Analysis of axonal morphology in constant darkness was performed on the second day after switching to DD (DD2). For the analysis of activity-dependent changes in axonal morphology, yw; Pdf-Gal4, UAS-mCD8GFP /UAS-TrpA1 and yw; Pdf-Gal4, UAS-mCD8GFP /UAS-TrpA1; UAS-Mef2RNAi/+ flies were entrained for 3 days using a 12:12 LD cycle at 21°C and collected for dissection at ZT14 immediately after

a 2 hr temperature elevation to 29°C. Imaging was performed with a Leica TCS SP2 confocal microscope using a 20× objective and a 4× digital zoom. Axons were traced using the Simple Neurite Tracer plugin for Fiji software VX-770 ( Longair et al., 2011). Quantitative analysis was performed with ImageJ 1.40 from NIH (http://rsb.info.nih.gov/ij). Axons of all s-LNv neurons in each brain hemisphere were analyzed as a group ( Fernández et al., 2008). For the Sholl’s analysis, 15 concentric circles spaced 10 μm apart were centered on the point where dorsal ramification opens. Total number of intersections selleck compound between axon branches and the concentric circles was computed using Sholl Analysis Plugin for ImageJ (Ghosh laboratory, UCSD). We have also modified this plugin

to additionally detect a 15° cone containing most of the intersections and to compute the fraction of the intersections outside of that “main projection direction” cone. Nearly identical results were seen when brains were stained with anti-GFP antibody using a standard immunohistochemistry

protocol. Immunostaining was performed as previously described in Tang et al. (2010). Briefly, fly heads were removed, fixed in 4% paraformaldehyde for 45 min at 4°C, and brains were dissected Astemizole in PBS. Brains were blocked in 10% normal goat serum (Jackson Immunoresearch) and subsequently incubated with primary antibodies at 4°C for 48 hr. Primary antibodies and their dilutions used were as follows: rabbit anti-GFP at 1:500 (Invitrogen), mouse anti-mCherry at 1:100 (Clontech), and mouse anti-PDF at 1:10 (from Developmental Studies Hybridoma Bank, University of Iowa). For detection of primary antisera, Alexa 488 goat anti-rabbit, Alexa 488 goat anti-mouse, and Alexa 633 goat anti-mouse (Invitrogen) were used at a dilution of 1:200. Brains were mounted in Vectashield Mounting Medium (Vector Laboratories). Locomotor rhythms of individual male flies were monitored for 4 days in LD conditions (12:12 LD intervals) followed by 4–9 days in DD conditions (constant darkness) using Trikinetics Drosophila Activity Monitors. Analyses of period length and rhythmic strength (assessed by by rhythmicity index [RI]; Levine et al., 2002) were performed with MATLAB-based software ( Donelson et al., 2012). Flies with an RI > 0.15 were considered rhythmic, with an RI = 0.1–0.15 weakly rhythmic, and with an RI < 0.1 arrhythmic.

e1-5.). Reprints are available from Hong Jiang, MD, Reproductive Medicine Centre, 105 Hospital of PLA, 424 Changjiang Rd, Hefei, China. [email protected]. “
“The recent introduction of cell-free DNA (cfDNA)-based noninvasive prenatal testing (NIPT) has offered pregnant women a more accurate BKM120 manufacturer method for detecting fetal aneuploidies than traditional serum screening methods.1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 and 12 NIPT noninvasively determines fetal chromosome copy number by interrogating

cfDNA isolated from maternal plasma, with the fetus contributing anywhere from <2% to >30% of the total cfDNA.3, 7 and 13 Other NIPT approaches use quantitative “counting” methods where fetal chromosome copy number is determined by comparing see more the absolute number of sequence reads from the chromosome(s) of interest (eg, chromosome 21) to reference

chromosome(s), and inferring fetal trisomy when this ratio is above a predetermined threshold. This approach cannot determine the source of DNA (fetal or maternal) and is therefore unable to detect additional fetal haplotypes associated with triploidy or vanishing twins. Vanishing twins were reported to account for 15% of false positives in a recent counting-based NIPT study.14 This likely results in unnecessary invasive prenatal testing. A more recent approach using a single-nucleotide polymorphism (SNP)-based method along with sophisticated informatics can resolve this potential source of false-positive results. This approach identifies the presence of additional fetal haplotypes, indicative of a triploid or dizygotic multifetal pregnancy, and determines parental origin.10 and 12 Using the SNP-based approach, the prevalence of cases found to have additional fetal haplotypes

within 30,795 consecutive cases undergoing routine clinical NIPT was determined, and is reported here. Clinical follow-up of these cases is also described. The current study included all samples from participating centers received for commercial testing from March 1, through Nov. 30, 2013, that received an NIPT result. This study received a notification of exempt determination from an institutional review board (Ethical and Independent Review Services, Bay 11-7085 no. 14064-01). All samples were analyzed at Natera’s Clinical Laboratory Improvement Act–certified and College of American Pathologists–accredited laboratory in San Carlos, CA. Analysis was performed for all samples on chromosomes 13, 18, 21, X, and Y, and included detection of trisomy 21, trisomy 18, trisomy 13, monosomy X, sex chromosome abnormalities (47,XXX/XXY/XYY), fetal sex, and additional fetal haplotypes. Maternal blood samples (>13 mL) were collected in Streck (Omaha, NE) blood collection tubes and processed at Natera (San Carlos, CA) within 6 days of collection.

However, there is also evidence suggesting that eCBs signal in a nonretrograde or autocrine manner, in which they can modulate neural function and synaptic transmission by engaging transient receptor potential vanilloid receptor type 1 (TRPV1) and also CB1Rs located on or within the postsynaptic cell (Figure 1B). Finally, recent studies indicate that eCBs can signal via astrocytes to indirectly modulate presynaptic or postsynaptic function (Figure 1C). This Review aims to highlight

the emerging mechanistic diversity of synaptic eCB signaling. The first demonstration of retrograde eCB signaling came from the discovery that eCBs mediate forms of short-term synaptic plasticity known as depolarization-induced suppression of inhibition (DSI) (Ohno-Shosaku et al., VE-821 concentration 2001; Wilson and Nicoll, 2001) and depolarization-induced

suppression of excitation (DSE) (Kreitzer and Regehr, 2001). Shortly after it was shown that eCBs also mediate presynaptic forms of long-term depression (eCB-LTD) at both excitatory (Gerdeman et al., 2002; Robbe et al., 2002) and inhibitory (Chevaleyre and Castillo, 2003; Marsicano et al., 2002) synapses. eCBs have since emerged as the best characterized retrograde messengers (Regehr et al., 2009), with numerous examples of short- and long-term forms of synaptic plasticity reported throughout the brain (Heifets and Castillo, 2009; Kano et al., 2009). CB1/CB2 receptors are Gi/o protein-coupled receptors that mediate RAD001 ic50 almost all effects of exogenous and endogenous cannabinoids. CB1Rs are one of the most widely expressed GPCRs in

the brain (Herkenham et al., 1990). Their localization to neuronal terminals (Katona et al., 1999, 2006) strongly suggests that CB1Rs play important roles in regulating synaptic function. Indeed, CB1R activation inhibits neurotransmitter release at synapses through two main mechanisms (Figure 2) (Chevaleyre et al., 2006; Freund et al., 2003; Kano et al., 2009). For short-term plasticity, in which CB1Rs are activated for a few seconds, the mechanism involves direct G protein-dependent (likely via the βγ subunits) inhibition of presynaptic Non-specific serine/threonine protein kinase Ca2+ influx through voltage-gated Ca2+ channels (VGCCs) (Brown et al., 2003; Kreitzer and Regehr, 2001; Wilson et al., 2001). For long-term plasticity, the predominant mechanism requires inhibition of adenylyl cyclase and downregulation of the cAMP/PKA pathway via the αi/o limb (Chevaleyre et al., 2006; Heifets and Castillo, 2009). Moreover, CB1Rs only need to be engaged during the induction, but not expression, phase of eCB-LTD. Induction also requires combined presynaptic firing with CB1R activation, thereby providing a mechanism for input specificity; that is, only active synapses detecting eCBs express long-term plasticity (Heifets et al., 2008; Singla et al., 2007). The expression mechanism for eCB-LTD may involve presynaptic proteins Rab3B/RIM1α (Chevaleyre et al., 2007; Tsetsenis et al.

For Dose 1 and Dose 2, early blood samples were taken at 2, 6 and 12 h after treatment,

for the remaining doses the 2 and 12 h plasma collections were eliminated. The highest plasma concentration of 0.82 μg/ml was measured at the 6 h time point after dose 1 (Fig. 3). Pharmacokinetic profiles for afoxolaner were observed to be predictable and reproducible following multiple dosing (Fig. 3). Mean afoxolaner plasma concentrations at 6 h were 0.82, 0.81, 0.97, 0.91, and 0.80 μg/ml for Doses 1 through 5, respectively. There was no apparent difference in the trough concentrations as check details mean minimum afoxolaner plasma concentrations (Cmin) collected at 30 days post-dose were 0.09, 0.09, 0.12, 0.10 and 0.15 μg/ml for Doses 1 through 5, respectively ( Fig. 3). These data indicate that steady state had been reached by the 2nd dose. No adverse clinical signs were observed during the study. A KD50 (50% knockdown concentration) value of 0.35 μg per cockroach was determined. At the higher injected dose,

symptoms were observed within 10 min, initially appearing as brief periodic BAY 73-4506 wing fluttering which progressed over time until the insects became uncoordinated and had difficulty remaining upright. Once prostrate, cockroaches displayed periodic volleys of leg tremors. The rapid onset and excitatory nature of toxicity suggested involvement of a neuronal target. By doing extracellular recordings on nerve 5 (N5) of the metathoracic Adenosine ganglion of American cockroaches, under control conditions, a single air puff to the cerci produced a rapidly adapting volley of action potentials with a spike frequency between 75 and 175 Hz. Injection of CPD I (10 μg) into the body cavity produced no significant effect on spontaneous action potential frequency. However, adaptation of the air puff-induced N5 activity was inhibited by CPD I, resulting in a strong increase of spike frequency (Fig. 4b). Similarly, bath perfusion of CPD I (10 mM) induced a strong increase in the air puff-induced spike frequency indicating increased excitability

due to blocking of inhibitory neuronal activity (Fig. 4c). The fact that the spontaneous action potential frequency remained unaffected suggested that action at the neurotransmitter receptors was a more likely target than action at voltage-gated ion channels. As the neurotransmitters involved in the cercal reflex include both excitatory nicotinic acetylcholine receptors (nAChRs) and inhibitory GABA receptors (GABARs), action of CPD I was investigated on both neurotransmitter receptors. Although no effect was observed on nAChRs (data not shown), the compound potently inhibited GABA-induced currents in American cockroach thoracic neurons. CPD I inhibited GABA-induced currents with an IC50 value of 10.8 nM (Fig. 5) with prolonged saline rinse (>15 min) resulting in partial recovery of the GABA response.

Such a pathway through the parameter space of network connectivity could be utilized during development, with the integrator network beginning in a more topographically organized form and moving to a more distributed connectivity pattern in the mature state, where the functional signatures of topography seem to be weaker (as discussed in Miri et al., 2011). In addition, our approach can be extended to allow greater heterogeneity in synaptic parameters or to model

circuits with nonmonotonic tuning curves (D.F., unpublished data). We have considered a single shape of synaptic activation function for all excitatory neurons, and a separate single shape for all inhibitory neurons, regardless of threshold. Relaxing this constraint might identify circuit architectures in which there are gradients Alisertib concentration in synaptic activation parameters as a function of neuronal threshold. Our work makes several predictions about the mechanisms of integration in the oculomotor integrator

and possibly other short-term memory circuits. First, in contrast to the previous spiking model of the oculomotor integrator based upon purely saturating synapses (Seung et al., 2000; Figure S4D), which modeled a single unilateral population and was generated before the inactivation experiments had been performed, our sensitivity analysis suggests that both inhibition and excitation are likely to be mediated by approximately linear or sigmoidal synaptic activation functions. Second, our quantitative fits to the drift rates following

inactivation suggest that the GSI-IX observed long integration time constants may not be solely due to network mechanisms, and instead suggest the presence of an intrinsic cellular or synaptic process with a time constant of order 1 s. Third, we suggest that integration depends critically upon the presence of a threshold mechanism. This could either take the form of a synaptic (or dendritic) threshold, as suggested by Aksay et al. (2007), or result from the circuit’s recurrent connectivity depending critically upon neurons with high eye-position thresholds, particularly for inhibition. Potential “synaptic” mechanisms consistent with a sigmoidal dependence upon presynaptic firing rate and an ∼1 s time constant are presynaptic facilitation (Wang MYO10 et al., 2006) or, postsynaptically, localized dendritic plateau potentials (Major et al., 2008 and Wei et al., 2001). The long time constants associated with these mechanisms could provide robustness against disruptions of circuit connectivity (Camperi and Wang, 1998, Goldman et al., 2003, Koulakov et al., 2002 and Mongillo et al., 2008). The high thresholds could be useful in filtering out low firing rates (Chichilnisky and Rieke, 2005), which are noisier in the oculomotor integrator than higher firing rates (Aksay et al., 2003).

Specifically, due to the Vm-to-spikes nonlinearity, the baseline elevation of membrane potential could have a reduced effect at the level of spiking activity when the response is weak (e.g., before stimulus

onset or at locations far from stimulus center), and an enhanced effect when the response is strong (e.g., near the peak of the stimulus evoked response). The possible effects of such nonlinearity on the predicted attentional modulations at the level of spiking activity are illustrated in Figure 7. This nonlinearity could lead to a small increase in the firing rates of V1 neurons in the absence of visual stimulation. Such an effect could be difficult to detect using single unit electrophysiology Lenvatinib price but may be more prominent in population responses. Consistent with this possibility, a recent study of attentional modulations at the level of single check details neurons in V1 found no baseline modulations in the absence of the stimulus or at very low stimulus contrasts (Thiele et al., 2009), while fMRI results in V1 show clear baseline modulations even when the stimulus is absent (e.g., Buracas and Boynton, 2007, Murray,

2008 and Pestilli et al., 2011). The attentional signals are initiated after fixation point dimming and shortly before the visually evoked responses ( Figure 6). This result implies that top-down modulations can be anticipatory in nature, consistent with previous studies (e.g., Ghose and Maunsell, 2002), and do not require extrastriate cortex to first process the visual stimulus. In fact, the attentional signal can be present even in the absence of the visual stimulus ( Figures 6B and 6E), consistent with some electrophysiological ( Luck et al., 1997, Reynolds et al., 2000 and Williford and Maunsell, 2006) and fMRI ( Kastner et al., 1999 and Ress et al., 2000) results. The attentional modulations in V1 operate at a larger spatial scale than the stimulus-evoked response. Because the baseline elevation extends beyond our imaged area, we cannot determine the exact spatial

extent of this top-down signal. However, Florfenicol since we observed no baseline elevation in attend-out trials, the top-down signal must have a limited spatial extent, at least in focal attention trials. In distributed attention trials, top-down signals could activate simultaneously four broad but separate V1 regions peaked at the representation of the possible stimulus locations, consistent with a recent finding in humans (Müller et al., 2003). Alternatively, a larger contiguous region could be activated, such as the V1 region corresponding to a ring at target eccentricity. The spatial extent and shape of the top-down attentional signal could be addressed in future VSDI experiments by systematically shifting the position of the stimuli relative to the position of the receptive fields in the imaged area.

Thus, SGN axons are arrayed in a high-to-low-frequency tonotopic map along the dorsoventral axis of the CN (Young and Oertel, 2004). Similar tonotopy is observed in CN neuronal responses themselves, determined both electrophysiologically (Luo et al., 2009) and by Fos induction ( Friauf, 1992; Saint Marie et al., 1999). We injected FosTRAP mice with 4-OHT during a 4 or 16 kHz continuous pure tone stimulus to TRAP CN neurons tuned to those frequencies. To increase the total number of TRAPed cells, we took advantage of TRAP’s ability to integrate IEG expression over

time by using a 4 hr pure tone stimulus during the TRAPing period. Then, 4–5 days later, we delivered a second 4 or 16 kHz stimulus for 1 hr, sacrificed the mice 1 hr learn more later, and processed the tissue for Fos immunostaining (Figure 5A). Thus, TRAPed cells represent neurons Wnt inhibitor activated by the first stimulus, and Fos protein immunopositive (Fos+) cells represent neurons activated by the second stimulus. Consistent with prior results, we found that 4 kHz stimulation during the second epoch induced Fos expression in clusters of cells in all three CN

subdivisions that were located more ventrally than the clusters that were Fos+ after 16 kHz stimulation. Similar results were observed for TRAPed cells. When the tone frequency was the same for the two stimulus epochs, the TRAPed and Fos+ populations overlapped, and the 4 kHz cluster was localized more ventrally than the 16 kHz cluster (Figure 5B, first and third columns). Within mice receiving stimuli of two different frequencies, the cells TRAPed by the 16 kHz stimulus Ketanserin were dorsal to Fos+ cells induced by the 4

kHz stimulus (Figure 5B, second column), whereas the reverse was true when the 4 kHz stimulus was TRAPed and the 16 kHz representation was revealed by Fos immunostaining (Figure 5B, last column). These qualitative impressions were confirmed by the quantification of the numbers of TRAPed and Fos+ cells in bins spanning the dorsoventral axis of the central dorsal cochlear nucleus (DCN; Figure 5C). In general, the populations of TRAPed cells were less sharply confined along the dorsoventral axis than the population of Fos+ cells. This may reflect the longer stimulus used for TRAPing (4 hr, versus 1 hr for Fos immunostaining) or some general noise in the TRAP approach. Regardless, this analysis supports the observations from individual sections that both TRAP and Fos immunostaining reveal similar tonotopic maps along the dorsoventral axis of the DCN. We also quantified the overlap between TRAPed and Fos+ cells for the different treatment groups across the entire extent of the DCN.

001). In principle, SOA alone could have provided information to guide betting; monkeys could have ignored their trial-by-trial decisions and just bet high more often if the masks appeared later or the task seemed easier. We analyzed the data from each SOA separately to address this potential

confound. Trial-by-trial analyses revealed that for each monkey, within each SOA, bets were correlated appropriately with decisions (χ2 test, p < 0.001 for each SOA and each monkey; details in Middlebrooks and Sommer, 2011). We quantified performance across SOAs using two phi correlation methods (Kornell et al., 2007; Zar, 1999). Phi correlation values could range from zero (random betting) to one (perfect association between decisions and bets). Both monkeys’ phi correlations, assessed with either method (Figure 1C; Figure S1 available online), were significant at each SOA and constant across SOAs Alpelisib manufacturer (one-way ANOVA,

p > 0.05). Another potential confound is the use of motor-related cues. Monkeys could possibly detect Y-27632 molecular weight their saccade latencies during the decision stage and use this information to help place bets. This explanation is feasible if latency distributions differ between correct-high versus correct-low trials and between incorrect-high and incorrect-low trials, but they did not (Table S1). All of these results replicate our prior findings (Middlebrooks and Sommer, 2011) and indicate that, within each trial during neuronal recordings, monkeys maintained information about their decision to guide their bet, a metacognitive strategy. We studied 87 neurons in the FEF (Monkey N: 35, Monkey S: 52), 112 in the PFC (N: 54, S: 58), and 133 in the SEF (N: 61, S: 72). As expected, neurons in all three areas were highly modulated during the task (Figure S2). The monkeys’ betting behavior did not vary significantly between recording sessions in the three unless cortical areas (phi correlations for Monkey N: FEF, 0.51; PFC, 0.49; SEF, 0.47; for Monkey S: FEF, 0.59; PFC, 0.54; SEF, 0.54;

no differences between areas by ANOVAs, p > 0.05, for both monkeys). Because the monkeys were well trained, the neuronal recording data included more correct-high and incorrect-low trials (the appropriate decision-bet pairings) than correct-low and incorrect-high trials (Table S2 shows the breakdown of trial outcomes). To test whether neurons encoded the decision, we compared all correct with all incorrect trials, regardless of subsequent bets (i.e., high and low bet trials pooled). First, we focused on neuronal activity related to the visual target. Using a similar masked target task, Thompson and Schall (1999) demonstrated that signals predictive of a monkey’s decision occur in the early visual responses of FEF neurons, prior to the start of motor-related processes (reviewed by Schall and Thompson, 1999; Schall, 2001; see also Schall et al., 1995; Sato and Schall, 2003).

These brain areas coincide with the so-called “default-mode network,” a system preferentially active when subjects engage in internal rather than external processes (Buckner et al., 2008). We hope to impress upon the reader the wealth of findings that can be revealed simply by unhiding data. To encourage the use of this approach, we provide sample MATLAB scripts for hue and transparency coding on our website (http://mialab.mrn.org/datavis). Along with increased annotation,

panel B also displays the beta parameters for individual subjects, averaged over VX-770 solubility dmso clusters of voxels passing significance (Figures 3Bb1 and 3Bb2). The 2D plots remove dependence on color mapping (which is more difficult for viewers to decode than position along an axis; selleckchem Cleveland and McGill, 1985) and allow us to access the data in greater detail. Scatter plots indicate the beta estimates

for each condition (rather than just the difference), reveal the degree of variability across subjects (and the absence of outliers), and validate our “paired” statistical approach, because beta values covary across conditions. A single figure may portray experimental data painstakingly collected over months or even years. Rather than use standard designs such as bar plots and thresholded maps that hide these data, we, as authors, peer reviewers, and editors, can establish new standards for visualizations that reveal data and inform readers. We thank Christian Habeck and James Moeller for commentary that helped to motivate this work, Tom Eichele for his contribution of the EEG data, and Kent Kiehl and Godfrey Pearlson for their contribution

of the fMRI data. We also thank Christian Habeck and Tom Eichele for valuable discussions throughout the completion of this work. “
“All sensory neurons not are alike. Each detects a physical stimulus and produces an electrical signal that gives rise to behavioral responses, conscious perceptions, or both. Many operate near the physical limits of detection and operate over a dynamic range of several orders of magnitude (Bialek, 1987 and Block, 1992). These properties suggest that they are endowed with a detector, an amplifier, and mechanisms for gain control. One of the most striking and well-understood examples is the ability of photoreceptors to detect single photons while retaining sensitivity to light intensities that vary by nine orders of magnitude (Rieke and Rudd, 2009). Each somatosensory neuron is distinct. The somatosensory system is a collection of neurons innervating the skin, muscle, joints, tendons, and internal organs that establish and maintain sensitivity across a range of stimulus intensities and frequencies. This collection includes nociceptors that require stimulation above a high threshold for activation.

It accepts that each neuron in the subpopulation is well approximated by a set of NLN parameters, but that many of these myriad parameters are highly idiosyncratic to each subpopulation. Our hypothesis is that each ventral stream cortical subpopulation uses at least three common, genetically encoded mechanisms (described below) to carry out that meta job description and that together, those mechanisms direct it to “choose” a set of input weights, a normalization pool, and a static Vismodegib price nonlinearity that lead to improved subspace

untangling. Specifically, we postulate the existence of the following three key conceptual mechanisms: (1) Each subpopulation sets up architectural nonlinearities that naturally tend to flatten object manifolds. Specifically, even with random (nonlearned) filter weights, NLN-like models tend to produce easier-to-decode object identity manifolds largely on the strength of the normalization operation (Jarrett et al., 2009, Lewicki and Sejnowski, 2000, Olshausen and Field, 2005 and Pinto et al., 2008b), similar in spirit to the overcomplete approach of V1 (described above). Experimental approaches are effective at describing undocumented behaviors of ventral stream neurons, but alone they cannot indicate when that search is complete.

Similarly, “word models” (including ours, above) are not falsifiable Y-27632 molecular weight algorithms. To make progress, we need to construct ventral-stream-inspired, instantiated computational models and compare their performance

with neuronal data and human performance on object recognition tasks. Thus, computational modeling cannot be taken lightly. Together, the set of alternative models define the space of falsifiable alternative hypotheses in the field, and the success of some such algorithms will be among our first indications that we are on the path to understanding visual object recognition in the brain. The idea of using biologically inspired, hierarchical computational algorithms to understand the neuronal mechanisms underlying invariant object recognition tasks is not new: “The mechanism of pattern recognition in the brain is little known, and it seems to be almost impossible to reveal it only by conventional physiological experiments…. If we could make a neural network model which has the same capability for pattern recognition as a human Rutecarpine being, it would give us a powerful clue to the understanding of the neural mechanism in the brain” ( Fukushima, 1980). More recent modeling efforts have significantly refined and extended this approach (e.g., Lecun et al., 2004, Mel, 1997, Riesenhuber and Poggio, 1999b and Serre et al., 2007a). While we cannot review all the computer vision or neural network models that have relevance to object recognition in primates here, we refer the reader to reviews by Bengio, 2009, Edelman, 1999 and Riesenhuber and Poggio, 2000, and Zhu and Mumford (2006).