skip to main content

This content will become publicly available on January 1, 2023

Title: Strategies and Discoveries Leading to the Synthesis of Trichoaurantianolide Natural Products
Several years ago, a small family of diterpenoid natural products attracted our attention as novel targets for synthesis studies. Initially, four compounds were independently characterized by the research teams of Vidari1 and Steglich.2 Trichoaurantianolides AeD (1e4 of Fig. 9.1) were isolated from fruiting bodies of the mushrooms Tricholoma aurantium and Tricholoma fracticum in 1995. Subsequent efforts of Stermer and coworkers3 described the isolation of the closely related lepistal (5) and lepistol (6) of Fig. 9.2 as the corresponding C8 deoxygenated compounds of this family. In addition, the corresponding acetate of trichoaurantianolide B was discovered and named as 6-O-aetyl- trichoaurantin (7).2 Structure assignments were based upon extensive nuclear magnetic resonance (NMR) studies, and the features of relative stereo- chemistry were confirmed by an X-ray crystallographic analysis of trichoaurantianolide B (2).1b,2 These original investigators described the trichoaurantianolides as examples of a new class of diterpenes named as neodolastanes that signified a structural relationship to the tricyclic metabo- lites of marine origins known as dolastanes as represented by dolatriol (8)4 and the clavularane 95 of Fig. 9.2. Neodolastanes were defined as substances in which the bridgehead methyl substituent appears in a vicinal relationship with respect to the isopropyl group as exemplified in 4,5-deoxyneodolabelline (10) more » of Fig. 9.2, a related class of marine natural products.6 Steglich and coworkers2 also indicated an assignment of absolute stereo- chemistry for 2 that was based on Hamilton’s applications of linear-hypothesis testing of crystallographic data. This seldom-used technique was in agreement with the proposed absolute configuration of 2 that was advanced by Vidari, based on an assessment of the observed Cotton effects in CD spectroscopy. In 2003, Ohta and coworkers7 reported the discovery of related neodolastanes tricholomalides A, B, and C (structures 11, 12, and 13 of Fig. 9.3) from Tricholoma sp. They concluded that the tricholomalides possessed the opposite absolute configuration claimed for the trichoaurantianolides. This conclusion was based upon the independent analysis of their circular dichroism studies. By application of the octant rule for substituent effects on cisoid a,b- unsaturated ketones,8 Ohta and coworkers suggested a revision of the prior assignment of absolute configuration for the trichoaurantianolides. This asser- tion was advanced in spite of the consistently positive specific rotations recorded in different solvents for trichoaurantianolides A, B, and C1,2 versus the negative values of tricholomalides A (11) and B (12) (compare values in Figs. 9.1 and 9.3). Note that tricholomalide C (13) only differs from trichoaurantianolide B (2) as a C-8 diastereomeric alcohol, presented in the antipodal series. The specific rotation of 13 was of little value since it was recorded as [a]0 (c 0.01, MeOH).7 In 2006, Danishefsky described a pathway for the total synthesis of racemic tricholomalides A and B, and this effort led to a revision of the relative C-2 stereochemistry (Fig. 9.3; revised structures 14 and 15).9 It seemed rather unusual that genetically similar fungi would produce closely related metabolites as enantiomers, but certainly this is not unprecedented. As a starting point, this issue lacked clarity, and we concluded that our synthesis plans must unambig- uously address the issues of absolute configuration. The chemistry of dolabellane and dolastane diterpenes has been reviewed.10 The proposed pathway for biosynthesis of the trichoaurantianolides and related compounds (Fig. 9.4) follows an established sequence from geranyl- geranyl pyrophosphate (16), which undergoes p-cation cyclization to initially form the eleven-membered ring of 17. The event is followed by a second cyclization to form the dolabellane cation 18, and this [9.3.0]cyclotetradecane skeleton is central to several families of natural products. Direct capture or elimination from 18 leads to the 3,7-dolabelladiene 19, which presents the most common pattern of unsaturation within this class. Compounds within this group are traditionally numbered beginning with C-1 as the bridgehead carbon bearing the methyl group rather than following the connectivity presented in ger- anylgeranyl 16. The cation 18 also undergoes a 1,2-hydrogen migration and elimination, which leads to a transannular cyclization yielding the 5e7e6 tri- cyclic dolastane 20. The secodolastanes, represented by 21, are a small collec- tion of marine natural products, which arise from oxidative cleavage of C10eC14 in the parent tricycle 20. In analogous fashion, the neodolabellane structure 22 is produced from 18 by stereospecific backbone migrations that result in the vicinal placement of the bridgehead methyl and isopropyl substituents. Transannular cyclizations, stemming from 22, yield the class of neodolastane diterpenes (23). Trichoaurantianolides and the related lepistal A (5) are the result of oxidations and cleavage of the C-ring (C4eC5) of 23, which leads to the features of an unusual butyrolactone system. The guanacastepenes, such as 24,11 and heptemerones, such as 25,12 are primary examples of the 5e7e6 neodolastane family, and these metabolites have also been isolated from fungi sources. A characteristic structural feature is the vicinal, syn-relationship of the bridgehead methyl and isopropyl sub- stituents as compared with the 1,3-trans relationship found in dolastanes (Fig. 9.2, structures 8 and 9). Guanacastepenes have proven to be attractive targets for synthesis studies.11,13 However, these fungal metabolites exhibit the antipodal, absolute stereochemistry as compared with neodolastanes from marine origins, such as sphaerostanol (26) (Fig. 9.5).14 « less
Authors:
Editors:
Harmata, Michael
Award ID(s):
1665356
Publication Date:
NSF-PAR ID:
10320996
Journal Name:
Strategies and tactics in organic synthesis
Volume:
15
ISSN:
1874-6004
Sponsoring Org:
National Science Foundation
More Like this
  1. Obeid, Iyad Selesnick (Ed.)
    Electroencephalography (EEG) is a popular clinical monitoring tool used for diagnosing brain-related disorders such as epilepsy [1]. As monitoring EEGs in a critical-care setting is an expensive and tedious task, there is a great interest in developing real-time EEG monitoring tools to improve patient care quality and efficiency [2]. However, clinicians require automatic seizure detection tools that provide decisions with at least 75% sensitivity and less than 1 false alarm (FA) per 24 hours [3]. Some commercial tools recently claim to reach such performance levels, including the Olympic Brainz Monitor [4] and Persyst 14 [5]. In this abstract, we describe our efforts to transform a high-performance offline seizure detection system [3] into a low latency real-time or online seizure detection system. An overview of the system is shown in Figure 1. The main difference between an online versus offline system is that an online system should always be causal and has minimum latency which is often defined by domain experts. The offline system, shown in Figure 2, uses two phases of deep learning models with postprocessing [3]. The channel-based long short term memory (LSTM) model (Phase 1 or P1) processes linear frequency cepstral coefficients (LFCC) [6] features from each EEGmore »channel separately. We use the hypotheses generated by the P1 model and create additional features that carry information about the detected events and their confidence. The P2 model uses these additional features and the LFCC features to learn the temporal and spatial aspects of the EEG signals using a hybrid convolutional neural network (CNN) and LSTM model. Finally, Phase 3 aggregates the results from both P1 and P2 before applying a final postprocessing step. The online system implements Phase 1 by taking advantage of the Linux piping mechanism, multithreading techniques, and multi-core processors. To convert Phase 1 into an online system, we divide the system into five major modules: signal preprocessor, feature extractor, event decoder, postprocessor, and visualizer. The system reads 0.1-second frames from each EEG channel and sends them to the feature extractor and the visualizer. The feature extractor generates LFCC features in real time from the streaming EEG signal. Next, the system computes seizure and background probabilities using a channel-based LSTM model and applies a postprocessor to aggregate the detected events across channels. The system then displays the EEG signal and the decisions simultaneously using a visualization module. The online system uses C++, Python, TensorFlow, and PyQtGraph in its implementation. The online system accepts streamed EEG data sampled at 250 Hz as input. The system begins processing the EEG signal by applying a TCP montage [8]. Depending on the type of the montage, the EEG signal can have either 22 or 20 channels. To enable the online operation, we send 0.1-second (25 samples) length frames from each channel of the streamed EEG signal to the feature extractor and the visualizer. Feature extraction is performed sequentially on each channel. The signal preprocessor writes the sample frames into two streams to facilitate these modules. In the first stream, the feature extractor receives the signals using stdin. In parallel, as a second stream, the visualizer shares a user-defined file with the signal preprocessor. This user-defined file holds raw signal information as a buffer for the visualizer. The signal preprocessor writes into the file while the visualizer reads from it. Reading and writing into the same file poses a challenge. The visualizer can start reading while the signal preprocessor is writing into it. To resolve this issue, we utilize a file locking mechanism in the signal preprocessor and visualizer. Each of the processes temporarily locks the file, performs its operation, releases the lock, and tries to obtain the lock after a waiting period. The file locking mechanism ensures that only one process can access the file by prohibiting other processes from reading or writing while one process is modifying the file [9]. The feature extractor uses circular buffers to save 0.3 seconds or 75 samples from each channel for extracting 0.2-second or 50-sample long center-aligned windows. The module generates 8 absolute LFCC features where the zeroth cepstral coefficient is replaced by a temporal domain energy term. For extracting the rest of the features, three pipelines are used. The differential energy feature is calculated in a 0.9-second absolute feature window with a frame size of 0.1 seconds. The difference between the maximum and minimum temporal energy terms is calculated in this range. Then, the first derivative or the delta features are calculated using another 0.9-second window. Finally, the second derivative or delta-delta features are calculated using a 0.3-second window [6]. The differential energy for the delta-delta features is not included. In total, we extract 26 features from the raw sample windows which add 1.1 seconds of delay to the system. We used the Temple University Hospital Seizure Database (TUSZ) v1.2.1 for developing the online system [10]. The statistics for this dataset are shown in Table 1. A channel-based LSTM model was trained using the features derived from the train set using the online feature extractor module. A window-based normalization technique was applied to those features. In the offline model, we scale features by normalizing using the maximum absolute value of a channel [11] before applying a sliding window approach. Since the online system has access to a limited amount of data, we normalize based on the observed window. The model uses the feature vectors with a frame size of 1 second and a window size of 7 seconds. We evaluated the model using the offline P1 postprocessor to determine the efficacy of the delayed features and the window-based normalization technique. As shown by the results of experiments 1 and 4 in Table 2, these changes give us a comparable performance to the offline model. The online event decoder module utilizes this trained model for computing probabilities for the seizure and background classes. These posteriors are then postprocessed to remove spurious detections. The online postprocessor receives and saves 8 seconds of class posteriors in a buffer for further processing. It applies multiple heuristic filters (e.g., probability threshold) to make an overall decision by combining events across the channels. These filters evaluate the average confidence, the duration of a seizure, and the channels where the seizures were observed. The postprocessor delivers the label and confidence to the visualizer. The visualizer starts to display the signal as soon as it gets access to the signal file, as shown in Figure 1 using the “Signal File” and “Visualizer” blocks. Once the visualizer receives the label and confidence for the latest epoch from the postprocessor, it overlays the decision and color codes that epoch. The visualizer uses red for seizure with the label SEIZ and green for the background class with the label BCKG. Once the streaming finishes, the system saves three files: a signal file in which the sample frames are saved in the order they were streamed, a time segmented event (TSE) file with the overall decisions and confidences, and a hypotheses (HYP) file that saves the label and confidence for each epoch. The user can plot the signal and decisions using the signal and HYP files with only the visualizer by enabling appropriate options. For comparing the performance of different stages of development, we used the test set of TUSZ v1.2.1 database. It contains 1015 EEG records of varying duration. The any-overlap performance [12] of the overall system shown in Figure 2 is 40.29% sensitivity with 5.77 FAs per 24 hours. For comparison, the previous state-of-the-art model developed on this database performed at 30.71% sensitivity with 6.77 FAs per 24 hours [3]. The individual performances of the deep learning phases are as follows: Phase 1’s (P1) performance is 39.46% sensitivity and 11.62 FAs per 24 hours, and Phase 2 detects seizures with 41.16% sensitivity and 11.69 FAs per 24 hours. We trained an LSTM model with the delayed features and the window-based normalization technique for developing the online system. Using the offline decoder and postprocessor, the model performed at 36.23% sensitivity with 9.52 FAs per 24 hours. The trained model was then evaluated with the online modules. The current performance of the overall online system is 45.80% sensitivity with 28.14 FAs per 24 hours. Table 2 summarizes the performances of these systems. The performance of the online system deviates from the offline P1 model because the online postprocessor fails to combine the events as the seizure probability fluctuates during an event. The modules in the online system add a total of 11.1 seconds of delay for processing each second of the data, as shown in Figure 3. In practice, we also count the time for loading the model and starting the visualizer block. When we consider these facts, the system consumes 15 seconds to display the first hypothesis. The system detects seizure onsets with an average latency of 15 seconds. Implementing an automatic seizure detection model in real time is not trivial. We used a variety of techniques such as the file locking mechanism, multithreading, circular buffers, real-time event decoding, and signal-decision plotting to realize the system. A video demonstrating the system is available at: https://www.isip.piconepress.com/projects/nsf_pfi_tt/resources/videos/realtime_eeg_analysis/v2.5.1/video_2.5.1.mp4. The final conference submission will include a more detailed analysis of the online performance of each module. ACKNOWLEDGMENTS Research reported in this publication was most recently supported by the National Science Foundation Partnership for Innovation award number IIP-1827565 and the Pennsylvania Commonwealth Universal Research Enhancement Program (PA CURE). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the official views of any of these organizations. REFERENCES [1] A. Craik, Y. He, and J. L. Contreras-Vidal, “Deep learning for electroencephalogram (EEG) classification tasks: a review,” J. Neural Eng., vol. 16, no. 3, p. 031001, 2019. https://doi.org/10.1088/1741-2552/ab0ab5. [2] A. C. Bridi, T. Q. Louro, and R. C. L. Da Silva, “Clinical Alarms in intensive care: implications of alarm fatigue for the safety of patients,” Rev. Lat. Am. Enfermagem, vol. 22, no. 6, p. 1034, 2014. https://doi.org/10.1590/0104-1169.3488.2513. [3] M. Golmohammadi, V. Shah, I. Obeid, and J. Picone, “Deep Learning Approaches for Automatic Seizure Detection from Scalp Electroencephalograms,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York, New York, USA: Springer, 2020, pp. 233–274. https://doi.org/10.1007/978-3-030-36844-9_8. [4] “CFM Olympic Brainz Monitor.” [Online]. Available: https://newborncare.natus.com/products-services/newborn-care-products/newborn-brain-injury/cfm-olympic-brainz-monitor. [Accessed: 17-Jul-2020]. [5] M. L. Scheuer, S. B. Wilson, A. Antony, G. Ghearing, A. Urban, and A. I. Bagic, “Seizure Detection: Interreader Agreement and Detection Algorithm Assessments Using a Large Dataset,” J. Clin. Neurophysiol., 2020. https://doi.org/10.1097/WNP.0000000000000709. [6] A. Harati, M. Golmohammadi, S. Lopez, I. Obeid, and J. Picone, “Improved EEG Event Classification Using Differential Energy,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium, 2015, pp. 1–4. https://doi.org/10.1109/SPMB.2015.7405421. [7] V. Shah, C. Campbell, I. Obeid, and J. Picone, “Improved Spatio-Temporal Modeling in Automated Seizure Detection using Channel-Dependent Posteriors,” Neurocomputing, 2021. [8] W. Tatum, A. Husain, S. Benbadis, and P. Kaplan, Handbook of EEG Interpretation. New York City, New York, USA: Demos Medical Publishing, 2007. [9] D. P. Bovet and C. Marco, Understanding the Linux Kernel, 3rd ed. O’Reilly Media, Inc., 2005. https://www.oreilly.com/library/view/understanding-the-linux/0596005652/. [10] V. Shah et al., “The Temple University Hospital Seizure Detection Corpus,” Front. Neuroinform., vol. 12, pp. 1–6, 2018. https://doi.org/10.3389/fninf.2018.00083. [11] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. https://dl.acm.org/doi/10.5555/1953048.2078195. [12] J. Gotman, D. Flanagan, J. Zhang, and B. Rosenblatt, “Automatic seizure detection in the newborn: Methods and initial evaluation,” Electroencephalogr. Clin. Neurophysiol., vol. 103, no. 3, pp. 356–362, 1997. https://doi.org/10.1016/S0013-4694(97)00003-9.« less
  2. Simmons, Lyle A. ; Bush, Karen (Ed.)
    ABSTRACT Unique DNA repair enzymes that provide self-resistance against therapeutically important, genotoxic natural products have been discovered in bacterial biosynthetic gene clusters (BGCs). Among these, the DNA glycosylase AlkZ is essential for azinomycin B production and belongs to the HTH_42 superfamily of uncharacterized proteins. Despite their widespread existence in antibiotic producers and pathogens, the roles of these proteins in production of other natural products are unknown. Here, we determine the evolutionary relationship and genomic distribution of all HTH_42 proteins from Streptomyces and use a resistance-based genome mining approach to identify homologs associated with known and uncharacterized BGCs. We find that AlkZ-like (AZL) proteins constitute one distinct HTH_42 subfamily and are highly enriched in BGCs and variable in sequence, suggesting each has evolved to protect against a specific secondary metabolite. As a validation of the approach, we show that the AZL protein, HedH4, associated with biosynthesis of the alkylating agent hedamycin, excises hedamycin-DNA adducts with exquisite specificity and provides resistance to the natural product in cells. We also identify a second, phylogenetically and functionally distinct subfamily whose proteins are never associated with BGCs, are highly conserved with respect to sequence and genomic neighborhood, and repair DNA lesions not associated with amore »particular natural product. This work delineates two related families of DNA repair enzymes—one specific for complex alkyl-DNA lesions and involved in self-resistance to antimicrobials and the other likely involved in protection against an array of genotoxins—and provides a framework for targeted discovery of new genotoxic compounds with therapeutic potential. IMPORTANCE Bacteria are rich sources of secondary metabolites that include DNA-damaging genotoxins with antitumor/antibiotic properties. Although Streptomyces produce a diverse number of therapeutic genotoxins, efforts toward targeted discovery of biosynthetic gene clusters (BGCs) producing DNA-damaging agents is lacking. Moreover, work on toxin-resistance genes has lagged behind our understanding of those involved in natural product synthesis. Here, we identified over 70 uncharacterized BGCs producing potentially novel genotoxins through resistance-based genome mining using the azinomycin B-resistance DNA glycosylase AlkZ. We validate our analysis by characterizing the enzymatic activity and cellular resistance of one AlkZ ortholog in the BGC of hedamycin, a potent DNA alkylating agent. Moreover, we uncover a second, phylogenetically distinct family of proteins related to Escherichia coli YcaQ, a DNA glycosylase capable of unhooking interstrand DNA cross-links, which differs from the AlkZ-like family in sequence, genomic location, proximity to BGCs, and substrate specificity. This work defines two families of DNA glycosylase for specialized repair of complex genotoxic natural products and generalized repair of a broad range of alkyl-DNA adducts and provides a framework for targeted discovery of new compounds with therapeutic potential.« less
  3. Abstract
    Excessive phosphorus (P) applications to croplands can contribute to eutrophication of surface waters through surface runoff and subsurface (leaching) losses. We analyzed leaching losses of total dissolved P (TDP) from no-till corn, hybrid poplar (Populus nigra X P. maximowiczii), switchgrass (Panicum virgatum), miscanthus (Miscanthus giganteus), native grasses, and restored prairie, all planted in 2008 on former cropland in Michigan, USA. All crops except corn (13 kg P ha−1 year−1) were grown without P fertilization. Biomass was harvested at the end of each growing season except for poplar. Soil water at 1.2 m depth was sampled weekly to biweekly for TDP determination during March–November 2009–2016 using tension lysimeters. Soil test P (0–25 cm depth) was measured every autumn. Soil water TDP concentrations were usually below levels where eutrophication of surface waters is frequently observed (> 0.02 mg L−1) but often higher than in deep groundwater or nearby streams and lakes. Rates of P leaching, estimated from measured concentrations and modeled drainage, did not differ statistically among cropping systems across years; 7-year cropping system means ranged from 0.035 to 0.072 kg P ha−1 year−1 with large interannual variation. Leached P was positively related to STP, which decreased over the 7 years in all systems. These results indicate that both P-fertilized and unfertilized cropping systems mayMore>>
  4. Recently, over 200 molecules have been detected in the interstellar medium (ISM), with about one third being complex organic molecules (COMs), molecules containing six or more atoms. Over the last few decades, astrophysical laboratory experiments have shown that several COMs are formed via interaction of ionizing radiation within ices deposited on interstellar dust particles at 10 K (H 2 O, CH 3 OH, CO, CO 2 , CH 4 , NH 3 ). However, there is still a lack of understanding of the chemical complexity that is available through individual ice constituents. The present research investigates experimentally the synthesis of carbon, hydrogen, and oxygen bearing COMs from interstellar ice analogues containing carbon monoxide (CO) and methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), or acetylene (C 2 H 2 ) exposed to ionizing radiation. Utilizing online and in situ techniques, such as infrared spectroscopy and tunable photoionization reflectron time-of-flight mass spectrometry (PI-ReTOF-MS), specific isomers produced could be characterized. A total of 12 chemically different groups were detected corresponding to C 2 H n O ( n = 2, 4, 6), C 3 H n O ( n = 2, 4, 6, 8),more »C 4 H n O ( n = 4, 6, 8, 10), C 5 H n O ( n = 4, 6, 8, 10), C 6 H n O ( n = 4, 6, 8, 10, 12, 14), C 2 H n O 2 ( n = 2, 4), C 3 H n O 2 ( n = 4, 6, 8), C 4 H n O 2 ( n = 4, 6, 8, 10), C 5 H n O 2 ( n = 6, 8), C 6 H n O 2 ( n = 8, 10, 12), C 4 H n O 3 ( n = 4, 6, 8), and C 5 H n O 3 ( n = 6, 8). More than half of these isomer specifically identified molecules have been identified in the ISM, and the remaining COMs detected here can be utilized to guide future astronomical observations. Of these isomers, three groups – alcohols, aldehydes, and molecules containing two of these functional groups – displayed varying degrees of unsaturation. Also, the detection of 1-propanol, 2-propanol, 1-butanal, and 2-methyl-propanal has significant implications as the propyl and isopropyl moieties (C 3 H 7 ), which have already been detected in the ISM via propyl cyanide and isopropyl cyanide, could be detected in our laboratory studies. General reaction mechanisms for their formation are also proposed, with distinct follow-up studies being imperative to elucidate the complexity of COMs synthesized in these ices.« less
  5. Two bis-carbamoylmethylphosphine oxide compounds, namely {[(3-{[2-(diphenylphosphinoyl)ethanamido]methyl}benzyl)carbamoyl]methyl}diphenylphosphine oxide, C 36 H 34 N 2 O 4 P 2 , (I), and diethyl [({2-[2-(diethoxyphosphinoyl)ethanamido]ethyl}carbamoyl)methyl]phosphonate, C 14 H 30 N 2 O 8 P 2 , (II), were synthesized via nucleophilic acyl substitution reactions between an ester and a primary amine. Hydrogen-bonding interactions are present in both crystals, but these interactions are intramolecular in the case of compound (I) and intermolecular in compound (II). Intramolecular π–π stacking interactions are also present in the crystal of compound (I) with a centroid–centroid distance of 3.9479 (12) Å and a dihedral angle of 9.56 (12)°. Intermolecular C—H...π interactions [C...centroid distance of 3.622 (2) Å, C—H...centroid angle of 146°] give rise to supramolecular sheets that lie in the ab plane. Key geometric features for compound (I) involve a nearly planar, trans- amide group with a C—N—C—C torsion angle of 169.12 (17)°, and a torsion angle of −108.39 (15)° between the phosphine oxide phosphorus atom and the amide nitrogen atom. For compound (II), the electron density corresponding to the phosphoryl group was disordered, and was modeled as two parts with a 0.7387 (19):0.2613 (19) occupancy ratio. Compound (II) also boasts a trans -amide group that approaches planarity with a C—N—C—C torsion angle of −176.50 (16)°. The hydrogen bondsmore »in this structure are intermolecular, with a D ... A distance of 2.883 (2) Å and a D —H... A angle of 175.0 (18)° between the amide hydrogen atom and the P=O oxygen atom. These non-covalent interactions create ribbons that run along the b -axis direction.« less