or inserted into the identical sites more than once in an investigated phylogeny. However, the frequency, evolutionary patterns or other characteristics of recurrent intron turnover remain unknown. We provide results for the first comprehensive analysis of recurrent intron turnover within a plant family and show that recurrent intron loss represents a considerable portion of all intron losses identified and intron loss events far outnumber intron gain events. We also demonstrate that recurrent intron loss is non-random, affecting only a small number of introns that are repeatedly lost, and that different lineages show significantly different rates of intron loss. Our results suggest a possible role of DNA methylation in the process of intron loss. Moreover, this study provides strong support for the model of intron loss by reverse transcriptase mediated conversion of genes by their processed mRNA transcripts.
| Spliceosomal introns (called introns hereafter) are noncoding DNA segments within eukaryotic genes that are removed after transcription. Although the presence of introns is one of the universal features of eukaryotes, and a large number of intron positions are highly conserved in orthologous genes across species, family and even kingdom boundaries [1], [2], intron functions and evolutionary origins continue to be a subject of much debate (see reviews in [3], [4]). The number of introns varies dramatically among organisms (see reviews in [5], [6]). Accumulating evidence suggests that the common ancestors of at least several eukaryotic supergroups were intron rich [1], [2], [7], [8] and the great interspecies difference in intron density was caused by considerably different rates of lineage-specific intron loss and/or gain [3], [5].
Patterns of intron loss and gain have been investigated extensively in numerous subclades of the eukaryotic tree of life with different levels of taxon sampling (see review in [3]). To date, vast numbers of single loss and gain events (events inferred as occurring only once in the phylogeny investigated (Fig. 1, top) have been well-documented. Some studies also document cases of recurrent loss [9]–[11] and/or recurrent gain (otherwise called parallel gain) [12]–[14], terms describing introns that are independently removed from or inserted into the identical sites more than once in an investigated phylogeny (Fig. 1, middle).
Early examples of potential recurrent intron gain came from small scale studies of single genes, including the Chironomus globin gene [14] and the fruit fly and plant xanthine dehydrogenase (xdh) genes [13]. One of the reasons that recurrent intron gain has attracted particular attention is that it has been proposed as a possible explanation for the presence of introns at the same sites in orthologous genes from distant lineages under the proto-splice site model [15], [16].
Recent studies employing genome-scale data have tried to evaluate the importance of recurrent gain through broad taxon sampling [17], [18]. These studies have argued that recurrent gain is rare and that shared introns are primarily due to evolutionary conservation. In 2009, [12] reported 4 cases of recurrent intron gain in a Daphnia pulex population based on the most parsimonious reconstruction of intron history and supporting structural evidence, suggesting that intron gain occurs with high specificity and at a high rate in this species.
Recurrent loss of introns has been reported in the mammalian glyceraldehyde-3-phosphate dehydrogenase gene [10], dipteran white gene [9], and Drosophila and mosquito multidrug resistance protein MRP1 genes [11]. Although the possibility of extensive recurrent loss in animal evolution has been proposed [4], little is known about the frequency, patterns or other characteristics of recurrent intron loss from orthologous genes since no comprehensive investigation of this phenomenon in any lineage of organisms has been reported yet.
Here we report the results of genome-wide computational identification and analysis of potential recurrent intron loss and/or gain events in five sequenced grass genomes by performing parsimonious reconstruction (Fig. 1) on trees of conserved genes and using Arabidopsis as the initial outgroup species (with additional outgroups used to confirm detected cases of intron presence/absence variation). The data show that recurrent intron loss accounts for at least 10% of all detected intron presence/absence variation sites. In contrast, we did not detect any clear case of recurrent gain. We further studied rate differentiation of recurrent loss in lineages, frequency of adjacent loss, position of lost introns in affected genes, expression patterns and functional enrichment of affected genes, intron size at turnover sites and their local DNA composition. The results of this comprehensive analysis yielded several observations that had previously been made in two-species comparisons, analyses of single gene families or broad multi-kingdom investigations, for instance that smaller introns were preferentially lost [10], [19], [20] and that rates of intron loss and/or gain varied between lineages [8], [21]. In addition, the observations that no recurrent intron gain events were detected, that genes with intron loss exhibited preferential expression in embryonic and/or gametophytic tissues, that lost introns had a high TG/CG ratio indicative of extensive CG methylation and that intron loss rates were similar across all chromosome regions have not been reported in any previous study of intron dynamics.
Our intron loss and gain detection method identified 990 intron sites at which an intron was polymorphic for presence/absence in at least one of the five grass genomes (called PA intron groups hereafter; Table 1) and 24,567 introns that were present at the same sites in homologous genes in all six genomes (conserved intron groups). The PA intron groups belong to 762 OrthoMCL clusters (that is, predicted orthologous gene families) of which the number of member genes in the six genomes ranges from 3 to 141 genes. Among them, 195 clusters (containing 230 PA intron groups) had 6 member genes and 179 (212 groups) out of the 195 clusters had single copy genes in each of the six species, which gave the peak value in the cluster size histogram (Figure S1). Information on intron turnover in the 179 single-copy OrthoMCL clusters is shown in Table S1.
By mapping the PA intron groups onto the phylogeny of the corresponding genes and performing parsimonious reconstructions, we tried to resolve the intron loss and gain history in PA intron groups (Fig. 1 and see Methods). An intron group was called well-resolved if the minimum number of intron loss or gain events gave a unique history. To study the effects of taxon sampling on the inference of recurrent intron turnover, we added orthologous genes from banana, spikemoss, and moss, and then re-analyzed the intron loss and/or gain history in groups that appeared to have undergone recurrent intron turnover based on the analysis of the six species. We found that (1) adding these outgroups helped reduce the number of unresolved histories by providing information on the ancient states of introns in some genes (examples are shown in Figure S2). (2) Moreover, adding the banana outgroup supported the conclusion for all but 4 recurrent loss groups that the intron was present in the common ancestor of the grasses. In these 4 cases, loss of introns occurred in Arabidopsis as well as one or more of the grasses. For these 4 cases, adding the outgroup did not change but further supported the intron turnover history that was modeled from analysis without the outgroup. (3) In a few cases, using additional outgroups suggested a different gain/loss history than that obtained using only Arabidopsis as outgroup. For instance, we found 2 cases (one example is shown in Figure S3) of an apparent recurrent intron gain that was the most parsimonious reconstruction in the gene clusters composed of the five grass and Arabidopsis genomes alone, but adding outgroups indicated that several recurrent losses were the correct interpretation of these data.
More precision can be gained if additional outgroups are used, but this essentially infinite task would still leave some ambiguities unresolved because of haplotype extinctions. As one further test to evaluate the accuracy of our reconstructions, we downloaded grape (Vitis vinifera) gene annotations [22] for 50 randomly selected PA intron groups and manually checked for any change in the reconstruction of intron turnover history. In all but 1 of these cases, the intron status of grape genes fully supported our reconstructions based on the 5 grasses and Arabidopsis alone. In the exceptional case, the most parsimonious reconstruction as a single gain event in rice was replaced by a more likely two losses, one in Arabidopsis and one in the common ancestor of the grasses. This difference and re-interpretation had already been detected by the comparison to M. acuminata, which also contains an intron in this location.
A summary of the properties of identified PA intron groups is shown in Table 1. We found 770 groups that experienced only one loss or one gain (652 losses and 118 gains; called single event groups) and 220 groups, accounting for 22% of all PA intron groups, that required at least two events to have independently occurred. Out of these 220, 113 groups were well-resolved, including 93 groups experiencing recurrent loss (recurrent loss groups), and 20 mixed events groups (i.e., loss followed by gain (3 groups) or gain followed by loss (17 groups)).
In the single event groups, the number of losses was found to be 5.5 times higher than that of intron gain (652∶118) and, in recurrent event groups, we did not detect any confirmed recurrent intron gains. When using the gene set only from the six genomes (five grasses plus Arabidopsis), 5 intron groups appeared to show recurrent gain. However, adding more distant outgroups indicated that 2 of them were more likely exhibiting recurrent loss and the other 3 were unresolved in the larger gene sets. In contrast, adding outgroups supported recurrent loss in most (93 out of 97) cases. In four cases, adding banana genes altered the gene tree topology and they became unresolved. Hence, recurrent gain, if it occurs at all, is much less frequent than single gain (0 vs 118 in our analysis).
Because the scale of our phylogenetically analysis limited the number of recurrent events that could be detected, it was not surprising to observe that most (82 out of 93; 88%) of the recurrent loss groups experienced only 2 independent losses of any specific intron (Table S2). In fact, the maximum number of recurrent events that can be detected in the five grass genomes if an orthoMCL cluster consists entirely of five orthologous genes is 3. According to the parsimony principle, events in paralogous genes that resulted from a species-specific duplication event were counted only once. An example is shown in Figure S4, where an intron was absent in both members of this rice tandem gene pair, but only 1 loss was counted.
We found that independent intron losses occurred between orthologs in 46 recurrent loss groups and between paralogs in 41 groups, respectively. Because this analysis only captured recent paralogs due to the homology criteria involved in creating OrthoMCL clusters, these relative numbers do not provide a comparable analysis of overall intron stability in orthologs versus paralogs. In the other 6 groups, loss or gain occurred between both orthologs and paralogs. Examples of these three patterns are shown in Figure S5.
Overall, we detected 858 intron loss events (652 single +206 recurrent loss events, Table 1) out of a total of 176,078 intron locations analyzed which were members of 25,557 intron groups (990 PA + 4,567 conserved groups). The 858 loss events resulted in the absence of 1635 introns in affected genes. 8782 (212 PA (Table S1) +8570 conserved) out of the 25,557 intron groups had gene tree topologies identical to the topology of the species tree. In this intron group subset, the total number of intron locations was 43,910. There were 182 intron loss events (including 37 recurrent losses, Table S1) and these resulted in 257 intron losses. To determine whether some introns are preferentially lost, we randomly assigned 257 intron absences to the 43,910 intron locations and counted the number of recurrent intron loss events. This random assignment process was repeated 10,000 times and the average and maximum number of recurrent intron losses were found to be 3.5 and 19, respectively. Because the actual number of recurrent losses in this data set was 37, this indicates that the observed recurrent loss is significantly higher (p-value <0.0001) than expected based on the hypothesis that intron loss was fully random in this intron group subset.
We identified 1026 intron loss and/or gain events in the 883 resolved PA intron groups, including 879 loss and 147 gain events. The number of events in the 93 recurrent loss groups was 206, or ∼20% of all events. We investigated lineage differentiation of the average frequency of intron loss (measured by the number of events/branch length) in the grass family and our results (Fig. 2 and Figure S6) revealed that (1) single loss occurred at a higher frequency than recurrent loss in all branches (counts/time are tested; log-linear model with time as offset, p-value <0.0001); (2) sorghum and maize had significantly (counts/time are tested; log-linear model with time as offset, p-value <0.0001) higher frequency of intron loss than foxtail millet, rice or Brachypodium in both single loss and recurrent loss groups; and (3) in the BEP clade, Brachypodium had a higher (not significant, p-value = 0.22) frequency than rice.
Another interesting trend was that the frequency of detected recurrent loss tended to be lower in more ancient branches. In the 5 terminal branches, the frequency was higher than or close to 1/MY, while in the three older branches, the Panicoid, BEP and grass values were 0.3, 0.2 and 0.2, respectively. This might be because the smaller number of ancient branches and useful outgroups are expected to yield a higher frequency of intron changes that are not resolvable at the older nodes. The lower intron loss frequency was less obvious in the single loss groups (Figure S6a), where only the branch representing the common ancestor of all grasses had a statistically significant lower (counts/time are tested; log-linear model with time as offset, p-value <0.0001) frequency than the other branches. The predicted lower intron loss frequency in ancient lineages could also be seen when calculating the frequencies for all single gain (Fig. 2c) or for all turnover events (Figure S6b).
Information on the presence or absence of a target intron in an ancestral organism can only be obtained from analyzing sister or ancestral taxa. Hence, the closer to the root of the phylogeny an event occurred, the fewer such taxa are available, thus making it more difficult to resolve more ancient events. This is expected to contribute to the lower observed frequency of intron loss observed, when compared to recent branches. Further study with increased taxon sampling will be needed to draw solid conclusions about intron loss and gain frequencies in the most ancestral stages of these lineages.
Because we know the total number of current introns that we were able to properly align, it is possible to estimate the rate of intron loss or gain events measured per intron per year in the terminal nodes. The overall number of analyzed introns in Brachypodium, rice, sorghum, foxtail millet and maize genomes were 31685, 30685, 32386, 33230 and 37119, respectively, and these numbers were used to derive the rates indicated in the terminal nodes for Fig. 2 and Figure S6.
Besides recurrent events, we also detected 10 OrthoMCL clusters (Table S3) that might have experienced retroposition or conversion between the intronless cDNA and the original gene copy. In their ancestral forms, the genes in these ten clusters contained at least 8 introns each. However, all introns were absent in all members of specific clades in the gene trees (Table S3). Both reverse-transcriptase-mediated (RT-mediated) intron loss (basically, cDNA-based gene conversion) [5], [23] and retroposition can lead to simultaneous loss of multiple introns, with all introns routinely being removed by the latter process. Retroposition [24], [25] usually generates an intronless copy that is located at a different locus from the original copy through the activity of retroelements like LINEs. RT-mediated intron loss, however, removes introns in the original gene but does not change its physical position in the genome. This observation provides the opportunity to discriminate between the two types of events. We found that intronless members of 2 gene clusters were located in syntenic blocks [26], [27] and thus were probably modified by cDNA conversion. Intronless members of the other 8 clusters were in new (non-syntenic) locations. In all of these cases. all introns were removed, suggesting retroposition. However, given the frequent movements of single genes to new locations that have been documented in the grasses [28]–[30], it is also possible that some of these are cases of cDNA conversion with prior or subsequent gene movement. As a control, we determined that only 33% of intron-retaining grass members of the 8 clusters were in non-sytenic blocks. Therefore, the 8/10 fully intronless genes in new locations were at a frequency suggesting that at least some lost their introns during retroposition.
In a gene, several introns may experience loss and/or gain events, and this corresponds to multiple PA intron groups in one OrthoMCL cluster. We found that 628 clusters contained 1 PA intron group and 134 contained 2 to 7 lost or gained introns. The PA intron groups at different locations exhibited identical presence/absence polymorphic patterns for introns in non-sister lineages in only 4 of these 134 clusters, suggesting recurrent loss of the same set of introns independently. In the other 130 multiple-intron-loss clusters, PA intron groups at different locations had at least two different presence/absence patterns. Furthermore, 87% of the 130 clusters contained 2 or 3 PA intron groups (Table S4) and the total number of PA intron groups in the 130 clusters was 364.
We further investigated the frequency of adjacent intron loss. We found 84 neighboring PA intron groups in the 25,557 intron groups (PA + conserved) analyzed. Among them, 57 were confirmed adjacent intron losses from the same lineage, suggesting origin in a single event. These 57 adjacent pairs belonged to 40 OrthoMCL gene clusters (Table S5). We found 11 cases of> = 3 adjacent intron losses. If adjacent triple PA intron groups (those with losses of three introns in a row in the same gene) were counted as 2 pairs, and if adjacent intron turnover events were independent, the probability of turnover of a neighbor intron could be described by a binomial distribution. A statistical test of the observed number of adjacent intron loss groups rejected the independent turnover hypothesis at the 0.05 level.
In cases where adjacent introns were lost from the same lineage, a size-limited model of intron loss (for instance, by nuclear cDNA conversion or aberrant double-strand break repair) would suggest that adjacent introns with a smaller intervening exon would be more likely to be lost in a single event. Surprisingly, we found that the sizes of intervening exons was not significantly different between adjacent introns lost in a single lineage (presumed single events) when compared to the intervening exon sizes of adjacent introns lost in two independent events (Figure S7).
When intron turnover occurred in a terminal node of the gene tree, the gene that underwent intron loss and/or gain was the affected gene of that event. When intron turnover occurred in an internal node, all descendants of that node were taken as affected genes. Therefore, one event might have more than one affected gene. The total number of genes that underwent intron turnover was 1798, including 308 involved in the category of recurrent loss, and 1142, 179, and 169 in single loss, single gain and gain/loss mixtures, respectively (Table 1). Some genes were counted more than once because intron turnovers occurred at multiple locations in some gene clusters. For example, if two introns in a gene family experienced recurrent loss and single loss, respectively, this gene was counted twice. If all genes were counted only once, the total number of affected genes of intron turnover was 1720.
We normalized for gene size and investigated the distribution of intron loss or gain along each gene (Fig. 3 and Figure S8). The locations of all introns in all gene models of the six genomes showed a relatively even distribution, with the two termini, i.e. (0, 0.1) and (0.9, 1.0) of the total length of a gene when calculating from the 5′ end, exhibiting lower values (Fig. 3a). Under-representation of intron loss and/or gain at the gene termini was also observed in recurrent (Fig. 3b) and PA intron groups (Fig, 3c): the percentage of recurrently lost and PA introns located at (0, 0.1) and PA introns at (0.9, 1.0) was lower than expected based on the mean and sd from 1000 resamplings from all introns. In the all loss (recurrent + single) groups, we observed (0.2, 0.3) had a percentage higher than expected based on the resampling results (Figure S8).
We investigated the distribution of intron turnover across chromosomes (Fig. 4 and Figure S9) by normalization of chromosome size. The centromere was located at 0; and then the short arms and long arms of chromosomes were normalized separately, with the short and long arm termini located at -1 and 1, respectively. Locations of genes belonging to the same intron group were counted independently. The distribution of the whole gene set of the five genomes (Fig. 4a) showed a smooth “V” shape with the lowest gene density located at the centromeric/pericentromeric region. The distributions of genes with detected recurrent loss (Fig. 4b) and total intron turnover (Fig. 4c) exhibited a similar overall trend. However, it should be noted that plant genes are highly mobile over evolutionary time, and even centromeres can be found in different positions in closely related lineages [31], so the current genomic location is not a perfect predictor for any single gene of its location when an intron was gained or lost.
Resampling showed that the percentage of genes that underwent recurrent intron loss located in (0.3, 0.4), the region close to the middle of a normalized long arm, was higher than expected from the resampling mean and sd (Fig. 4b). The density ratio calculated as the density (number of genes per normalized unit length of a chromosome) of genes with recurrent intron loss divided by the density of the entire gene set exhibited a peak at this region (Figure S10), suggesting hotspots for intron recurrent loss events at these regions. When considering all PA introns, a higher than expected percentage of affected genes was located at the chromosome arm ends (−1, −0.9) and (0.9, 1) (Fig. 4c).
We found that 698 (91.6%) of the 762 OrthoMCL clusters with intron turnover events were derived from genes with multiple introns and the other 64 (8.4%) were from ancestral genes with a single intron. The 93 resolved recurrent loss intron groups belonged to 88 gene clusters, 4 (4.5%) of which were from single-intron ancestral genes. The frequency of all loss and recurrent loss occurring in single intron genes is 8.4% and 4.5%, respectively. A Pearson's Chi-square test (p-value = 0.2918) indicated that, compared to the PA intron groups, the number of single-intron genes was neither over-represented nor under-represented in the recurrent loss groups.
A total of 1026 confirmed intron turnover events (Table 1) were found among all introns analyzed in the grasses (588,669) and 65 of these events were in one of the 18,226 single-intron genes. A Pearson's Chi-square test (p-value = 1.75e-8) indicated that the rate of intron turnover in single intron genes (65/18,226 or 0.36%) is significantly higher than the rate of intron turnover overall (1026/588,669 or 0.17%).
We investigated the codon phase distribution of intron turnover events and found that 58% (575/990 PA intron groups), 20% (195) and 22% (220) of turnover events involved introns in phase 0 (intron-exon boundaries located between two codons), phase 1 (between the first and second base of a codon), and phase 2 (between the second and third base of a codon), respectively. The excess of phase 0 introns has been well-documented (see review in [3]) and intron phase distribution in rice has been estimated at 57∶22∶21 for phase0: phase1: phase2 [32], very close to our estimation for intron turnover in the five grass genomes (56∶21∶23). Statistical analysis (Pearson's Chi-square) indicated no significant difference between the overall intron phase and PA intron phase data.
We investigated the expression patterns of the 289 rice genes that exhibited intron turnover because publicly available expression data are relatively abundant for rice. A total of 283 out of the 289 genes matched probes in the rice 57K Affymetrix GeneChip (http://www.affymetrix.com/). 283 probes were selected through PLExdb's “Gene List Suite” tool using these genes as queries. Each probe corresponded to a single gene. When a gene matched multiple GeneChip probes, the probe with highest BLAST bit score was used to represent the gene. We extracted gene expression information for eleven stages in early embryogenesis and six for pollen development from multiple experiments deposited in PLEXdb (Table S6; see Methods). Since expression values are not comparable between different experiments, we only focused on whether genes that underwent intron turnover were detectably expressed or not in a given tissue. As shown in Table 2, compared to total nuclear genes, significantly higher (Pearson's Chi-square test, see Table 2) percentages of genes with intron loss events were expressed in both early embryogenesis and pollen development both when considering the average percentage of genes expressed in all developmental stages and the percentage of genes expressed in at least one of the developmental stages. Genes in the PA category, which includes all genes that underwent intron loss and/or gain, also were expressed significantly more frequently in these two expression categories than total nuclear genes. The fact that genes that have undergone intron loss exhibit a higher frequency of germ line or early embryogenesis expression than observed for the complete transcriptome suggests that expression in heritable tissues is associated with transmission of the intron removal outcome.
GO analysis of genes in all intron gain and/or loss categories indicated numerous terms that were over-represented at a level considered statistically significant by the GO analysis software (see Methods). These over-represented categories are shown in Figure S11. Among them, the top five most significant GO terms were from the following functional categories: ‘catalytic activity’, ‘oxidoreductase activity’, ‘metabolic process’, ‘omega-3 fatty acid desaturase activity’ and ‘positive regulation of protein modification process’.
We compared the size and composition of PA intron groups to that of all intron groups. The average sizes of introns in recurrent loss intron groups, all PA groups and conserved groups were 212 bp, 266 bp and 360 bp, respectively. The average size of PA introns (total intron count = 7950) and recurrent loss introns (1097) were both significantly shorter than that of conserved introns (201,660) (Mann-Whitney test, p-value <2.2e-16). The preference for loss of short introns has been reported previously [10], [19], [20]. It is not clear why smaller introns might be more easily lost, although two distinct possibilities come to mind. Smaller introns may be less likely to contain important regulatory modules, so that their loss would be less likely to detrimentally affect gene function. A second possible explanation is that intron removal (perhaps by a cDNA conversion process or aberrant NHEJ repair) is tolerated only if complete because partial removal would leave an unspliceable intron fragment, and thus smaller introns would be more likely to be fully removed in any time-constrained process. A third possibility is selection for smaller transcript size or fewer introns so that genes might be more rapidly transcribed and/or matured to increase gene expression [33]. However, we observed that small genes were just as likely to lose and/or recurrently lose introns as large genes (Figure S12), indicating that transcription rate is not a clear factor in any possible selection for intron loss. We also did not observe any correlation between organismal genome size and overall intron number or intron loss rate (Figs. 2, Figure S6), nor exceptionally frequent loss of introns from genes with large intron numbers (Figure S13), so a model suggesting selection against introns per se is not supported by these observations. The observation in Figure S13 that introns are most frequently lost from genes with few introns does not provide any obvious mechanistic model for this preferential loss, but does suggest that many of these genes with very few introns are the ongoing products of especially frequent intron loss. Consistent to our observation of non-random loss of introns, Figure S13 suggests that not all introns in a gene have an equal likelihood of being removed.
Although recurrently lost introns, like single-loss introns, are smaller in size than the average intron, this fact does not by itself explain recurrent loss. That is, if one compares the size of recurrently lost introns, there is no significant different in their size compared to single loss introns (Figure S14).
We investigated the nucleotide compositions of different categories of intron groups (recurrent loss, PA, and conserved intron groups) as well as their flanking exonic sequences in the five grass genomes. As depicted in Table S7, the mean G+C content of recurrent loss introns, PA introns and conserved introns identified in OrthoMCL groups were very similar (39%) and lower than the genome-wide G+C content (44%) or the average G+C content of exons (55%). The lower G+C content of introns compared to exons in plants is a well-known phenomenon (e.g. [34]–[36]) and is believed to be related to the high G+C content of plant triplet codons [37], [38]. The mean G+C contents of the last 20 bp of the upstream flanking exon and the first 20 bp of the downstream flanking exon were similar in recurrent loss and PA intron categories (55%). This value is identical to the genome-wide G+C average for exons. Interestingly, the mean G+C content of the 20 bp of exonic sequences flanking conserved introns (47%) was about 8% less than that of the 20 bp of exonic sequences flanking PA introns. Pearson's Chi-square test indicates that this difference is highly significant (p-value <2.2e-16).
We investigated dinucleotide frequencies for all PA and conserved introns. The three intron categories (recurrent loss, PA and conserved) exhibited similar frequencies of different dinucleotides (Table S8). In all three categories, TT, AT, AA, TG and TA were the most abundant dinucleotides, while GC, CC, GG and CG were the least abundant dinucleotides. A similar tendency was found in the genome-wide dinucleotide frequency, where CG had the lowest frequency and AA, TT and AT had the highest frequencies. While CG was also the least frequent dinucleotide in exons flanking conserved introns, it was TA that had the lowest frequency in exons flanking lost introns, in agreement with the higher AT-richness of exons flanking conserved introns mentioned above. Within introns, TG was the fifth (recurrent loss), fourth (PA) and third (conserved) most abundant dinucleotide across the three different categories. Interestingly, the TG/CG ratio was 2.4, 2.9 and 4.2 in recurrent loss, PA and conserved categories (Tables S9 and S10). The differences in the TG/CG ratios among the three intron categories were highly significant in terms of Pearson's Chi-square test with Bonferroni correction (Table S10). A low “CG” and high “TG” frequency suggests the process of “C” to “T” transition that is enhanced by 5-methylation at cytosine bases [39], [40]. Introns in general are observed to be relatively less methylated than exons in plants, especially at cytosine bases [41]. This result suggests that introns with a history of less CG methylation are more likely to be removed.
Flanking exons of conserved introns also had a relatively higher TG/CG ratio than flanking exons of PA and recurrent loss introns (TG/CG = 2.0 vs. 0.9 and 0.77; Table S9). Hence, these results suggest that conserved introns and their flanking exons were relatively highly methylated while intron turnover tended to occur where the degree of cytosine-5 methylation in both the flanking exons and introns was relatively low.
Although the GT….AG terminal intron dinucleotides are most abundant in all studied eukaryotes, including plants, rarer junction sequences are also found [42]. Table S11 shows the terminal intron dinucleotides associated with conserved introns and PA introns. No significant differences were observed between conserved or PA introns, regardless of whether the intron location had the added precision of confirmation by transcript analysis.
Many previous parsimonious reconstructions of intron loss and/or gain [2], [43] were based on Dollo parsimony, which assumes that every intron arose only once along the tree and thus explicitly excludes parallel gains. Recent studies, however, suggest that this assumption is not valid [8]. Prohibiting recurrent intron gain is also a characteristic of some probabilistic methods [44]–[46]. However, one of our aims was to provide an unbiased assessment of the frequency of intron gain. Hence, a cladistic parsimonious counting strategy (See Fig. 1 and Methods) was employed that initially modeled the fewest intron turnover events in the tree but did not restrict the occurrence of recurrent gain or loss. The parsimony method was valid as a first step in the analysis as we had no prior knowledge on the frequency of gain or loss. Our results indicate that intron loss is much more frequent than intron gain (858 losses (206 recurrent events +652 single events): 118 gains = 7.2: 1) (Table 1), consistent with results from several previous investigations in plants [32], [47], [48].
Use of a simple cladistic approach to predict the origin of a rearrangement is not valid when the rates of the two mechanisms of rearrangement are quite different. In our case, for example, where intron gain occurs at a 7-fold lower rate than intron loss, parsimony tends to overestimate gain and underestimate loss since it cannot be guaranteed that any of the events judged as intron gains are not actually cases of multiple independent losses that could not be resolved due to a lack of phylogenetic power. In fact, in all unresolved and mixed groups, the gain events predicted by the parsimony principle could also be explained by two or more recurrent intron losses. In 18% (21 out of 118) of single gain groups, the gain can be substituted by 2 losses. If every gain is allowed to be substituted by at most 3 losses, the percentage reaches 31% (37 out of 118). Given the at least 7.2 fold higher frequency of losses than gains that we observed under the parsimony model, even models with 3, 4, 5 or 6 recurrent losses are more likely than single gain events, thus allowing all of the gains to be potentially explained by recurrent loss.
The origin of new introns requires the insertion of precisely positioned functional splicing signals inside (and, presumably by chance, flanking) the new intron. Unlikely as this seems, many models (see [49] for a short summary) of frequent intron gain have been proposed, including (1) modification of self-splicing introns [50]; (2) intronization of coding sequences [51], [52]; (3) transposon insertions that become precise introns [53]–[55]; (4) tandem duplication of coding sequence with an AGGT motif to create an intron terminal signal [50]; (5) nonhomologous end-joining (NHEJ) of DNA segments [12], [56]; and (6) intron transposition [57]–[59]. In plants, only one of these models, a transposon insertion turned intron, has been supported by in vivo evidence [53]–[55]. One case has been observed in which a new intron did not dramatically debilitate gene function because it was precisely processed [55]. However, none of the proposed intron gains that have been fixed in a species have been associated with a transposable element insertion/structure, so rare intraspecies variation for this trait may be more an indication of mutational neutrality than a major mechanism of evolutionary change by intron gain.
The rarity of intron gain implies that the chance of recurrent gain should be even lower. Previous reports of recurrent intron gains [12]–[14] were based exclusively on a cladistic approach with few investigated taxa. The main reason that these investigators preferred recurrent gains is that the intron distribution in their phylogeny could be explained more parsimoniously by invoking less independent intron gains than losses: 2 gains v.s. 8 losses in the globin gene analyzed by [14] and 2 gains v.s. 14 losses in the xdh genes analyzed by [13]. As mentioned above, when the frequency of loss is much higher than gain — and this has been supported by the current study as well as by previous analyses [10], [32], [47], [48] — the cladistic approach tends to mistake recurrent losses as individual or recurrent gains. Furthermore, as is shown in our study, using a single species as outgroup may lead to incorrect assumptions on the ancestral state of an intron. In short, more convincing evidence (such as that provided in [12]) of recurrent intron gain is needed to determine whether this proposed phenomenon has a solid foundation in plants. However, phylogenetically deep and/or molecularly detailed studies of intron gain have recently been published for a number of organisms [12], [56], [60]–[64], so it will be interesting to see if any future plant studies uncover similar phenomena.
Including outgroups, the phylogenetic scope of the three previously reported cases of recurrent intron loss were in single genes in vertebrates [10], Bilateria [9] or Pancrustacea [11]. In our research, the scope is extended to the last common ancestor of monocots and dicots. During the ∼150 MY evolution of the grass family since their divergence from a common ancestor with Arabidopsis, we have identified 93 (initial parsimony result) to 200 (all possible cases, including those currently unresolved) polymorphic intron sites that are likely to have experienced recurrent losses. The rate of recurrent intron loss thus is 2.4−5.2×10−5/MY/intron site (total analyzed intron sites = 25,535). The counting of recurrent loss is affected by the depth of phylogeny and taxon sampling. For example, recurrent losses in rice and maize will appear as a single loss in each species if BEP clade and PACCAD clade species were investigated separately. Here BEP is an abbreviation representing the clade of grass subfamilies Bambusoideae, Ehrhartoideae, and Pooideae; and PACCAD an abbreviation representing the clade of grass subfamilies Panicoideae, Arundinoideae, Chloridoideae, Centothecoideae, Aristidoideae and Danthonioideae. As a further investigaton of this point, we extended the intron loss analysis for six sets of orthologous genes that contained a total of 9 recurrent loss groups and 18 intron loss events by manually aligning the orthologs from the five grass species. These were analyzed with the orthologs from one additional species (2 genes) or two additional species (4 genes) for which sequence information was available, and this identified 3 additional recurrent loss events, all in the same gene (unpub. obs). Thus, adding more taxa will lead to the detection of more recurrent losses, indicating that the current estimates are very conservative minima.
Our results show that recurrent loss accounts for a considerable proportion of all detected intron losses. Moreover, loss is not random, but involves a small subset of introns that are lost over and over again in multiple lineages. Our analyses indicate that some of these introns may have been lost repeatedly because of their (1) small size, (2) expression in tissues that contribute to the germ line or (3) history of less 5-methylation in regional CG dinucleotides. Fawcett and colleagues [48] suggested that the high rate of intron loss in a small genome might be caused by strong selection for genome reduction in their study of two Arabidopsis species. However, consistent with previous studies [10], [19], [20], our results shows that short introns are preferentially lost, a result not consistent with selection for intron loss as a mode of genome size reduction. Moreover, we observed a higher frequency of intron loss in larger compared to smaller grass genomes. Both observations suggest that selection is not on the basis of an effect on genome size. Given that the loss of a single intron in plants will only decrease genome size by a few hundred bp in most cases, it is difficult to see how this would be detected and/or significant within a several gigabase genome. Another theory is that loss of an intron could lead to more efficient transcription [33], but this model is also inconsistent with a preference for short intron removal.
Based on our observation that lowly methylated introns are more likely to be removed than highly methylated introns, it is possible that there might be selection related to the epigenetic status of genes. Methylation of DNA is associated with specific chromatin compositions/conformations and an epigenetic status that tends to have negative effects on transcription [65], [66]. Hence, loss of an intron with a particular level of DNA methylation might be expected to have a selected epigenetic outcome, perhaps leading to an altered level or timing of gene expression.
Although natural selection might explain some or all recurrent losses, it is also possible that coincidental features of intron structure, gene location and/or gene expression might explain a high rate of removal for specific introns. Two general mechanisms for intron loss have been proposed, but neither has gained comprehensive support yet [49]. One proposed model is intron loss by genomic deletion [4], [5] via NHEJ [48] or other molecular mechanisms, leading to exact [48], [67] or inexact [20], [68] intron removal. The other proposed mechanism is an RT-mediated intron loss model. The standard version of this mechanism predicts that introns located at 3′ ends of genes are more likely to be lost because reverse transcriptase reads/polymerizes from 3′ to 5′ along the RNA template and often produces incomplete transcripts [5], [69], [70]. However, a lack of 3′ bias of intron loss has been reported in several species, including plants [32], [48], animals [43], [71] and fungi [46], [60]. So, modified RT-mediated intron loss models involving self-primed reverse transcription have been proposed [72]–[74], but there is not yet any compelling evidence to support these models [43], [60], [75].
Our data indicated that large grass genomes (sorghum and maize) have a significantly higher intron removal rate and that short introns are more commonly removed. These observations are compatible with the hypothesis that RT-mediated intron loss plays a role in driving intron loss because large genomes contain more class I TEs and thus may have higher reverse transcriptase activity. Furthermore, smaller introns are more likely to be lost by RT if the enzyme is prone to incomplete template coverage. In addition, several other results in this study provide indirect support for an RT-mediated intron loss mechanism, including (1) adjacent intron loss occurring at higher frequency than expected by chance, (2) genes exhibiting intron loss are enriched in germline and early embryogenesis transcriptomes; and (3) deletions of introns are exact. The lack of a 3′ bias for intron loss detected in our study, however, conflicts with the simplest models of cDNA conversion for intron loss, as does the observation that the concurrent loss of adjacent introns does not seem to be highly affected by intervening exon size. Another model, involving ncRNAs that direct DNA rearrangement (including deletion) [76], does not necessarily require RT activity or 3′-end priming at an mRNA polyA, but still would predict a more frequent loss of adjacent introns if they were separated by smaller exons. To increase our understanding of these issues, studies are needed to investigate de novo intron loss, and these studies would be best performed with introns in genes that exhibit a history of recurrent loss in other lineages and in genetic backgrounds that are enriched for RT and/or for NHEJ activities.
The genomic sequences and annotation data for maize (Zea mays, version 5b.50) and rice (Oryza sativa, IRGSP/RAP Build5) were downloaded from MaizeSequence (http://www.maizesequence.org/index.html), IRGSP (http://rgp.dna.affrc.go.jp/E/IRGSP/) and RAP-DB (http://rapdb.dna.affrc.go.jp/) databases. Data for the sorghum (Sorghum bicolor, version 1.4), foxtail millet (Setaria italica, JGI release 164) and Brachypodium (Brachypodium distachyon, JGI release 114) genomes were obtained from the Phytozome website at DOE-JGI (http://www.phytozome.net/). Data for Arabidopsis thaliana (version TAIR10) were retrieved from TAIR (http://www.arabidopsis.org/). Raw reads for the five grass genomes were downloaded from the NCBI Trace Archive Database (ftp://ftp.ncbi.nih.gov/pub/TraceDB/) and the reads for Arabidopsis thaliana were from the 1001 Genomes project website (http://1001genomes.org/). Besides these six species, sequences and annotation of banana (Musa acuminata, version 1), was downloaded from The Banana Genome Hub (http://banana-genome.cirad.fr/); data of spike moss (Selaginella moellendorffii, JGI release 91) and moss (Physcomitrella patens, JGI release 152) were also downloaded from Phytozome v8.0.
The estimated positions of maize, sorghum, foxtail millet, Brachypodium and Arabidopsis centromeres were directly extracted from several earlier publications [77]–[81]. The positions of rice centromeres were collected from the Rice Genome Annotation Project (http://rice.plantbiology.msu.edu/annotation_pseudo_centromeres.shtml), and sequences around these regions were compared to IRGSP/RAP Build5 to find the corresponding locations.
Intron group identification: The workflow of intron group identification is shown in Figure S15. Firstly, the representative gene repertoires of the five grasses and Arabidopsis were extracted and orthologous clusters (called OrthoMCL clusters) were built using OrthoMCL [82] with default parameter settings. Here, we used the longest transcript as a representative sequence for each gene. Secondly, protein alignment-guided multiple sequence alignments (MSA) of coding DNA sequences (CDS) of each OrthoMCL cluster were constructed using TranslatorX [83] with default parameter settings. Next, OrthoMCL clusters too large or small (>200 or <3 member genes) or clusters of which>30% of the members matched known transposon proteins were excluded. This last step was designed to remove from consideration those gene models where transposable elements (TEs) were found inside coding sequence, because we have noted that most of these are mis-annotations (especially pseudogenes mis-annotated as genes) ([84], H. Wang and J. Bennetzen, unpub. res.). The analyzed genes that remained included many cases of introns (both conserved and PA variants) that contained TEs.
Subsequently, we extracted intron positions from gene models annotated in the various genomes and mapped these positions to the above protein-guided CDS alignment. An intron in the genomic sequence thus was mapped to a position between two consecutive bases corresponding to the last base of exon k (base i) and the first base of exon k+1 (base i+1) in a CDS (Figure S15, top right). In the CDS alignment, whenever a position between two consecutive bases had break points, we called this position an intron group. If all member genes in the intron groups had break points, this indicated that each member gene had an intron at the same position and this group was judged a conserved intron group. If only some member genes had break points, the group was called a PA intron group (Figure S15, middle left). We identified all intron group candidates for all OrthoMCL clusters with this method. We further required that the up- and downstream exons of detected intron groups were well-aligned (flanking exons in each member gene exhibited ≥60% identity to consensus sequences of the flanking exon alignments) (Figure S15, middle right). This homology restriction led to the selection of conserved intron sites and excluded artificial intron groups caused by MSA algorithms. One possible issue was “intron sliding” [3], [85], where an intron was not actually lost or gained, but moved one to several nucleotides away due to a shift in the intron/exon boundaries. With an inappropriate detection method, “intron sliding” might be perceived as an intron gain, intron loss or (most likely) reciprocal intron gain and loss in two different lineages. Hence, we required that the alignment be perfect at the exact ends of the intron/exon junction, and that no two intron PAs were allowed to be within <20 bp of each other. We manually checked all 990 of our PA intron groups to see if any were due to intron sliding, and none were. Next, we excluded intron groups containing very short introns (≤10 bp) to avoid artifacts generated by incorrect intron annotation (Figure S15, bottom). Finally, selected intron groups were compared with raw reads to confirm that the presence or absence of introns in the groups was not caused by assembly errors (Figure S15, bottom and Figure S16).
Another possible issue with these analyses involves the quality of the gene annotations that we accepted from the published genomes investigated. In particular, it was not clear whether we would see any differences in intron presence/absence variation properties in cases where introns were confirmed by transcript analysis. Transcript data (e.g. ESTs) covered the breakpoint plus at least 20 bp upstream and downstream of the intron boundary for 946 of the 990 PA intron groups in at least one of the species investigated. Although the number of PA introns without transcript support was too small to allow statistically significant values to be obtained when comparing PA intron properties to those PA introns with transcript support, in all cases the trends were in the same direction.
Gain or loss event resolutions: For every PA intron group, we mapped the intron presence/absence pattern on the corresponding gene tree. The history of intron turnover of the group was reconstructed according to the parsimony principle which assumes that the history with the lowest number of intron turnover events has the highest likelihood of representing the true chain of events (Fig. 1). If the parsimonious reconstruction corresponded to more than one possible intron loss and/or gain history, the intron group was called an unresolved group. For every group for which recurrent events were inferred from the six genome analysis, we added in orthologous genes from banana, spike moss and moss and redid the intron loss and/or gain history inference. This analysis allowed us to confirm reconstructions based on fewer species and demonstrate that none of the initial recurrent gains calls were supported by the broader cladistics analysis. Detailed information for all intron loss events identified in this study is provided in Table S12.
TreeBeST (http://treesoft.sourceforge.net/treebest.shtml) was used to build the gene trees for the OrthoMCL clusters. The program constructed trees with the Maximum Likelihood method under guidance of the species tree. The species tree topology used in this study was (((((((Zmay, Sbic)Andropogoneae, Sita)Panicoid,(Bdis, Osat)BEP)Grass, Muca)Monocot, Atha)Angiosperm, Smoe), Ppat). The 4-letter genome codes used were Zmay: Zea mays; Sbic: Sorghum bicolor; Sita: Setaria italica; Bdis: Brachypodium distachyon; Osat: Oryza sativa; Muca: Musa acuminata; Atha: Arabidopsis thaliana; Smoe: Selaginella moellendorffii; Ppat: Physcomitrella patens.
Using previous estimations of the divergence time of the five grass species, i.e. 150 million years (MY) between Arabidopsis and grasses [86], [87]; 60 MY between BEP and Panicoids [87], [88], 47 MY between rice and Brachypodium [77], 26 MY between foxtail millet and Andropogoneae [78], and 12 MY between sorghum and maize [89], branch lengths of the grass species tree could be scaled as evolutionary time. The mean rates of intron loss or gain in branches were calculated as the number of events divided by the branch length.
GO annotation and enrichment analysis of genes exhibiting intron loss or gain were performed in AgriGO (http://bioinfo.cau.edu.cn/agriGO/) [90] using “suggested backgrounds” as references. These backgrounds were the GO annotation of whole gene sets of organisms. For Brachypodium, sorghum and foxtail millet, only one background was available. For rice and maize, we chose the annotation labeled as “MSU 7.0 nonTE” and “Zea mays ssp.”, respectively. In all analyses, statistical tests were performed using the Fisher exact test and the multi-test adjustment method according to Yekutieli [91]; the significance level was set to 0.05; and complete GO was chosen as gene ontology type.
Rice expression data were downloaded from PLEXdb (http://www.plexdb.org/). Normalized expression data from various experiments (Table S6) were extracted for early embryogenesis and germ line cells. Details of these experiments can be found at the PLEXdb website under the “Expression Atlases” link. We identified probes corresponding to the rice genes exhibiting intron turnover with the “Gene List Suite” tool (http://www.plexdb.org/modules/glSuite/gl_main.php).
|
10.1371/journal.pbio.1001810 | Dual PDF Signaling Pathways Reset Clocks Via TIMELESS and Acutely Excite Target Neurons to Control Circadian Behavior | Molecular circadian clocks are interconnected via neural networks. In Drosophila, PIGMENT-DISPERSING FACTOR (PDF) acts as a master network regulator with dual functions in synchronizing molecular oscillations between disparate PDF(+) and PDF(−) circadian pacemaker neurons and controlling pacemaker neuron output. Yet the mechanisms by which PDF functions are not clear. We demonstrate that genetic inhibition of protein kinase A (PKA) in PDF(−) clock neurons can phenocopy PDF mutants while activated PKA can partially rescue PDF receptor mutants. PKA subunit transcripts are also under clock control in non-PDF DN1p neurons. To address the core clock target of PDF, we rescued per in PDF neurons of arrhythmic per01 mutants. PDF neuron rescue induced high amplitude rhythms in the clock component TIMELESS (TIM) in per-less DN1p neurons. Complete loss of PDF or PKA inhibition also results in reduced TIM levels in non-PDF neurons of per01 flies. To address how PDF impacts pacemaker neuron output, we focally applied PDF to DN1p neurons and found that it acutely depolarizes and increases firing rates of DN1p neurons. Surprisingly, these effects are reduced in the presence of an adenylate cyclase inhibitor, yet persist in the presence of PKA inhibition. We have provided evidence for a signaling mechanism (PKA) and a molecular target (TIM) by which PDF resets and synchronizes clocks and demonstrates an acute direct excitatory effect of PDF on target neurons to control neuronal output. The identification of TIM as a target of PDF signaling suggests it is a multimodal integrator of cell autonomous clock, environmental light, and neural network signaling. Moreover, these data reveal a bifurcation of PKA-dependent clock effects and PKA-independent output effects. Taken together, our results provide a molecular and cellular basis for the dual functions of PDF in clock resetting and pacemaker output.
| Circadian clocks provide a mechanism for predicting and adapting behavioral and physiological processes to 24-hour rhythms in the environment. In animal nervous systems, cell-autonomous molecular oscillators are coupled via neural networks that control daily patterns of activity. A major neuropeptide synchronizing neural oscillators in the Drosophila clock network is PIGMENT DISPERSING FACTOR (PDF). Here we identify a fork in the processing of the PDF signal in circadian neurons to independently reset the molecular clock and regulate neuronal activity. We show that the cAMP-activated protein kinase A (PKA) in circadian neurons is necessary and sufficient for many PDF-dependent behaviors. In addition, we find that a PDF>PDF receptor>PKA pathway targets the clock component TIMELESS to control molecular oscillators, and that this process may be influenced by rhythmic expression of PKA. We show that this pathway splits at the level of cAMP generation, with PDF and cAMP acutely increasing the activity of clock neurons in a PKA-independent manner. Thus, PDF operates via dual signaling pathways: one via PKA to reset clocks and the other via cAMP to acutely control activity. These results have broad implications given the conserved involvement of neuropeptide signaling in synchronizing clocks in circadian neural networks.
| Circadian clocks endow organisms with the ability to predict and respond adaptively to daily changes in the environment. In many taxa, these clocks consist of cell-autonomous molecular feedback loops, producing ∼24-hour oscillations at the mRNA and protein levels. In insects and mammals these clocks are also connected in neural networks that stabilize and synchronize these molecular feedback loops and communicate timing information to regulate daily behavior. How network and cell-autonomous mechanisms collaborate to produce robust circadian rhythms remains a major question.
In Drosophila, the molecular circadian clock consists of a set of interlocked transcriptional feedback loops in which the basic helix-loop-helix per-arnt-sim (bHLH-PAS) domain transcription factor CLOCK (CLK) forms a heterodimer with CYCLE (CYC) and binds E-boxes in the promoter regions of period (per), timeless (tim), vrille (vri), Par-domain protein 1ε (Pdp1ε) and clockwork orange (cwo), promoting their transcription (reviewed in [1]). PDP1ε and VRI feed back to regulate the Clk and cryptochrome (cry) promoters [2],[3], while CWO feeds back to regulate CLK/CYC activation at E-boxes [4]–[7]. PER and TIM proteins dimerize in the cytosol and are each required for their subsequent localization to the nucleus where PER inhibits CLK/CYC–mediated activation [8]–[14]. The CRY photoreceptor mediates light resetting via TIM degradation [15]–[18]. Clock function is evident as 24-h oscillations in the mRNA and protein levels of most of these clock components. The activity, stability, and subcellular localization of these proteins are largely controlled post-translationally by daily phosphorylation rhythms and subsequently by ubiquitin/proteasome dependent degradation [17],[19]–[25]. In contrast to transcriptional regulators, significant oscillations have not been described for these post-translational regulators with the exception of the PP2A subunits tws and wdb [26].
In insects and mammals, intercellular signaling among pacemaker neurons in neural networks has been found to be critical for synchronizing molecular clocks. The Drosophila pacemaker network is comprised of ∼150 neurons of which specific subgroups regulate discrete aspects of behavior in light-dark (LD) and constant darkness conditions (DD) [27]. Two of these groups—all but one sLNv and all lLNvs (small and large ventral-lateral neurons) express the neuropeptide PIGMENT DISPERSING FACTOR (PDF). The s-LNvs rhythmically express PDF in the dorsally projecting terminals that terminate near the DN1 [28]. Loss of function of either pdf or its receptor pdfr results in strongly reduced morning anticipation, an evening activity peak that is phase-advanced by 1 h relative to wild-type, and strongly reduced DD rhythmicity [29]–[41]. Ablation of PDF neurons results in similar phenotypes suggesting that PDF is the major transmitter of these neurons [37]. Transgenic rescue of pdfr mutants showed morning anticipation could be attributed to function in the DN1p neurons, while evening anticipation phenotypes mapped to non-PDF neurons, including the PDF(−) sLNv, the CRY+ subset of the LNd, and the DN1 [35],[41].
PDF coordinates molecular oscillations between disparate circadian pacemaker neurons and mediates pacemaker neuron output downstream of the clock. tim and cry mRNA oscillations in pacemaker neurons are damped in pdf01 mutants [42]. The timing of nuclear entry of PER protein in sLNv becomes phase-dispersed in DD in pdf01 flies [43]. PER expression in the LNd and DN1 cells is phase advanced on the first day of DD and subsequently damps [34],[43]. Analysis of TIM protein levels and a PER-luciferase fusion reporter suggested that the clocks in the different cell groups in the network can both advance or delay in response to PDF signaling [44].
Interestingly, while pdfr mutants exhibit notably reduced morning anticipation, PER oscillations in sLNvs and DN1s are comparable to wild-type flies under LD conditions [35], suggesting that PDF also mediates pacemaker neuron output in clock target neurons. Acute activation and silencing of neuronal activity observed by PDF injection in cockroaches are consistent with this latter mechanism [45]. However, the underlying signaling pathways mediating these dual PDF functions within non-PDF clock neurons, i.e., clock resetting and neural output, have not been identified.
PDFR is expressed in most of the pacemaker neurons and most of these neurons respond to PDF application in ex vivo preparations with increases in cAMP levels [46],[47]. We tested the function of the cAMP-dependent protein kinase A (PKA) in clock neuron-driven behavior. cAMP is the canonical activator of PKA activity. cAMP binds the PKA regulatory (R) subunit releasing the catalytic (C) subunit to phosphorylate substrates (reviewed in [48]). While PKA signaling has been implicated in circadian clock function [49],[50], its precise role in mediating PDFR signaling has not been defined.
Using the Gal4-UAS system, we expressed a type I regulatory subunit of PKA that is defective for cAMP binding (U-PKA-R1dn), thereby rendering endogenous catalytic subunits insensitive to cAMP activation [51],[52]. Initially we drove expression using circadian drivers, such as cwo-G4 (also known as c632a; [5]) and examined behavior under standard 12:12LD conditions. cwo-G4 is a GAL4 enhancer trap insertion just upstream of the transcription start site of core clock gene cwo and drives expression in clock neurons [5]. In addition to expression in the PDF clock neurons [5], we examined cwo-G4 driven nuclear green fluorescent protein (GFP) expression in pacemaker neurons using PER co-labeling. We observed GFP expression mainly in the circadian pacemaker neurons, with limited expression in other brain regions (Figure S1; unpublished data). Among pacemaker neurons, we observed GFP in non-PDF clock neurons, including all of the LNds and the PDF(−) s-LNv, as well as a number of DN1s and DN3s (Figure S1). Broad expression of PKA-R1dn with cwo-G4 (Figure 1C) was able to phenocopy many features seen in pdfr mutants (Figure 1D). These flies exhibited reduced morning anticipation and phase advanced evening activity under LD cycles, and were nearly arrhythmic in constant darkness, thus mimicking the three canonical behavioral phenotypes of pdf and pdfr mutants (Figure 1A–1E; Tables 1 and 2).
Given that the majority of PDFR functions mapped to non-PDF clock neurons, we then asked whether these PKA-R1dn effects were due to its expression in non-PDF neurons by blocking PDF neuron expression with pdf-G80. Importantly, we did not detect cwo-G4 driven GFP expression in the PDF(+) neurons, verifying effective inhibition of cwo-G4 activity in those neurons by pdf-G80 (Figure S1). We observed once again that these flies behaved similarly to pdf and pdfr mutants, exhibiting the three hallmark phenotypes of reduced morning anticipation, phase-advanced evening activity, and reduced rhythmicity in constant darkness (Figure 1F–1J; Table 1), providing strong evidence for PKA function in non-PDF neurons in mediating PDF effects in circadian neurons. The addition of pdf-G80 did improve the rhythmicity (power-significance [P-S]) in constant darkness, suggesting that PKA activity in the PDF neurons contributes to a portion of this phenotype (Table 2, p<0.005). Nonetheless, the morning and evening phenotypes continue to map to non-PDF neurons.
Given that cwo-G4 also drives limited expression in some non-circadian areas we decided to examine PKA function using an independent circadian driver. Previous work from our laboratory demonstrated that morning and evening anticipation phenotypes of pdfr mutant behavior can be rescued by expressing U-pdfr in non-PDF neurons using cry13-G4 and pdf-G80 [35]. Here we expressed U-PKA-R1dn in PDF(−) circadian cells using this Gal4/Gal80 combination and found that inhibition of PKA activity also results in morning and evening anticipation similar to pdfr- (Figure 2A, 2B, 2D, 2E; Table 1). If PDFR and PKA are operating in the same pathway, we would expect that U-PKA-R1dn expression in a pdfrhan5304 (pdfr-) mutant background would exhibit phenotypes comparable to either PKA-R1dn or pdfr mutants alone. Indeed, we found no differences in morning or evening behaviors among pdfr-, PKA-R1dn expression, or PKA-R1dn expression in the pdfr- background using cry13-G4/pdf-G80 (Figure 2B, 2C; Table 1), suggesting that PDFR and PKA operate in a common pathway within these cells.
Rescue or suppression of mutant receptor function by expression of activated downstream signaling components is a powerful in vivo method to elucidate signaling pathways [53]. We also attempted to suppress pdfr mutant phenotypes by expressing a PKA catalytic subunit (U-PKA-mC*), which is defective for regulatory subunit binding and is therefore constitutively active [52]. We expressed PKA-mC* in PDF(−) circadian neurons in pdfr mutants and observed that it rescued the reduced morning anticipation and phase-advanced evening activity onset characteristic of pdfr mutants (Figure 2; Table 1). There were no notable effects of PKA-mC* expression in a wild-type background in LD (Figure 2F–2J; Table 1). While PKA-mC* rescued LD phenotypes, we did not observe rescue of DD rhythms, suggesting that PKA-mC* is not expressed at the appropriate levels, cAMP inducibility may be required, and/or that non-PKA signaling pathways may contribute to PDFR function in DD (Table 2). We obtained similar results in the absence of pdf-G80, thus allowing PKA-mC* expression in the PDF neurons, suggesting that lack of PKA activity in PDF neurons is not responsible for the lack of DD rhythm rescue by PKA-mC* (Table 2). Taken together, our genetic evidence, especially our ability to bypass PDFR function with an activated form of PKA, indicate that PKA is a major mediator of PDFR signaling in non-PDF clock neurons. PKA activity is both necessary and sufficient for the execution of most PDFR-mediated behaviors.
We have previously shown that we can rescue pdfr mutant morning anticipation and DD rhythmicity, but not evening anticipation, by expressing wild-type pdfr only in DN1p circadian neurons using the Clk4.1-G4 driver [41]. To test whether PKA also functions in the DN1p we expressed PKA-R1dn using Clk4.1-G4 and found modestly reduced morning anticipation and DD rhythmicity, but no change in evening anticipation, complementing our observations for pdfr rescue (Figure 3A–3D; Tables 1 and 2) [35],[41]. Given that cwo-G4/pdf-G80 driven PKA-RIdn exhibited more robust morning and DD phenotypes (Tables 1 and 2), we hypothesize that non-PDF, non-DN1p cells may also contribute to these phenotypes and/or this driver combination more strongly inhibits PKA within the DN1p than Clk4.1-G4. To address the neural substrates of PKA function in evening anticipation, we used the mai179-G4 driver in combination with pdf-G80, which drives expression in the single PDF(−) sLNv, and the CRY(+) subset of LNd with variable expression in a 1–2 DN1s [30],[54],[55]. Here we found that PKA inhibition phase advances evening anticipation similarly to pdfr mutants (Figure S2). In keeping with a model in which the CRY+ LNd and fifth PDF(−) sLNv selectively regulate evening anticipation, PKA-R1dn driven by mai179-G4/pdf-G80 had no effect on morning anticipation (Figure S2; Table 1) or DD rhythms (Table 2).
Rhythms in PDF levels are apparent in the terminals of LNv neurons and rhythmic PDF release is thought to contribute to the temporal encoding of the PDF signal; however, rhythmic PDF may not be necessary for rhythmic behavior or clock function [28],[40],[56]. We hypothesized that rhythmic control of signal processing within cells receiving the PDF signal may contribute to the robustness of this pathway. To determine whether PKA subunit transcripts are under circadian clock control in DN1p clock neurons, we expressed UAS-membrane GFP (U-mGFP) using Clk4.1-G4 and isolated these neurons by fluorescence-activated cell sorting (FACS). After RNA isolation and linear amplification (see Materials and Methods), we examined the transcript levels of the three catalytic PKA subunits (C1, C2, and C3) and two regulatory subunits (R1 and R2) by quantitative real-time (RT)-PCR. Transcript levels of three PKA subunits (PKA-C1, PKA-R1, and PKA-R2) oscillate in phase, with coincident peaks in the mid-day (ZT4) (Figure 3E–3G). Peak transcript levels were reduced and rhythms were not detected in the per01 mutants consistent with circadian clock control. The two other PKA transcripts (C2, C3) were near the limits of quantitative detection. Thus, not only is PKA activity likely controlled via rhythmic inputs of PDF-driven cAMP production but PKA is also rhythmically controlled at the level of gene expression. We hypothesize that these dual mechanisms collaborate to provide a robust time-of-day PKA signal to synchronize non-PDF to PDF oscillators.
The ability of PDF neurons to reset molecular clocks in non-PDF neurons has been powerfully demonstrated by selective manipulation of circadian period in PDF neurons and examination of molecular oscillations in non-clock cells [39],[41],[55]. To determine the direct molecular targets of PDF neuron signaling, we rescued clock function selectively in PDF neurons of arrhythmic per01 mutants and assayed molecular oscillations in per-less non-PDF neurons. By examining molecular changes in per01 non-PDF neurons, we removed the possibility that identified changes would be indirect through a functioning circadian clock and/or per. In addition, we examined molecular changes on the first day of constant darkness, removing the possibility that light signaling via CRY is responsible for any changes. We rescued per in PDF cells (per01;pdf-G4/+;U-per16/+; “pdfPER” flies) and examined non-PDF-expressing circadian cells (LNd and DN1) in brains stained for the clock components TIM (Figure 4) and PDP1ε (Figure 5). First we demonstrated that transgenically supplied PER cycles in the sLNvs and rescues oscillations in TIM levels and nuclear localization in those cells (CT24), indicating at least a partially rescued sLNv clock consistent with prior studies (Figure 4A) [55].
We then examined the consequences of the rescued sLNv clock on non-PDF per01 DN1 and LNd neurons. Consistent with the fact that these cells lack per or a fully functioning circadian clock, TIM is predominantly expressed in the cytoplasm at all times of day (Figure 4) [13]. However we found stark changes in the levels and phase of TIM oscillation in DN1 neurons despite the lack of PER. In per01 controls, we observed a low amplitude TIM oscillation with an inappropriate day-time peak (Figure 4). In pdfPER rescue flies, TIM cycling amplitude increases with elevated peak levels and its oscillation phase is synchonized with that of TIM in the PDF neurons (Figure 4). TIM levels and phase in LNds are also modified by the PDF-neuron clock, but these effects are much smaller than those in the DN1s (Figure 4), consistent with our prior finding that the DN1 are more strongly reset by PDF neurons than the LNd [41]. To determine if these effects on TIM are specific, we also assayed a second clock component PDP1ε, a core circadian transcription factor directly activated by CLK/CYC [3]. We found that PDP1ε exhibits comparable levels between per01 and pdfPER rescue flies with only a modest change at ZT18 in the LNd (Figure 5). The absence of strong effects on PDP1ε suggests that PDF-mediated inputs to the molecular clock are unique to TIM. This result also argues strongly against large PDF effects on CLK/CYC driven transcription, a major determinant of PDP1ε levels [3]. The strength of the effects on TIM evident in the absence of a functioning clock, per and light, suggest that TIM is a direct target of PDF signaling.
Our results suggest PKA mediates PDF effects and that PDF targets TIM. To test whether PKA can influence TIM levels, we expressed PKA-R1dn in non-PDF circadian cells using cwo-G4 in combination with pdf-G80 in a per01 mutant background and examined TIM levels (Figure 6). Flies were entrained and dissected at four time points across the LD cycle, and stained for TIM. We observe reduced peak levels of TIM in both the LNd and DN1 at ZT24 with a non-significant trend developing by ZT12 during the light period. In these per01 flies, TIM remains in the cytoplasm throughout the 24-h cycle. These results indicate that PKA activity positively regulates TIM accumulation in the LNds and DN1s in the absence of a functioning clock suggesting a direct effect on TIM.
We have shown that the PDF-cell clock specifically regulates TIM in non-PDF circadian cells (Figures 4 and 5) and that reducing PKA activity in non-PDF cells leads to reduced TIM levels (Figure 6). Our behavior data show that PKA acts in the PDF signaling pathway downstream of PDFR (Figures 1 and 2). We therefore determined whether loss of PDF would mimic reduced PKA activity and result in reduced TIM. Here we examined loss of PDF (pdf01) in an arrhythmic per01 mutant background and examined the effects of the PDF peptide on TIM at the end of DD1 (CT24). We compared per01 and per01;;pdf01 flies and observed that TIM is strongly reduced in the absence of PDF in both LNd and DN1 neurons (Figure 7). We observe a nearly 50% reduction in TIM staining intensity in the absence of PDF, which is comparable to or even larger than the effect than we observed with PKA inhibition. One possibility is that expression of PKA-R1dn may not completely interrupt the signaling cascade, whereas pdf01 is a confirmed null mutation [37]. These results provide further independent support for the hypothesis that the PDF>PDFR>PKA signaling pathway influences the core molecular clock by promoting the accumulation or stability of TIM.
Reduced morning anticipation in pdfr mutants coupled to an absence of significant core clock effects in LD suggested that PDF morning function is via effects on neuronal output [35]. We hypothesized that these effects may be mediated by direct effects on neuronal activity. Our previous work had identified the DN1p as functional targets of PDF on morning anticipation using rescue of pdfr mutants [41]. To determine if this reflects a direct interaction, we selectively labeled the DN1p using U-GFP in combination with Clk4.1-G4 and the LNv using anti-PDF and found that the arbors of each extensively co-mingle (Figure 8A). To test if DN1p neurons are direct targets of PDF neurons, we employed GFP reconstitution across synaptic partners (GRASP) [57]. Here we expressed one fragment of GFP (GFP11) on the extracellular surface of the LNv neurons using pdf-LexA and its complementary fragment (GFP1-10) on the extracellular surface of the DN1p neurons using Clk4.1-G4. We observe robust fluorescence in the dorsal terminals consistent with extensive physical contacts (Figures 8B, 8C, and S4). Co-labeling of sLNv-DN1p GRASP with PDF finds that PDF signal is in close proximity to GRASP signals suggesting that sites of physical contact are potential release sites for PDF (Figure 8B and 8C).
To examine PDF signaling mechanisms that control neuronal output, we first performed live imaging on the DN1p neurons on explanted brains. Using the Clk4.1-G4 driver in combination with the FRET sensor U-Epac1(50A) [58], we measured the variation of [cAMP] following focal PDF application specifically in the DN1p neurons. Prior studies had used bath application of PDF to examine changes in cAMP and thus, the observed effects could be due to indirect activation [47]. Following focal PDF application to the DN1p neurons (Figure S3A), we observed a decrease in the ratio YFP/CFP indicating an increase of [cAMP] (Figure S3B–S3D). Thus, as suggested by prior studies [47], we confirm that PDF increases cAMP levels in the DN1p.
To resolve whether PDF acutely controls DN1p neuronal activity, we developed patch clamp electrophysiology of the DN1p subset of pacemaker neurons and assayed the response of these cells to focal PDF application. We performed cell-attached recordings in combination with live calcium imaging on the DN1ps by simultaneously recording firing frequency and [Ca2+]i using Clk4.1-G4 driving expression of the GCaMP6f calcium indicator [59]. Focal PDF application acutely stimulates the DN1ps by increasing the instant firing frequency of the neurons (Figure 9A and B). This increase in neuronal activity is directly correlated with an increase in [Ca2+]i as measured by the GCaMP6f calcium indicator (Figure 9C).
Using whole-cell current clamp recordings, we found that PDF both acutely depolarizes and increases action potential firing rates (Figure 10A). This excitatory effect of PDF is dependent on its receptor as we could not detect any effect of PDF on the membrane potential or firing frequency in cells lacking PDFR (Figure 10A). The PDF-evoked depolarization was present after blocking action potential firing, and thus most synaptic transmission, with the voltage gated sodium channel blocker TTX indicating that PDF acts on the DN1ps directly (Figure 10B). Surprisingly the PKA inhibitor H89 did not block these effects indicating that PDF activates a PKA-independent pathway to acutely activate neurons (Figure 10C). To independently confirm the dispensability of PKA signaling we recorded from DN1p neurons expressing the dominant-negative PKA-R1 (Figure 10D and 10E). PDF application still depolarizes and increases in firing frequency further supporting the hypothesis that PDF acts independently of PKA activation to regulate membrane excitability.
We next examined the potential role of cAMP as the intracellular component mediating the acute PDF effect on membrane activity. First, we demonstrated that the adenylate cyclase inhibitor MANT-GTPγS blocks the PDF induced excitation (Figure 11A–11C). Conversely, forskolin (an adenylate cyclase activator) and direct dialysis of cAMP into the cell induces a depolarization similar to the PDF evoked response (Figure 11D and 11E). Finally, the cAMP induced depolarization and activation was present in the neurons expressing PKA-R1dn (Figure 11E). Taken together these data indicate that cAMP, rather than other upstream signaling components is responsible for the PDF effects on excitability.
In voltage clamp mode, PDF application acutely induces an inward current at negative potentials and a positive shift in the reversal potential (Figure 12A–12C). This inward current is TTX-insensitive (Figure 12C) and is attenuated after reduction of extracellular sodium (Figure 12F). Furthermore, focal application of the adenylate cyclase activator forskolin or direct intracellular dialysis of cAMP induce an inward current like PDF (Figure 12D and 12E). The properties of the observed current—neuropeptide induction, TTX resistance as well as PKA independence—are consistent with a cyclic-nucleotide-gated channel (CNG) [60]. Thus, using this novel patch clamp analysis and focal application of PDF, we demonstrate that PDF acts as an excitatory neurotransmitter that acutely increases firing rate and calcium, likely in a PKA-independent manner. The rapidity of PDF effects on excitablility argues strongly that they are direct and not via clock resetting.
Intercellular communication has emerged as a critical element in circadian pacemaker function in multicellular animals. PDF acts as a master neural network regulator coordinating molecular oscillations between disparate circadian pacemaker neurons in Drosophila. Yet how PDF signaling resets circadian clocks as well as acutely regulates neural activity has not been clearly defined. Here we provide evidence that the PDF signaling pathway works through two mechanisms to regulate circadian behavior; a clock resetting pathway that targets the core clock protein TIMELESS (TIM) via PKA to maintain synchronous molecular oscillations throughout the pacemaker network, and a neural activity pathway that acutely increases the firing rate of pacemaker neurons independent of PKA (Figure 13).
Here we provide in vivo genetic evidence for a role for PKA in mediating PDF neuropeptide effects on behavior, including demonstration of clock control of PKA subunits in PDF target neurons. While prior work had demonstrated a role for PKA in circadian behavior, these studies observed effects under conditions of PKA overexpression in mutant conditions [61], failed to link PKA to PDF receptor signaling [49],[61]–[64], or impaired cAMP or PKA throughout the fly [49],[63] or the circadian network [62],[65]. We show that inhibition of PKA in a subset of non-PDF neurons can mimic the advanced circadian activity in the evening observed in PDF or PDF receptor mutants. Moreover, expression of an activated form of PKA can rescue most pdfr mutant phenotypes, providing powerful genetic evidence that PKA is mediating PDFR signaling in non-PDF circadian pacemaker neurons (Figure 13).
Using refined cell-specific manipulations, we dissected the functional neuroanatomy of PKA function. These studies demonstrated DN1p PKA contributes to morning anticipation and DD rhythms and PDF(−) sLNv and CRY+ LNd functions in evening anticipation, similar to the division of labor we previously observed for pdfr rescue [35],[41]. Nonetheless, the finding of modest effects of DN1p PKA on morning anticipation in the face of a prominent role for non-PDF cells (Table 1) suggest that other non-PDF cells make a contribution and/or that the DN1p function is via PDF effects on neuronal excitability upstream of PKA. While a previous study linked a PDF-coupled adenylate cyclase (AC3) function in the PDF neurons to morning anticipation in sLNv, it is not known if this adenylate cyclase may also couple to other receptors that may mediate these effects, for example, by regulating PDF release rather than PDF receptor signaling [31]. We did observe that PDF neurons also contribute to DD rhythmicity effects of PKA (Table 2), consistent with the more distributed function of PDFR in DD rhythmicity. Taken together, these data provide a circuit map for PKA function in mediating PDF effects in the central pacemaker network.
A central feature of core circadian clock components is their time-of-day dependent expression, providing the mechanistic basis of biological timekeeping. To address whether the clock actively controls PKA expression or activity, we assessed transcript levels using FACS isolation of the DN1p. Here we show that both regulatory subunits (R1 and R2) and one of the three catalytic subunits (C1) of PKA show robust rhythms with a peak during the mid-day. Coordinate oscillations of both regulatory and catalytic subunits of PKA should result in daily increases in the sensitivity to PDFR activation. Rhythmic PKA expression also provides a mechanistic basis for rhythmic behavior under conditions when PDF oscillations are not apparent [29],[56]. PKA rhythms are abolished in mutants of the classical core clock component per01, indicating these oscillations are clock controlled. Given that these PKA transcripts peak in the mid day (ZT4; Figure 3E–3G) at a time when CLK activity is low [66] and that peak PKA transcript levels are reduced in the per01 mutant suggest that it is not directly CLK-activated. Interestingly, expression of a bacterial sodium channel in the larval sLNv can induce PKA-C1 transcript expression [62], suggesting that clock-driven changes in neuronal activity may mediate PKA transcript rhythms. While the peak of PKA subunit transcription in the DN1p in the mid day is not coincident with the requirement of PDFR signaling for morning anticipatory behavior, the rate of accumulation and half-life of PKA protein in these cells is not known. Regardless, these results indicate that pacemaker neurons rhythmically control their sensitivity to PDF inputs, suggesting that rhythmic PDF-driven cAMP production and rhythmic PKA transcription collaborate to provide a robust time-of-day specific signal to synchronize non-PDF to PDF oscillators (Figure 13).
In addition to demonstrating a key role for PKA in PDFR signaling, we also reveal important in vivo evidence that PDF neuronal signaling selectively targets the circadian clock component TIM in non-PDF neurons, providing a molecular basis for network influence on core molecular clocks (Figure 13). To address the core clock target of PDF signaling, we set up a complex genetic scenario in which we used the arrhythmic per01 mutant as a background and rescued per (and thus, clock function) only in PDF neurons. We then asked how rescued clock function in PDF neurons impacts molecular clock components in per01 non-PDF clock neurons. We found that these rhythmic PDF neurons are able to drive high amplitude and appropriately phased DN1 molecular oscillations in the core clock component TIM but not in another clock component PDP1ε, suggesting that TIM is the specific target of PDF signaling in the molecular clockwork. We observe reduced effects on TIM in the LNd, perhaps due to the fact that the rescue of per01 in PDF neurons is not complete (Figure 4) [55]. Moreover, it is unlikely that PDF is inducing an intact clock in per01 non-PDF neurons. In addition to the weak or absent PDP1ε oscillations, clock-driven oscillations in TIM nuclear localization are not observed with TIM remaining cytoplasmic, consistent with studies indicating that PER is required for TIM nuclear localization [10],[13]. Loss of PDF in a per01 background results in reduced TIM levels in both the LNd and DN1 (Figure 7). These robust TIM effects in the LNd may reflect a more extreme perturbation of PDF signaling in the null mutant and/or the potential of non-PDF transmitters to influence LNd TIM levels in the PDF cell rescue. Nonetheless, our data support the view that PDF signaling is specifically regulating TIM rather than reconstituting a clock in per01 target neurons.
Our work suggests that PKA is an important intermediary between PDF and TIM (Figure 13). Inhibition of PKA in non-PDF neurons in per01 mutants reduces TIM levels in LNds and DN1s (Figure 6), consistent with a role for PKA in promoting TIM accumulation or stability. The more strongly evident effects of PKA on TIM than in the PDF cell rescue context may reflect the incomplete PDF cell rescue, more robust PKA manipulation with dominant negative expression, and/or PDF or PKA-independent effects of PDF cells on LNd TIM levels. PKA effects on TIM in the absence of per or a fully functioning clock suggest that these effects are direct. PKA has also been implicated in activating CLK driven transcription. However, these effects are modest and observed under conditions of PKA overexpression in cultured S2 cells [67]. Moreover, PKA does not phosphorylate CLK in vitro [67]. Our finding that PDP1ε in non-PDF cells is not strongly affected by rescue of the molecular clock in PDF neurons further supports the hypothesis that CLK activity is not an in vivo target of PDF/PKA. TIM contains numerous consensus PKA phosphorylation sites and is robustly phosphorylated by PKA in vitro [68]. Our findings that reduction of PKA function reduces TIM levels suggests a positive role for PKA in TIM accumulation or stability. Our finding that TIM levels in LNd and DN1 neurons are reduced in the absence of PDF peptide provides strong independent verification for the PDF>PDFR>PKA pathway in targeting TIM to influence the core molecular clock. Notably, we did not observe any effects on TIM in DD as assayed by Western blot after PKA inhibition in the eye (unpublished data), suggesting PKA pathway function in the core clock may be restricted to the pacemaker neurons. Comparable changes in TIM due to light pulses are associated with significant phase shifts [69] that are comparable to, or even exceed, those evening phase effects observed in pdf mutants or with PKA-R1dn expression, suggesting that these TIM effects are biologically meaningful.
The finding that TIM responds to PDF and PKA could explain observed interactions between PDF and CRY signaling. Altering the pace of PDF-cell clocks can reset non-PDF clocks. However under standard LD conditions, PDF-cell clocks are only able to reset evening phase after mutation of the CRY photoreceptor, indicating that CRY antagonizes PDFR signaling [70]. Our identification of TIM as a common target of CRY and PDFR signaling provides a plausible mechanism for these phenotypes: CRY-mediated degradation of TIM may render pacemaker neurons insensitive to PDF receptor inputs thus explaining the CRY-dependence of PDF effects on evening phase.
Thus, TIM is a multimodal integrator of core clock, environmental, and network pathways of entraining and maintaining clocks in the pacemaker network: (1) tim is transcribed by the CLK/CYC heterodimer and is thus regulated directly by the core feedback loop; (2) TIM protein levels are controlled by environmental light via CRY-mediated degradation; and (3) we demonstrate that TIM responds to network signals via PDF signaling, likely directly mediated post-transcriptionally by PKA.
In addition to elucidating signaling mechanisms that link PDF to core clocks, we also defined mechanisms by which PDF acutely regulates neuronal activity. While our work suggests that PDF acts via changes in protein abundance to reset clocks, our previous work suggested that PDF also has effects on pacemaker neuron output, specifically morning behavior, that are independent of resetting clocks [35]. In fact, PDF injection in cockroaches acutely regulates neuronal activity [45]. However, the precise nature and mechanism by which PDF achieves these effects are not clear. Here we have developed patch clamp electrophysiology of the DN1p subset of neurons and assayed the response of these cells to focal PDF application. We focally applied PDF to these neurons and found that PDF both acutely depolarizes and increases action potential firing rates in a PDFR dependent manner, indicating that PDF is acutely excitatory and providing a mechanistic basis for effects on pacemaker neuron output (Figure 9A). Consistent with our data in the DN1p, membrane-tethered PDF peptide expressed in the PDF+ LNv depolarizes the sLNv [29]. Surprisingly PKA inhibiton (by H89 or the expression of a dominant negative PKA) did not block these effects, while adenylate cyclase inhibition did block them, indicating that PDF activates a cAMP-dependent, PKA-independent pathway to acutely activate neurons (Figures 10B–10E and 11A–11E). We note that genetic inhibition of PKA in the DN1ps only modestly reduces morning anticipation, suggesting a potential role for this PKA-independent pathway in morning behavior (Figure 3B and 3C). Given the properties of the PDF-induced current we hypothesize that PDF-driven cAMP activates a cyclic nucleotide gated channel to acutely depolarize and activate target neurons (Figure 13). Our model is consistent with the role of G-alpha-s and cAMP in mediating PDF effects in the sLNv on morning and evening activity allocation [29]. However, the role of PKA was not examined this study.
While we cannot exclude the possibility of direct or indirect cross talk between pathways, these data reveal a bifurcation of the PDF receptor signaling pathway: a PKA-dependent fork contributes to synchronization of the molecular clocks via regulation of TIM and a PKA-independent fork acutely induces neuronal activity (Figure 13), thus providing mechanistic bases for the dual functions of PDF in the Drosophila circadian pacemaker network.
Fly lines carrying U-PKA-mC* and U-PKA-R1dn (also known as BDK33) were a generous gift from Daniel Kalderon [52]. UAS-CD4::spGFP1-10 and LexAop-CD4:spGFP11 flies were the gift of Kristen Scott [71]. The latter transgene was recombined with pdf-LexA (a gift of Michael Rosbash [72]). Lines carrying combinations of these and other transgenes or mutants were constructed using standard genetic crosses. Tim was genotyped for the s/ls alternative start site polymorphism using previously described primers (Peschel 2004). TIM staining in the per01;pdf-G4;U-per16 rescue context was completed three times: in two trials the tim genotypes were s/ls, and in one trial the control was ls/ls, while the pdfPER flies were s/ls. The staining results were comparable between these conditions and were combined. Strains for TIM staining with PKA-R1dn expression (Figure 6) were s/ls.
Fly behavior was recorded using the Drosophila Activity Monitoring system (Trikinetics) and analyzed using ClockLab and the Counting Macro as described [73]. Briefly, male flies were fed on 5% sucrose-agar medium in 5LD7DD conditions at 25C. LD eductions were obtained using averaged data in 30-minute bins across days 2–5 of the behavior run. DD period and rhythmicity data were calculated in ClockLab with period measurements taken only from flies in which the Power-Significance (P-S) ≥10.
Morning anticipation was calculated using a variant of the method described in [32]. Activity from each of four days of LD behavior recorded for each individual fly were analyzed such that the morning index (MI) = ((total activity 3 h prior to lights-on)/(total activity 6 h prior to lights-on)) − (0.5). 0.5 was subtracted so flat activity over the six hours analyzed is equal to 0. In cases where no activity counts occurred in the 6 hours before lights-on, resulting in an undefined 0/0, the ratio was set to 0.5, indicating no change in activity over that time period.
The timing of evening activity onset was calculated as previously described [35] with the onset time defined as the first time point in the four 30-min bin sliding window with the largest increase in activity prior to lights-off.
Genotypes were compared by Student's two-tailed t-test.
Flies to be stained were entrained for five to seven 12-h light, 12-h dark (LD) cycles at 25°C and either dissected and fixed at the indicated timepoint for LD staining (Figure 6) or transferred to constant darkness and dissected and fixed for DD1 staining (Figures 4 and 5). Brains were dissected in PBS (pH 7.5) and fixed in 3.7% formaldehyde in PBS for 1 h shaking at room temperature. Brains were then washed 3× in PBS and primary antibody solution was added. Guinea pig anti-TIM (1∶2,000), rabbit anti-PER (1∶16,000), and mouse anti-PDF (1∶500) (Developmental Studies Hybridoma Bank) were incubated overnight shaking at 4°C in a solution of PBS, 10% goat normal serum (GNS), and 0.3% Triton X-100. For stains involving rabbit anti-PDP (1∶200) brains were dissected and fixed as above, except after fixation and 3× washes with PBS, brains were subject to a 1-h permeablization in PBS +1% Triton X-100 and primary antibody solution was incubated for 3 days in PBS with 0.3% Triton X-100 and 10% GNS. After the primary incubation, for all stains, brains were washed 3× in PBS +0.3% Triton X-100 and secondary antibodies (for PDF, PER, TIM staining: anti-mouse Alexa647, anti-guinea-pig Alexa 488, anti-rabbit Alexa 594; for PDF, PDP1ε staining: anti-mouse Alexa 594, anti-rabbit Alexa 488) (all dyes from Molecular Probes - Invitrogen) were each added at 1∶500.
Brain images were taken on a Nikon E800 laser-scanning confocal microscope using a 60× A 1.40 N.A. objective with laser, filter, and gain settings remaining constant within each experiment. Channels were scanned sequentially. Confocal Z-stacks were analyzed in NIH ImageJ software. Intensity measurements were taken from single confocal sections at approximately the middle of each cell. Nearby areas of similar area to the cells being measured were selected for each cell group in each hemisphere as a measurement of background staining. The background measurement for each cell group in each hemisphere was subtracted from the intensity measurement for each cell in that group. Background-subtracted values were then averaged across all brains in that experiment. Image measurements were normalized prior to combining data from independent experiments. For Figures 4 and 5, each experiment was individually normalized such that pdfPER Rescue at CT6 = 1. For Figure 6, data were normalized to Control at ZT6 = 1. For Figure 7, data were normalized to per01 at CT24 = 1. Data from independent experiments were combined post-normalization to obtain the final graphs. Images from Figures 4, 5, and 6 are displayed using the inverted 5 Ramps lookup table within ImageJ for ease of viewing images with low signal. Staining data were statistically analyzed by one-way ANOVA and Tukey's pairwise comparisons.
GRASP signal and mouse anti-PDF stained brains were fixed, mounted, and imaged as above, except using anti-mouse Alexa 594 to label PDF. Both 40× and 60× objective images were collected in 1 micron steps through the region containing staining, or through the entire dorsal brain for non-labeled parental control lines.
Cells were processed as described previously [74]. Before FACS cell sorting cells were filtered using 100 micron filter. Propidium iodide (Sigma, 130 ng/ul) was added to distinguish between dead and alive cells. Cells were sorted on Aria II FACS Cell Sorter (BD Biosciences) into an extraction buffer from the PicoPure RNA extraction kit (Arcturus). Transcripts were obtained from 40 to 45 brains (yielding 300–500 DN1p neurons) per time-point. Subsequently, the cells were lysed and stored at −80°C until RNA extraction as described previously [74].
Cells were processed as described previously [75]. cDNA from two independent replicates per genotype were analyzed per time-point on a BioRad CFX384 real-time PCR system. mRNA was quantified as described previously [74]. One-way ANOVA was used to determine statistically significant differences between time-points within each genotype (p<0.05). The following primers were used to examine pka expression: pka-R1, F primer, 5′-ACTTTGGCGAGATTGCTCTG-3′; R primer, 5′-CGGACAACGATACGAAACTG-3′; pka-R2, F primer, 5′-CTACGAACGCATGAATCTGG-3′; R primer, 5′-GCCGAAGTACTGTCCCTTGC-3′; pka-C1, F primer, 5′-ATCGCTGGCATCGTAGTCG-3′; R primer, 5′-AAGGCGCTTGGTTAAGACG-3′.
Brains from male adults Drosophila (7–14 days old) were removed from their heads in ice-cold recording solution. After removing the connective tissue, air sacs, and trachea with fine forceps, the brains were transferred to a recording chamber and were held ventral side down by a harp slice grid (ALA scientific). No enzymatic treatment was used to avoid altering ion channels function on the cell surface. Brains were allowed to rest in continuously flowing oxygenated saline (95% oxygen and 5% carbon dioxide) for at least 10 min and no more than 2 h before recording. Perfusion with oxygenated saline was continued throughout the recording period. Whole brain electrophysiology and imaging experiments were performed on an Ultima two-photon laser scanning microscope (Prairie Technologies) equipped with galvanometers driving a Coherent Chameleon laser. Fluorescence was detected with photomultiplier tube. Images were acquired with an upright Zeiss Axiovert microscope with a 40×0.9 numerical aperture water immersion objective at 512×512 pixel resolution and 1-µm steps. Current-clamp recordings were performed with pipettes (10–14 MΩ) filled with internal solution. To visualize the recorded cell, Alexa Fluor 594 biocytin (10 µM) was added into the intracellular solution. Recordings were made using Axopatch 200B patch-clamp amplifier, Digidata 1320 A, and pCLAMP software (Axon Instruments). The extracellular recording solution contains in mM: 101 NaCl, 1 CaCl2, 4 MgCl2, 3 KCl, 5 glucose, 1.25 NaH2PO4, and 20.7 NaHCO3 (pH 7.2, 250 mOsm). The internal solution contains in mM: 102 K-gluconate, 0.085 CaCl2 1.7, MgCl2, 17 NaCl, 0.94 EGTA, 8.5 HEPES, 4 Mg-ATP, 0.3 Tris-GTP, and 14 phosphocreatine (di-tris salt) (pH 7.2, 235 Osm). For simultaneous cell attached and live calcium-imaging recordings, the Drosophila DN1ps neurons were visualized with GCaMP6f indicator [59]. The x-y images of GCaMP6f fluorescence were acquired at 10–20 Hz. GCaMP6f fluorescence was excited at 840 nm and was captured at wavelengths between 490 and 540 nM using a bandpass filter.
Changes in intracellular cAMP concentration were imaged using the Epac1-cAMPs indicator. Whole brain imaging experiments were performed using hemolymph-like HL3 saline [76] (in mM: NaCl 70, KCl 5, CaCl2 1.5, MgCl2 20, NaHCO3 10, D-trehalose dihydrate 5, sucrose 115, Hepes 5, pH adjusted at 7.1 with NaOH 1 M). After dissection, whole brains were placed with HL3 solution in the experimental chamber (POC-R perfusion chamber, Zeiss) and placed on the stage of an Axiovert 200 M inverted microscope attached to a Zeiss 510 Meta/ConfoCor3 Laser Scanning unit (Zeiss) available through the Northwestern University Biological Imaging Facility. The x-y confocal images of Epac1-camps fluorescence were acquired at 2–4 Hz using a Zeiss planApochromat 20×0.8 N.A. objective. Epac1-camps fluorescence were excited at 454 nm by a 200 mW argon ion laser and were captured at wavelengths between 470 and 500 nM for CFP and between 510 and 550 nM for YFP using a bandpass filter. The pinhole was set to provide a confocal optical slice of 10 µm. Epac1-camps fluorescence intensity was normalized to the average fluorescence intensity in the images captured before neurotransmitter application and the ratio YFP/CFP was calculated.
PDF (50 µM, dissolved in recording solution, GenScript) or forskolin (Sigma) was applied focally for 10 s to the recorded cells via pressure ejection (0.5–1 psi) from a glass pipette (5–10 µM) placed in the vicinity of the cell. TTX (Tocris) and/or H89 (Sigma) were bath applied by exchanging the recording solution. cAMP (10 µM, Sigma) and MANT-GTPγS (500 nM, Sigma) were diluted into the intracellular solution.
|
10.1371/journal.ppat.1007578 | Oncogenic KSHV-encoded interferon regulatory factor upregulates HMGB2 and CMPK1 expression to promote cell invasion by disrupting a complex lncRNA-OIP5-AS1/miR-218-5p network | Kaposi’s sarcoma (KS), a highly disseminated tumor of hyperproliferative spindle endothelial cells, is the most common AIDS-associated malignancy caused by infection of Kaposi’s sarcoma-associated herpesvirus (KSHV). KSHV-encoded viral interferon regulatory factor 1 (vIRF1) is a viral oncogene but its role in KSHV-induced tumor invasiveness and motility remains unknown. Here, we report that vIRF1 promotes endothelial cell migration, invasion and proliferation by down-regulating miR-218-5p to relieve its suppression of downstream targets high mobility group box 2 (HMGB2) and cytidine/uridine monophosphate kinase 1 (CMPK1). Mechanistically, vIRF1 inhibits p53 function to increase the expression of DNA methyltransferase 1 (DNMT1) and DNA methylation of the promoter of pre-miR-218-1, a precursor of miR-218-5p, and increases the expression of a long non-coding RNA OIP5 antisense RNA 1 (lnc-OIP5-AS1), which acts as a competing endogenous RNA (ceRNA) of miR-218-5p to inhibit its function and reduce its stability. Moreover, lnc-OIP5-AS1 increases DNA methylation of the pre-miR-218-1 promoter. Finally, deletion of vIRF1 from the KSHV genome reduces the level of lnc-OIP5-AS1, increases the level of miR-218-5p, and inhibits KSHV-induced invasion. Together, these results define a novel complex lnc-OIP5-AS1/miR-218-5p network hijacked by vIRF1 to promote invasiveness and motility of KSHV-induced tumors.
| Kaposi's sarcoma-associated herpesvirus (KSHV) infection caused Kaposi’s sarcoma (KS), a highly disseminated tumor that frequently occurs in patients with AIDS. KSHV-encoded viral interferon regulatory factor 1 (vIRF1) is an oncogenic protein, which has been shown to be vital in KSHV evasion of innate antiviral response and induction of tumorigenesis but its role in KS tumor invasiveness and motility remains unclear. A growing volume of literatures has proposed that lncRNAs could function as tumor suppressors or oncogenes, and numerous lncRNAs might act as competing endogenous RNAs (ceRNAs) that competitively bind to microRNAs (miRNAs), hence exerting influence on posttranscriptional regulation. However, whether cellular lncRNAs are involved in the progression of KS is still unknown. Here, we revealed a previously undefined role of vIRF1 in cell motility and proliferation, and described the cross-regulatory network of cellular lncRNAs and miRNAs involved in the pathogenesis of KS. We found that the crosstalk between miR-218-5p and lnc-OIP5-AS1 contributed to vIRF1-induced cell motility and proliferation via increasing HMGB2 and CMPK1 expression. In summary, this study constitutes an important discovery related to KS pathogenesis, particularly in the invasiveness and motility of KS tumors.
| Kaposi’s sarcoma-associated herpesvirus (KSHV), also known as human herpesvirus 8 (HHV-8), is a double-stranded DNA virus, which belongs to γ-herpesvirus. KSHV was initially identified in an AIDS-associated Kaposi’s sarcoma (AIDS-KS) lesion, and has since been strongly linked to Kaposi’s sarcoma (KS), primary effusion lymphoma (PEL), a subset of multicentric Castleman’s disease (MCD), and KSHV-associated inflammatory cytokine syndrome (KICS) [1]. Like other herpesviruses, the life cycle of KSHV consists of two phases, latent and lytic phases, both of which contribute to KSHV-induced pathogenesis, tumorigenesis and angiogenesis [2, 3]. KSHV genome contains over 90 open reading frames [4], some of which are homologous to human genes. To establish a successful persistent infection, KSHV encodes these homologous proteins to regulate cell growth, immune response, inflammatory response and apoptosis, and thus escape the immune antiviral response of host cells [5]. Moreover, these homologous proteins are also in favor of KSHV-induced tumorigenesis. For examples viral interferon regulatory factors (vIRFs) [6], viral interleukin-6 (vIL-6) [7], viral G protein-coupled receptor (vGPCR) [8], viral Bcl-2 (vBcl-2) [9], viral FLICE inhibitory protein (vFLIP) [10] and viral cyclin (vCyclin) [11] have been shown to be pro-oncogenic or promote tumorigenesis.
The cellular IRFs (IRFs 1~9) are a family of cellular transcription proteins that regulate the expression of interferon and interferon-stimulating genes (ISGs) in innate immune response, among which IRF3 and IRF7 play key roles in the induction and secretion of type I interferon [12]. vIRF1 (449 amino acids), as one of the KSHV vIRFs (vIRF1 to vIRF4), is encoded by KSHV ORF-K9, which has 26.6% and 26.2% of protein homology to cellular IRF3 and IRF7, respectively [13]. vIRF1 has been shown to compete with IRF3 to interact with CBP/p300 coactivators by blocking the formation of CBP/p300-IRF3 complexes, thereby inhibiting IRF3-mediated transcription and signal transduction of type I interferon [14]. However, vIRF1 could not block IRF-7-mediated transactivation [14]. In the other hand, vIRF1 represses tumor suppressor gene p53 phosphorylation, leading to an increase of p53 ubiquitination by reducing ATM kinase activity [15]; vIRF1 could also directly bind to p53 and effectively inhibit p53-mediated apoptosis by reducing its acetylation and inhibiting the transcription of p53 activation [16, 17]. In addition, vIRF1 restrains TGF-beta signaling via direct interaction with Smads (Smad3 and Smad4) to disturb Smad3/Smad4 complexes from binding to DNA and suppresses IRF-1-induced CD95/CD95L signaling-mediated apoptosis [18, 19]. As the first identified oncogenic protein encoded by KSHV, vIRF1 has been reported to transform mouse embryonic fibroblasts (NIH3T3) cells [6], however, its role in KSHV-induced tumor invasiveness and motility and its underlying mechanism remains totally unclear.
Less than 2% of the human genome encodes protein-coding genes, while the vast majority of the genome is transcribed as non-coding RNAs [20]. Based on the size, non-coding RNAs could be vaguely divided into three groups: microRNAs (miRNAs), long non-coding RNAs (lncRNAs), and circular RNAs (circRNAs) [21]. Many miRNAs (~22 nucleotides in length) have been well-characterized and shown to repress gene expression by inhibiting the translation or destabilization of mRNA transcript via binding to mRNA sequences [22]. LncRNAs (>200 nucleotides in length) have indispensable roles in diverse biological processes, including chromatin remodeling, X chromosome inactivation, genomic imprinting, nuclear transport, transcription, RNA splicing and translation [23–25]. A growing volume of literatures support the notion that both lncRNAs and miRNAs could function as tumor suppressors or oncogenes involved in the regulation of cell proliferation, metastasis, apoptosis, and invasion [24, 26, 27]. More interestingly, emerging evidence indicates that numerous lncRNAs might act as competing endogenous RNAs (ceRNAs) that competitively bind miRNAs, hence exerting influence on posttranscriptional regulation [28].
Recently, several oncogenic viruses have been shown to encode lncRNAs and are thought to participate in enhancing viral replication, promoting oncogenesis and contributing to pathogenesis [29–31]. KSHV encodes an lncRNA, known as polyadenylated nuclear RNA (PAN RNA). PAN is multifunctional, regulating KSHV replication, viral and host gene expression, and immune responses [32–35]. However, whether cellular lncRNAs are involved in the progression of KS is still unknown.
In the present work, we aimed to elucidate the role of vIRF1 in cell migration, invasion and proliferation. We found that vIRF1 promoted cell migration, invasion and proliferation by epigenetically silencing miR-218-5p and activating lncRNA-OIP5-AS1 transcription. Further, we uncovered that the crosstalk between miR-218-5p and lnc-OIP5-AS1 contributed to vIRF1-induced cell motility and proliferation via increasing HMGB2 and CMPK1 expression. Our novel findings illustrated a critical role of vIRF1 in the invasiveness, motility and development of KS tumor.
Previous works showed that vIRF1, as a homologue of cellular IRFs, disrupted immune antiviral response of host cells and contributed to KSHV-induced tumorigenesis [5]. However, its role on tumor invasiveness and motility remains unclear. To determine whether vIRF1 had a role in cell motility, we transduced HUVECs with lentiviral vIRF1 at a MOI of 2. vIRF1-transduced HUVECs showed a vIRF1 mRNA expression level similar to that of KSHV-infected HUVECs (S1 Fig, Fig 1A and 1B). We then examined the effect of vIRF1 on cell migration and invasion. In transwell migration and Matrigel invasion assays, overexpression of vIRF1 enhanced cell migration and invasion (Fig 1C, 1D and 1E). In plate colony formation assay, vIRF1 clearly enhanced cell proliferation (Fig 1F and 1G).
To assess the mechanism mediating vIRF1 promotion of cell motility and proliferation, we performed microarray-based miRNA expression profiling and identified a set of miRNAs that were differentially expressed between vIRF1- and pHAGE-transduced HUVECs (GEO accession number GSE119034). As a known tumor suppressor [36], miR-218-5p was significantly down-regulated in vIRF1- transduced cells, and hence was selected for further validation by qRT-PCR. As shown in Fig 1H and 1I, downregulation of miR-218-5p was observed in both vIRF1-transduced and KSHV-infected HUVECs. Then, we sought to determine whether the downregulation of miR-218-5p might contribute to vIRF1 promotion of cell motility, and proliferation. As expected, overexpression of miR-218-5p in vIRF1-transduced HUVECs reversed vIRF1-enhanced cell migration and invasion (S2 Fig, Fig 1J and 1K) as well as cell proliferation (Fig 1L).
Next, we conducted mass spectrometry analysis to investigate the direct targets of miR-218-5p. As shown in Table 1, there were a series of proteins that were up-regulated by > 1.5 folds in cells overexpressing vIRF1. Using bioinformatics analysis, we predicted four proteins that might be the potential targets of miR-218-5p, and hence chose them for further luciferase reporter assay. We confirmed that miR-218-5p decreased the 3’UTR reporter activities of both high mobility group box 2 (HMGB2) and cytidine/uridine monophosphate kinase 1 (CMPK1) (Fig 2A), which was further shown in a dose-dependent fashion (Fig 2B). Indeed, overexpression of miR-218-5p suppressed the levels of HMGB2 and CMPK1 proteins in a dose-dependent manner (Fig 2C). Conversely, blocking the miR-218-5p function with a specific inhibitor elevated the expression levels of HMGB2 and CMPK1 proteins in a dose-dependent manner (Fig 2D). To further confirm that miR-218-5p directly targeted HMGB2 and CMPK1, we performed mutagenesis with miR-218-5p (Fig 2E and 2F). The mutant mimic did not have any effect on the 3’UTR reporter activities of HMGB2 and CMPK1 (Fig 2G), and the levels of HMGB2 and CMPK1 proteins (Fig 2H). Moreover, the mRNA and protein levels of HMGB2 and CMPK1 were significantly up-regulated in cells expressing vIRF1 or infected by KSHV (Fig 3A–3D). In IHC staining, there were more HMGB2- and CMPK1-postive cells in KS lesions than in normal skin tissues (S3 Fig and Fig 3E).
Previous studies have shown that HMGB2 and CMPK1 are abundantly up-regulated in various malignant tumors, and are closely associated with tumor development and poor prognosis [37–46]. To determine if upregulation of HMGB2 and CMPK1 was necessary for vIRF1-induced cell motility, and proliferation, we silenced HMGB2 and CMPK1 expression in vIRF1-transduced HUVECs with a mixture of siRNAs, respectively (S4 Fig), and observed diminished vIRF1-induced cell migration, invasion and proliferation (Fig 4A–4I). Moreover, knock-down of HMGB2 and CMPK1 also inhibited KSHV-induced cell migration and invasion (Fig 4J–4L).
MiR-218-5p is expressed from two separate loci, pre-miR-218-1 and pre-miR-218-2, which are co-expressed with their host genes SLIT2 and SLIT3, respectively [47]. The expression of miR-218-5p depends on the promoter activity of its host genes, and hypermethylation of the promoter inhibits miR-218-5p expression [48]. Therefore, we examined the expression of SLIT2 and SLIT3. The expression of SLIT2 mRNA was low in both vIRF1-transduced and KSHV-infected cells while no expression of SLIT3 was detected in HUVECs (Fig 5A and 5B). These results indicated that vIRF1 might suppress miR-218-5p by reducing the expression of primary form pre-miR-218-1 and its host gene SLIT2 by epigenetically silencing their promoter. Indeed, the expression of pre-miR-218-1 was significantly suppressed by both vIRF1 and KSHV infection (Fig 5C and 5D). Consistent with these results, methylation-specific PCR showed that the promoter of pre-miR-218-1 was more hypermethylated in vIRF1-expressing cells and KSHV-infected cells than normal cells (Fig 5E and 5F). Moreover, treatment with a potent inhibitor of DNA methylation, 5-aza, not only blocked vIRF1 suppression of miR-218-5p (Fig 5G), but also decreased the expression of its targets HMGB2 and CMPK1 (Fig 5H). These results revealed that vIRF1 silencing of miR-218-5p expression was due to DNA methylation on its promoter.
DNA methyltransferase 1 (DNMT1) mediates DNA methylation and has been reported to cause miR-218-5p silencing [49]. We found that DNMT1 was up-regulated by 1.62-fold in vIRF1-transduced HUVECs (Table 1). Moreover, Western-blotting confirmed that DNMT1 protein was up-regulated in vIRF1-expressing cells and KSHV-infected cells (Fig 6A and 6B). Knock-down of DNMT1 expression with specific siRNAs (siDNMT1) reduced the hypermethylation of the promoter of pre-miR-218-1 in vIRF1 expressing cells (Fig 6C), and enhanced the expression of SLIT2, pre-miR-218-1 and miR-218-5p (Fig 6D). Meanwhile, vIRF1 induced expression of HMGB2 and CMPK1 was also abolished following DNMT1 inhibition (Fig 6E).
vIRF1 binds to p53 and represses p53-dependent transcription and apoptosis [16]. p53 transcriptionally suppresses the DNMT1 promoter by interacting with specificity protein 1 (Sp1) and forma complex [50]. Based on these studies, we sought to elucidate whether vIRF1 inhibition of p53-dependent transcription was responsible for vIRF1-induced DNMT1 up-regulation and therefore was involved in the inhibition of miR-218-5p. Indeed, overexpression of p53 in vIRF1-transduced HUVECs reduced the expression levels of DNMT1, HMGB2 and CMPK1 (Fig 6F), as well as hypermethylation of the promoter of pre-miR-218-1 in vIRF1-expressing cells (Fig 6G), hence causing an increase of both pre-miR-218-1 and miR-218-5p expression in vIRF1-infected HUVECs (Fig 6H).
These results indicated that vIRF1-induced miR-218-5p inhibition via aberrant DNA methylation at the pre-miR-218-1 promoter by inhibiting p53 to cause DNMT1 upregulation.
Numerous studies have shown that lncRNAs can act as ceRNAs to regulate the functions of miRNAs. To identify lncRNAs which may serve as ceRNAs and interact with miR-218-5p, we utilized online software programs starbase v2.0 (http://starbase.sysu.edu.cn/) and LncBase Predicted v.2 (http://carolina.imis.athena-innovation.gr/diana_tools/web/index.php?r=lncbasev2%2Findex-predicted) to search for lncRNAs that have complementary base pairing with miR-218-5p. Considering the abundance in the cytoplasm, and high score of predicted binding sites, we identified lncRNA OIP5 antisense RNA 1 (lncRNA-OIP5-AS1) as a potential candidate. We found that there were four putative miR-218-5p-binding sites in lnc-OIP5-AS1 (Fig 7A) and that lnc-OIP5-AS1 was indeed up-regulated in both vIRF1-transduced and KSHV-infected HUVECs (Fig 7B and 7C). We also found that vIRF1 was capable of activating the luciferase activity of lnc-OIP5-AS1 promoter (S5 Fig). We generated four luciferase reporter constructs, each of which contains only one putative miR-218-5p-binding site. Of these, miRNA-218-5p mimics reduced the luciferase activities of lnc-OIP5-AS1 (S3) and lnc-OIP5-AS1 (S4) reporters (Fig 7D) in a dose-dependent fashion (S6 Fig). In contrast, the miR-218-5p mutant mimic lacking the seed sequence did not reduce the luciferase activities of both lnc-OIP5-AS1 (S3) and lnc-OIP5-AS1 (S4) reporters (Fig 7E). To confirm the physical interaction between miR-218-5p and lnc-OIP5-AS1, we performed RNA pull-down experiments. Biotin-labeled mimics were incubated with HUVECs lysates, isolated with streptavidin agarose beads and then analyzed by RT-qPCR. We observed that lnc-OIP5-AS1 enriched miR-218-5p but not the miR-218 mutant (Fig 7F). These results indicated that lnc-OIP5-AS1 could directly bind miR-218-5p.
To determine whether lnc-OIP5-AS1 could act as a ceRNA to abrogate the function of miR-218-5p by releasing its binding to the targeted transcripts, we knocked down lnc-OIP5-AS1 with specific Smart Silencer in HUVECs and performed an RNA immunoprecipitation (RIP) assay on Ago2 (S7 Fig). We found that silencing of lnc-OIP5-AS1 in HUVECs increased the HMGB2 and CMPK1 transcripts in the Ago2 complex (Fig 7G). Furthermore, overexpression of lnc-OIP5-AS1 (S3) and lnc-OIP5-AS1 (S4) fragments abolished the inhibition of HMGB2 and CMPK1 3’UTR reporter activities by miR-218-5p (Fig 7H). These results demonstrated the sequestration of miR-218-5p by lnc-OIP5-AS1, which relieved the inhibition of the HMGB2/CMPK1 transcripts by miR-218-5p.
Intriguingly, overexpression of miR-218-5p significantly reduced the level of vIRF1-induced lnc-OIP5-AS1 (Fig 7I). Conversely, inhibition of lnc-OIP5-AS1 reversed vIRF1 inhibition of the expression of both pre-miR-218-1 and miR-218-5p (Fig 7J). We further performed knock down of Dicer to prevent maturation of miR-218-5p from pre-miR-218-1 and then examined the effect of silencing lnc-OIP5-AS1 on miR-218-5p stability (S8 Fig). We found that suppression of lnc-OIP5-AS1 reduced the degradation of miR-218-5p (Fig 7K). As a result, silencing of lnc-OIP5-AS1 attenuated vIRF1-induced DNMT1, HMGB2 and CMPK1 expression (Fig 7L). These results indicated that lnc-OIP5-AS1 could not only inhibit the function of miR-218-5p by acting as a ceRNA but also reduce the level of miR-218-5p by direct binding to induce miR-218-5p degradation. Consistent with these results, silencing of lnc-OIP5-AS1 inhibited vIRF1-induced cell migration, invasion and proliferation (Fig 7M and 7N).
To further dissect the functions of vIRF1 in the context of KSHV genome, we constructed a KSHV mutant with ORF-K9 deleted using a two-step red recombination system as previously described [51–53]. Positive colonies were screened and verified by PCR (S9A and S9B Fig). Restriction analysis showed that the RGB-K9-mutant had a band shift of about 1.3 kb compared to the wild-type RGB-BAC16, indicating that the K9 mutant bacmid was successfully generated (S9C Fig). The RGB-K9-mutant was transfected into iSLK cells and selected to obtain stable producer iSLK cells. As expected, we did not detect the expression of vIRF1 in iSLK-RGB-K9 mutant cells, while the levels of vIRF4 and ORF57 had minimal changes (S9D Fig). Similarly, HUVECs infected by the mutant virus had no vIRF1 expression (S9E Fig) but had minimal changes in the levels of vIRF4 and ORF57 (S9F Fig). As expected, expression levels of phosphorylated p53, acetylated p53, and p21 were increased in vIRF1_mutant cells compared to those of KSHV_WT virus-infected HUVECs (S9G Fig). Because HUVECs transduced with lentiviral vIRF1 at a MOI of 2 showed a vIRF1 mRNA expression level similar to that of wild type KSHV-infected HUVECs (S1 Fig), we transduced vIRF1_mutant cells with 2 MOI of lentiviral vIRF1. We found that loss of vIRF1 not only reduced cell migration and invasion (Fig 8A) but also decreased the level of hypermethylation in the pre-miR-218-1 promoter (Fig 8B). Importantly, complementation with vIRF1 in vIRF1_mut-infected HUVECs was sufficient to rescue cell migration and invasion induced by KSHV (Fig 8A), and reverse the level of hypermethylation in the pre-miR-218-1 promoter induced by KSHV (Fig 8B). We also observed a significant decrease of lnc-OIP5-AS1 expression, and an increase of miR-218-5p and pre-miR-218-1 expression in the mutant cells compared to both KSHV_wild-type infected cells and vIRF1-transduced mutant cells (Fig 8C). Furthermore, deletion of vIRF1 reduced the expression of DNMT1, HMGB2 and CMPK1 while complementation with vIRF1 was sufficient to rescue the expression levels of these proteins (Fig 8D). Meanwhile, inhibition of miR-218-5p with a specific inhibitor in both mutant cells and vIRF1-transduced mutant cells increased cell migration and invasion (Fig 8E). Similar increase in cell migration and invasion was also observed in the mutant and vIRF1-transduced mutant cells following overexpression of lnc-OIP5-AS1 (S3) and lnc-OIP5-AS1 (S4) fragments (Fig 8F). Taken together, these results demonstrated that vIRF1 mediated KSHV-induced cell migration, and invasion by down-regulating miR-218-5p and up-regulating lnc-OIP5-AS1.
KSHV K9/vIRF1 was initially characterized as an early lytic gene. However, subsequent studies have shown that it is also expressed during viral latency. vIRF1 has two transcription start sites, one is distal to the AUG, which is active during latency in PEL, and another is a more proximal site, which is induced upon lytic reactivation [54]. Hence, vIRF1 might have a dual modes of expression during latent and lytic replication [55–57]. Furthermore, K9/vIRF1 mRNA is expressed in all KS tumors (total 21 KS clinical biopsies) and preferentially transcribed during latent infection of either endothelial/mesenchymal lineage cells, which strengthens the role of K9/vIRF1 in KS tumorigenesis [58]. In the current study, we found that vIRF1 promoted endothelial cell migration and invasion, as well as proliferation. Further, deletion of vIRF1 from the KSHV genome reduced KSHV-induced cell migration, invasion and proliferation. However, we could not assess the expression level of the endogenous vIRF1 protein because there is currently no vIRF1 antibody available. Despite the limitation, this work still revealed a novel role of vIRF1 in cell migration, invasion and proliferation, which is an important part of KS pathogenesis, particularly in the invasiveness and dissemination of KS tumors.
MiR-218-5p, a vertebrate-specific intronic miRNA co-regulated with its host genes SLIT2/SLIT3, functions as a tumor suppressor by modulating multiple pathways [36]. It is downregulated in numerous human cancers, such as colorectal, prostate, pancreatic, gastric, and thyroid cancers [47, 49, 59–63]. The mechanism of silencing miR-218-5p and its host genes, SLIT2/SLIT3 is through promoter hypermethylation. For instance, human papillomavirus type 16 oncogene E6 reduces the level of miR-218-5p and SLIT2 through promoter hypermethylation [64]. However, significance of miR-218-5p in the development of KS remains undefined. In this study, we revealed that both KSHV infection and vIRF1 expression reduced the level of miR-218-5p at least in part by silencing of pre-miR-218-1/SLIT2 via promoter hypermethylation. Further, we demonstrated that vIRF1 increased DNMT1 expression by inhibiting p53 transcriptional activity, leading to a higher level of DNA methylation of the pre-miR-218-1 promoter.
Lnc-OIP5-AS1 located at chromosome 15q15.1, known as cyrano, is ~8,000 nucleotides in length and abundant in the cytoplasm. It was originally characterized in zebrafish and displayed crucial effects in embryonic nervous system development [65]. It was also reported to play a vital role in embryonic stem cells (ESCs) self-renewal maintenance [66]. With regard to its role in cancer, lnc-OIP5-AS1 exhibits multifaceted and complex features. For example, lnc-OIP5-AS1 is shown to be a tumor suppressor and inhibit HeLa cells proliferation by interacting with the RBP HuR to reduce HuR’s availability for binding target mRNAs, or associating with GAK mRNA to impair GAK mRNA stability [67, 68]. On the contrary, lnc-OIP5-AS1 can exert oncogenic functions in several other cancers. It was consistently up-regulated in renal cell carcinoma, glioblastoma, and gastric cancer [69–71]. Silencing of lnc-OIP5-AS1 repressed YAP-Notch signaling pathway activity leading to decrease of glioma cells’ proliferation, migration in vitro and tumor formation in vivo [72]. Moreover, lnc-OIP5-AS1 was highly expressed in lung adenocarcinoma tissues and cells, and the loss of lnc-OIP5-AS1 inhibited lung adenocarcinoma cell proliferation, migration and invasion [73]. In our report, we found that both KSHV infection and vIRF1 expression increased the expression of lnc-OIP5-AS1 in endothelial cells. Silencing of lnc-OIP5-AS1 suppressed cell migration, invasion and proliferation. Intriguingly, we found that vIRF1 activated the transcription of lnc-OIP5-AS1, however, the precise mechanism remains unknown.
The cross-regulatory interactions between lncRNAs and miRNAs have been recognized to regulate their downstream targets of either lncRNAs or miRNAs [74, 75]. Several miRNAs including miR-7 [66], miR-410 [76], miR-424 [67], and miR-448 [73] have been identified to interact with lnc-OIP5-AS1. On the other hand, miR-218-5p has been reported to interact with lnc-MALAT1, participating in choriocarcinoma growth [77]. In the current study, we revealed the crosstalk between miR-218-5p and lnc-OIP5-AS1, confirmed a direct interaction between miR-218-5p and lnc-OIP5-AS1, and unearthed the fateful consequences of this interaction. We showed that lnc-OIP5-AS1 functioned as a ceRNA and sequestered miR-218-5p to relieve its binding and targeting of HMGB2 and CMPK1 transcripts. Further, lnc-OIP5-AS1 could inhibit miR-218-5p expression through regulating miR-218-5p stability. Once entering the RNA-induced silencing complex (RISC), miRNAs become extremely stable due to the protection of both ends of miRNAs by AGO proteins from 3ʹ–5ʹ exoribonucleases-mediated degradation. Therefore, we speculated that lnc-OIP5-AS1 might block miR-218-5p from loading onto AGO proteins, and hence accelerate its degradation. Interestingly, by an unclear mechanism, lnc-OIP5-AS1 also increased DNMT1 expression to promote DNA methylation of the pre-miR-218-1 promoter, leading to decreased level of miR-218-1. On other hand, the lnc-OIP5-AS1/miR-218-5p interaction also resulted in miR-218-5p suppression of lnc-OIP5-AS1 expression albeit the precise mechanism is unknown.
In conclusion, our study revealed that vIRF1 promoted cell migration, invasion and proliferation by a p53- and lnc-OIP5-AS1-mediated down-regulation of miR-218-5p, leading to increased expression levels of its target genes HMGB2 and CMPK1 (Fig 8G). This process was mediated by the complex crosstalk between miR-218-5p and lnc-OIP5-AS1. These novel findings extend the cross-regulatory network of cellular lncRNAs and miRNAs involved in the pathogenesis of KS.
The clinical section of the research was reviewed and ethically approved by the Institutional Ethics Committee of the First Affiliated Hospital of Nanjing Medical University (Nanjing, China; Study protocol # 2015-SR-116). Written informed consent was obtained from all participants, and all samples were anonymized. All participants were adults.
The iSLK cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 1% penicillin-streptomycin, 1 μg/ml puromycin and 250 μg/ml G418. The established iSLK-RGB-BAC16 and iSLK-RGB-K9 mutant cells were cultivated in DMEM supplemented with 10% fetal bovine serum (FBS), 1 μg/ml puromycin, 250 μg/ml G418, and 1.2 mg/ml hygromycin B [53]. HEK293T and continuous cell lines human umbilical vein endothelial cells (catalog #CRL-1730, ATCC, Manassas, VA, USA) were maintained as previously described [78]. The latter were only used for plate colony formation assay to evaluate the ability of cell proliferation. Primary human umbilical vein endothelial cells (HUVECs), which were used for all assays except for luciferase and plate colony formation assays, were isolated and cultured as previously delineated [79].
Flag-vIRF1 was cloned by inserting the coding sequences into plasmid pHAGE-CMV-MCSIzsGreen as previously described [78, 80]. The respective sequences of HMGB2 3’UTR, CMPK1 3’UTR and lncRNA-OIP5-AS1 fragments containing putative miR-218-5p binding sites (S1: 365–1167; S2: 3078–3539; S3: 4293–4701 and S4: 7775–8169) were amplified by PCR and inserted into the pGL3-Control plasmid (Promega, Madison, WI, USA), respectively. The pCMV6-Entry-C-Myc-p53 construct and pCMV6-Entry-C-Myc construct were provided by ORIGENE (Beijing, China). The DNA fragment of lnc-OIP5-AS1 covering –1500 bp to 0 bp of the transcription start site was amplified and subcloned into pGL3-Basic plasmid (Promega, Madison, WI, USA).
HUVECs were transfected using the Effectence transfection reagent (Qiagen, Suzhou, Jiangsu, China), while HEK293T cells were transfected using the Lipofectamine 2000 Reagent (Invitrogen, Carlsbad, CA, USA). 5-aza (Decitabine), a potent inhibitor of DNA methylation, was from Selleck Chemicals (Shanghai, China). siRNAs were synthesized from Genepharma (Suzhou, China), the sequences of siRNAs are listed in S1 Table. LncRNA Smart Silencer was obtained from RiboBio (Guangzhou, China). Antibodies against KSHV LANA, HMGB2, CMPK1, DNMT1 and Dicer were from Abcam (Cambridge, MA, USA). Anti-Flag was obtained from Cell Signaling Technologies (Beijing, China). Anti-Myc, anti-α-Tubulin, and anti-GAPDH were from Santa Cruz Biotechnology (Dallas, TX, USA). Anti-rabbit immunoglobulin G (IgG), anti-mouse IgG, anti-phosphorylated p53, anti-acetylated p53, anti-p53, and anti-p21 were purchased from Beyotime Institute of Biotechnology (Nantong, Jiangsu, China). Western-blotting analysis was conducted as previously described [81, 82]. In this study, all Western blotting results were independently repeated at least three times unless otherwise stated.
Cell migration, invasion and colony formation assays were executed as previously described [83–85].
Luciferase reporter assay was conducted using the Promega dual-luciferase reporter assay system according to the previous study [86].
Methylation-specific PCR (MS-PCR) was adopted using DNA Bisulfite conversion kit (TIANGEN BIOTECH, Beijing, China) and Methylation-specific PCR kit (TIANGEN BIOTECH, Beijing, China) according to the manufacturer’s instructions. MS-PCR primers were designed as previously described [87].
HUVECs were collected, washed, and re-suspended with lysis buffer (Thermo Fisher Scientific, Waltham, America). After incubating for 5 min, the lysates were precleared by centrifugation at 14,000 rpm for 10 minutes, and then were added to streptaviden magnetic beads (Thermo Fisher Scientific, Waltham, America), which were incubated with Biotin-labeled miR-218-5p, miR-218-5p mut 2, or Neg. Ctrl (Genepharma, Suzhou, China) for 4 hours. The bound RNAs in the pull-down material were quantified by qRT-PCR.
HUVECs were transfected with lnc-OIP5-AS1 Smart Silencer or its Neg. Ctrl for 48 h, and used for RIP experiments with an anti-Ago2 antibody (MERCK, Darmstadt, Germany) and the Magna RIPTM RNA-Binding Protein Immunoprecipitation Kit (MERCK, Darmstadt, Germany), according to the manufacturer’s instructions. The levels of lnc-OIP5 AS1, HMGB2 or CMPK1 were examined by qRT-PCR.
A KSHV mutant with ORF K9 deleted was constructed as described in previous studies [52, 88]. In brief, using the bacterial artificial chromosome (BAC) technology and the Escherichia coli Red recombination system, together with PCR, restriction digestion, and sequencing for strict quality control, a KSHV ORF K9 mutant (called RGB-K9-mutant) was constructed by removing K9 coding sequence (CDS) from the wild-type recombinant KSHV RGB-BAC16 [53]. RGB-BAC16 and RGB-K9 mutant DNA were transfected into iSLK cells and selected using 1 μg/ml puromycin, 250 μg/ml G418, and 1.2 mg/ml hygromycin B for 3 weeks to establish stable viral producer cell lines, iSLK-RGB-BAC16 and iSLK-RGB-K9 mutant cells. To produce virus stocks for infection, iSLK-RGB-BAC16 and iSLK-RGB-K9 mutant cells were plated at 30 to 40% confluence and induced with both Doxycycline (Dox) (1 μg/ml) and sodium butyrate (NaB) (1 mM). After induction for 4 or 5 d, the supernatant was harvested, centrifuged, filtered, and concentrated by ultracentrifugation (25, 000 g at 4°C for 3 h) using SW32 Ti rotor (Beckman Coulter Inc, USA). The pellet was resuspended, supplemented with 8 μg/mL polybrene and then incubated with 105 HUVECs in a 6-well plate for 4 h. The primers for construction and identification of K9 mutant bacmid were designed as previously described [89] and the sequences of the primers could be found in S2 Table.
RNA was extracted using RNA Isolator Total RNA Extraction Reagent (Vazyme Biotech Co., Ltd, Nanjing, China) from cells. Total RNA was reverse transcription by HiScript Q RT SuperMix (Vazyme Biotech Co., Ltd, Nanjing, China). Real time quantity PCR was performed by AceQ qPCR SYBR Green Master Mix (Vazyme Biotech Co., Ltd, Nanjing, China). The sequences of the primers for PCR could be found in S3 Table.
The extraction of genome DNA was performed by using TIANamp Genomic DNA Kit (TIANGEN BIOTECH, Beijing, China) according to the user’s guide. Briefly, cells were trypsinized, and neutralized by 20% FBS DMEM. The suspension was centrifuged and the supernatant was discarded.
Mass spectrometry analysis was adopted according to the previous study [86].
The KS clinical specimens were kindly offered by Jiangsu Province Hospital. All samples were anonymized and all participants are provided with informed consent. IHC was carried out as previously described with specific antibodies [85, 90].
All data are appeared as the means ± SD with at least three replications. Statistical analysis was on account of Student’s t-test and the criterion for statistical significance was adopted as P values of < 0.05.
Microarray data have been submitted and can be accessed by GEO accession number GSE119034.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.