Zeolites and related microporous materials have been widely used as selective catalysts, absorbents, and ion-exchangers1,2,3,4. Recently zeolites have gained interest as catalysts for biomass conversion5,6,7,8 and automotive exhaust remediation9,10. The unique properties of zeolites are associated with the well-defined pores and channels of molecular dimensions. Because of the close structure-property relationship, numerous efforts have been made to synthesize zeolitic materials with new pore structures and compositions and subsequently explore their properties and applications. Currently, only 252 zeolite framework topologies have been approved by the International Zeolite Association11, even though millions of possible zeolite topologies have been identified by computational methods12,13. The huge gap between experimental output and theoretical prediction indicates that our understanding of zeolite crystallization is still limited. It is, therefore, important to push the field by developing rational strategies to synthesize new zeolites. Besides the conventional hydro-/solvothermal synthesis, several new synthetic strategies have been proposed for targeted zeolite synthesis. These include bottom-up approaches that use, for example, pre-designed organic structure-directing agents (OSDAs)14,15,16,17 and promotion of specific cavities as in the synthesis of embedded isoreticular zeolites18,19, and top-down approaches, i.e., topotactic transformations. The most common top-down approach to synthesize new zeolites has been two-dimensional–three-dimensional (2D–3D) layer condensation from a layered precursor20,21,22,23. Recently, a new approach of using known zeolites as precursors to synthesize new zeolites via 3D–2D–3D topotactic transformation has gained significant attention24,25,26,27. The most successful example of these methods is the assembly–disassembly–organization–reassembly (ADOR) synthesis route. The unique concept of the ADOR process is to exploit chemically selective weak T–O bonds (T = Al, Si, Ge, etc.), e.g., Ge–O bonds in germanosilicates, in known zeolites and subsequently promote the formation of new T–O bonds. The ADOR strategy provides a synthetic tool for the generation of zeolites with novel topologies and compositions that cannot be synthesized by conventional hydrothermal methods. To the best of our knowledge, these targeted synthesis strategies have only been demonstrated on silica-based zeolite syntheses (e.g., silicates, aluminosilicates and germanosilicates), but never on phosphate-based frameworks.

Aluminophosphates (known as AlPO4-n), first synthesized by Union Carbide28, are an important class of zeolitic molecular sieves with wide applications. AlPO4 molecular sieves have a unique feature, which allows substitution of the Al or P by main block elements (e.g., Si) or transition metals to engender specific properties in the material29,30. For example, silicoaluminophosphate SAPO-34 (framework type CHA) is one of the most widely used solid catalysts for biomass conversion, with a global market of tens of billions of US dollars per year. Currently, only 20% of the zeolite framework types can be synthesized as AlPO4 phases11, and examples of rational synthesis of AlPO4 materials are very rare31. In AlPO4 frameworks, the AlO4 and PO4 tetrahedra always alternate according to Löwenstein’s rule32. Nevertheless, Al can adopt penta-coordination with a pair of Al atoms bridged by a hydroxyl group as found in AlPO4-21 (AWO)33 and hexa-coordination with two additional water molecules, for example, in AlPO4-C (APC)34. Upon the loss of water molecules by heating, AlPO4-21 and AlPO4-C transform into new fully tetrahedrally coordinated zeolite frameworks AlPO4-25 (ATV) and AlPO4-D (APD), respectively. In 2014, we reported the synthesis and structure of a new AlPO4 zeolite denoted PST-635. The framework of PST-6 (PSI) has a one-dimensional (1D) pore system containing parallel 8- and 10-ring channels. PST-6 was obtained by calcination of PST-5 at 500 °C, which was synthesized using diethylamine (DEA) as an OSDA.

Both PST-5 and PST-6 are obtained as nano-size plate-like crystals (Supplementary Fig. 1). The structure of PST-6 was successfully solved by combining electron diffraction and powder X-ray diffraction (PXRD). However, the poorer stability of PST-5 under the electron beam prevented its structure determination, and therefore the understanding of its structure transformation to PST-6. Solid-state nuclear magnetic resonance (NMR) spectroscopy indicated the presence of bridging hydroxyl groups in PST-5.

Recent developments of fast 3D electron diffraction (3DED) or microcrystal electron diffraction (MicroED) data collection techniques (<5 min) using continuous rotation (denoted cRED here) have made it possible for studying beam sensitive nano- and micrometer-sized crystals18,36,37,38,39,40. Here, we demonstrate the ab initio structure determination of PST-5 using cRED data collected under low-dose condition. By detailed investigations of the structure transformation of PST-5 to PST-6, we propose an approach to generate novel zeolite structures from known zeolites via 3D–3D topotactic transformation. The transformation is initiated via the weak Al–O bonding in penta-coordinated AlO4(OH) pairs or hexa-coordinated AlO4(OH)2 units. By adopting this approach, we have, in addition, predicted a series of targeted hypothetical zeolite structures.


Structure of PST-5

A high-resolution cRED dataset of PST-5 (with a resolution better than 0.85 Å) was collected from a sub-micrometer-sized crystal using low dose and fast data collection (in total 15 e Å−2 and 3 min) on a JEOL JEM2100 transmission electron microscope (Fig. 1a and Supplementary Fig. 2). PST-5 crystallizes in an orthorhombic space group Pb2b (No. 27), with the unit cell parameters a = 36.5956(2) Å, b = 21.80273(9) Å, c = 10.26929(4) Å (Supplementary Tables 1 and 2). Its framework structure was first solved ab initio by direct methods. To confirm the model and obtain more accurate positions and atomic displace parameters (ADPs) of individual atoms, the model was further refined against the cRED data (Supplementary Table 2). Details of structure determination are given in the Supplementary Information. The framework of PST-5 exhibits a totally new topology (Supplementary Figs. 4 and 5, see Supplementary Information for more details) with a fully four-connected 3D net and a 2D pore system containing parallel 8-ring (5.5 × 2.6 Å) and 10-ring (5.8 × 4.4 Å) channels along the c-axis, and another 8-ring (3.3 × 3.3 Å) channel along the b-axis (Fig. 2). PST-5 is one of the most complex zeolite structures, with 18 Al, 18 P, and 76 O atoms in the asymmetric unit. Four of the O atoms belong to hydroxyl groups and act as bridging atoms to form Al–OH–Al linkages. All 18 P atoms and 10 out of the 18 Al atoms are tetrahedrally coordinated with O atom to form TO4 units (T = Al, P). The remaining eight Al atoms have trigonal bipyramidal-coordination and form four symmetry-independent AlO4-OH-AlO4 pairs in the structure. Each AlO4-OH-AlO4 pair becomes part of a 3- and a 5-ring (Supplementary Fig. 6). Solid-state NMR spectroscopy confirmed the presence of Al–O–Al linkages (Supplementary Figs. 79), and a sharp Infrared (IR) band appearing at 3398 cm−1 indicates the presence of bridging OH groups (Supplementary Fig. 10). The structure model obtained by Rietveld refinement against synchrotron PXRD data agrees well with that obtained from cRED data, which further confirms the framework structure of PST-5. In addition, the OSDA positions in the channels could be located from both cRED and PXRD data, and subsequently refined by Rietveld refinement. The final refined unit cell composition is |(C2H5)2NH2+)16 |[Al72P72O288(OH)16] (Fig. 1b, Supplementary Figs. 11 and 12 and Supplementary Table 3, see Supplementary Information for more details). Raman spectrum shows that DEA molecules are protonated during PST-5 crystallization35. Therefore, the negative framework charges created by the bridging Al–OH–Al groups are balanced by the protonated form of the occluded OSDA molecules (DEA∙H+), as reflected by quite similar numbers of OSDA molecules and bridging O atoms.

Fig. 1: Reconstructed 3D reciprocal lattice and Rietveld refinement.
figure 1

a Reconstructed three-dimensional reciprocal lattice of PST-5 from the continuous rotation electron diffraction (cRED) data viewed along the [100] direction. Inset is the crystal of PST-5 from which the cRED data was collected, scale bar = 500 nm. b Powder X-ray diffraction (PXRD) profiles of Rietveld refinement for PST-5 (λ = 1.5474 Å): observed, calculated and difference profiles are shown in red, black, and blue, respectively. The bars represent the positions of allowed reflections.

Fig. 2: Comparison of the structures of PST-5 and PST-6.
figure 2

Structural models of PST-5 (a1) and PST-6 (b1) viewed along the c-axis. Both structures are built from the same building layer containing 4-, 6-, 8-, and 10-rings (if the OH groups, shown in red are not considered). The difference is the orientations of the (Al,P)O4 tetrahedra in the two structures. The red atoms in (a1) correspond to the OH groups bridging two Al atoms in PST-5. These Al atoms are five-coordinated. Structural models of PST-5 (a2) and PST-6 (b2) viewed along the b-axis, showing the transforms of double-crankshaft chains (dcc) and double-4-rings (d4r) in PST-5 (a1) to narsarsukite-type chains (nsc) in PST-6 (indicated by purple boxes). c Building units in PST-5 (dcc and d4r) and PST-6 (nsc).

3D–3D topotactic transformation of PST-5 to PST-6

In situ PXRD patterns indicate that the structural transformation of PST-5 to PST-6 starts at 200 °C, and finishes at 500 °C (Supplementary Fig. 13), which is accompanied by notable changes in the PXRD pattern. The observed pattern after calcination of PST-5 at 500 °C agrees very well with the simulated pattern from the PST-6 model, confirming the calcined PST-5 (i.e., PST-6) has the PSI framework (Supplementary Fig. 14). In situ IR studies show that dehydroxylation of bridging OH groups in PST-5 starts above 200 °C and finishes at 300 °C (Supplementary Fig. 10, see SI for more details). Both structures are built from a similar building layer containing 4-, 6-, 8-, and 10-rings if not considering the OH groups (Fig. 2a1, b1 and Supplementary Fig. 15). Interestingly, the transformation of PST-5 to PST-6 consists not only in the loss of the hydroxyl groups, as observed in the transformation of PST-13 to PST-1441. Additionally, it involves T–O bond-breaking and reforming, which leads to totally different connections of the building layers in the two structures (Fig. 2a2, b2) and thereby different pore systems; 1D for PST-6, in contrast to 2D for PST-5.

The mechanism for the 3D–3D topotactic transformation of PST-5 to PST-6 is investigated based on their structure relationships, as well as molecular dynamics simulations. PST-5 and PST-6 show large structure similarities when viewed along the c-axis (Fig. 2a1, b1). The framework of PST-5 consists mainly of two types of building units; double-crankshaft chain (dcc) and double-4-ring (d4r), both contain 4-rings, as shown in Fig. 2c and highlighted in Fig. 2a1. The framework of PST-6 consists mainly of one type of building unit; a narsarsukite-type chain (nsc), as shown in Fig. 2c and marked in Fig. 2b1. The transformation of PST-5 to PST-6 can be described by the dcc-nsc and d4r-nsc transformations. It is worth noting that the dcc-nsc transformation was also observed in AlPO4-21 to AlPO4-2533, and AlPO4-C to AlPO4-D34. However, the d4r-nsc transformation has not been reported before.

The key to the chain transformation is the penta-coordinated Al atoms, where Al–O bond lengths (on average 1.83(9) Å) are longer than those of tetra-coordinated Al atoms (on average 1.72(1) Å). The longer bond lengths indicate weaker bonds, which are more easily broken. Therefore, the built-in weakness associated with penta-coordinated Al atoms in the structure may be utilized to initiate the topotactic transformation. Upon thermal treatment, the bridging hydroxyl groups are removed through a dehydroxylation to form the fully four-connected framework PST-6 (Fig. 2 and Supplementary Fig. 6).

By detailed inspection of the structural relationships between PST-5 and PST-6, the geometry changes of penta-coordinated Al atoms initiate a domino geometrical inversion of nearby PO4 tetrahedra. Upon heating, the penta-coordinated Al atoms lose the bridging hydroxyl groups, which also leads to cleavage of the Al–O–P bond to the adjacent PO4 tetrahedra. The AlO4 and PO4 tetrahedral pairs with resulting broken bonds then invert their orientations and reconnect to the next nearest pairs of AlO4 and PO4 tetrahedra (Fig. 3, indicated by arrows). As a result, the double-crankshaft chains (dcc) in PST-5 transforms to the narsarsukite-type chains (nsc) in PST-6, which is composed of one chain with tetra-coordinated Al atoms, and another chain originating from penta-coordinated Al atoms (Fig. 3a1, b1). The inversion further extends to neighboring atoms, which results in a domino-like alteration of the tetrahedral conformation. There are two symmetry-independent dcc chains in PST-5 with different orientations. After being transformed to PST-6, they orient in the same way. The reconstructed PST-6 (PSI) material was found to contain stacking faults35, due to changes in both orientation and tetrahedral conformation. There are eight unique penta-coordinated Al atoms that can initiate the topotactic transformation. When the transformation starts at different positions, it will also generate disorder after the reconstruction.

Fig. 3: Chain transformations between PST-5 and PST-6.
figure 3

a1 Double-crankshaft chain (dcc) and a2 double four-ring (d4r) unit in PST-5, and b1 and b2 narsarsukite chain (nsc) in PST-6. The removal of bridging hydroxyl groups associated with the dcc and d4r initiates the geometry changes of penta-coordinated Al atoms, resulting in cleavage of Al–O–P bonds (corresponding O atoms are shown in white) followed by inversion and reconnection of the tetrahedra (indicated by arrows). Both dcc and d4r transform to nsc in PST-6. Green, penta-coordinated (and penta-coordinated derived) AlO4 tetrahedra; yellow, tetra-coordinated AlO4; blue, PO4. Oxygen atoms associated with OH or bond-breaking in the chains are shown in white and the rest O in red.

To study the possible mechanisms for the transformation of PST-5 to PST-6, periodic ab initio molecular dynamics simulations were carried out on a model PST-5 system. In nearly every case, one bond from penta-coordinated Al was broken, forming tetra-coordinated Al and corresponding P-OH groups (Supplementary Figs. 16 and 17). In one case, the Al–O–P linkage was broken within a 3- (–Al–O–P–Al–OH–) ring, but otherwise the framework interruption occurred between the penta-coordinated Al and neighboring P atoms, which were part of a double-crankshaft chain (dcc) or a double-4-ring (d4r) structural unit but not in a 3-ring. The simulations clearly confirm that weaker bonds in the PST-5 framework are associated with penta-coordinated Al atoms and are labile in the presence of an extra-framework proton donor. The initial broken bonds are those in the dcc and d4r, largely the same as those implicated in the transformation of PST-5 to PST-6 at the experimental annealing temperature. Coupled with the topological strain inherent within the dcc and d4r due to their fused 4-ring motifs, the penta-coordinated Al atoms are thus critical in introducing reactivity into the framework. It should be noted that the molecular dynamics simulations of an idealized, fully four-connected, AlPO4 PST-5 framework, without penta-coordinated Al atoms and bridging OH groups, showed no cleavage of any framework bonds over the same timescale at 500 °C.

New zeolite generation

Inspired by the 3D–3D topotatic transformation in PST-5/PST-6, we investigated the existing structures in the Zeolite Structure Database11. 12 framework types are found to contain infinite dcc, and they can be divided into three groups based on the structural complexity (Table 1): I) built solely from dcc that are connected directly (APC, GIS, MER, PHI, SIV, GME, Supplementary Fig. 18); II) built from dcc that are bridged by isolated 4-rings (ATT, AWO, UEI, Supplementary Figs. 19 and 20); III) built by dcc and other building units (PST-5, DON, STO, Supplementary Fig. 21). Interestingly, regardless of the complexity, the dcc-nsc structural transformation has been observed in all groups, from AlPO4-C (APC) to AlPO4-D (APD) in group I, AlPO4-21 (AWO) to AlPO4-25 (ATV) in group II, and PST-5 to PST-6 (PSI) in group III (Table 1 and Supplementary Table 4). The three parent frameworks all contain dcc in which penta- (AlPO4-21 and PST-5) or hexa- (AlPO4-C) coordinated Al atoms are present. By applying similar dcc to nsc transformations observed in the three known zeolite pairs to other dcc-containing zeolites, we generated a series of novel hypothetical nsc-containing zeolite structures, as shown in Fig. 4, Supplementary Figs. 1822, Table 1, and Supplementary Table 5 (see Supplementary Information for more details)42. Notably, the same parent zeolite can generate more than one hypothetical zeolites (e.g., UEI to UEI_H1-5), and the same zeolite can be generated from different parent zeolites (e.g., GIS_H1 from both GIS and SIV). The transformation can also generate new channel dimensionality (Supplementary Fig. 18 and Table 1). All these hypothetical nsc-containing zeolites fulfill the criteria for zeolite frameworks realizable for target synthesis (Supplementary Table 6)43. Most of them were calculated to have much lower framework energies than the parent dcc-containing zeolites. It is worth mentioning that because each TO4 tetrahedral unit not belonging to dcc can have two possible connectivities, either upwards or downwards, the number of hypothetical zeolites generated via this approach can be enormous for zeolites in group II and III (see Supplementary Information for more details).

Table 1 Observed (in bold) and predicted 3D–3D topotactic transformations of zeolites through chain transformation from dcc to nsc and d4r to nsc.
Fig. 4: Illustration of observed and predicted 3D–3D topotactic transformations.
figure 4

a1a10 Projections of the APC, GIS, MER, GME, ATT, AWO, UEI, PST-5, DON, and STO frameworks. All of them contain double-crankshaft chain (dcc). Among them, APC, AWO, UEI, and PST-5 contain penta- or hexa-coordinated atoms. b1, b6, and b8 Projections of APD, ATV, and PSI (PST-6) frameworks after topotactic transformation by changing dcc to narsarsukite chain (nsc). They have the same framework projections as APC, AWO, and PST-5, respectively, but differ in the 3D connectivity along the projections. b2b5, b7, b9, and b10 Selected hypothetical structures of GIS_H1, MER_H1, ATT_H1, UEI_H1, DON_H1, and STO_H1 generated by replacing dcc with nsc. Magenta: dcc; cyan: nsc; yellow: double four-ring (d4r). The solid and hollow circles indicate vertices that connect upwards and downwards, respectively.

We have demonstrated that many hypothetical zeolites can be predicted from known zeolite frameworks based on the dcc-nsc transformation. The 3D–3D topotactic transformation would provide a simple approach to generate novel zeolite frameworks, which may be inaccessible by conventional hydrothermal/solvothermal synthesis. We have also shown by molecular dynamics simulations that the weakened bonding, such as penta- or hexa-coordinated Al atoms, is crucial to initiate the 3D–3D topotactic transformation. Therefore, to synthesize new structures by this approach, it is important to introduce penta- or hexa-coordinated Al atoms in dcc-containing zeolite frameworks, and additional exploratory work is needed. Previous studies showed that certain types of OSDAs, like easily protonatable amines, could promote the formation of bridging Al–OH–Al bonds in AlPO4 molecular sieves, and penta- or hexa-coordinated Al atoms have been found in c.a. 25% of these materials, including AlPO4-17, AlPO4-41, AlPO4-EN3, AlPO-CJB1, VPI-5, IST-1, PST-13, etc.41,44,45,46,47,48,49,50. The 3D–3D topotactic transformation may also be applicable to other zeolite compositions. For example, the gallophosphate Mu-18 (UEI) contains both dcc and penta-coordinated atoms (Fig. 4a7)51. We anticipate that with the discovery of many new zeolites, the approach described here will be more feasible.


We have solved the structure of PST-5, a new 2D medium-pore AlPO4 molecular sieve, using continuous rotation electron diffraction (cRED). Based on the investigations of the dcc to nsc transformations observed in PST-5/PST-6, AlPO4-C/AlPO4-D, and AlPO4-21/AlPO4-25 pairs, we propose an approach to generate novel zeolite structures, which enriches the structural diversity. The 3D–3D topotactic transformation relies on the chemical weakness in the dcc. PST-5 transforms dcc and d4r to nsc, changes its overall interlayer connectivities and finally generates PST-6. The penta-coordinated Al atoms, as observed in PST-5, could be built into a precursor material by using pre-designed OSDAs, which balance the framework energy. This approach via the 3D–3D topotactic transformation can be considered complementary to the existing strategies of synthesizing novel zeolite structures. By understanding the 3D–3D topotactic transformation, we anticipate that our approach can be generalized and provide new insights into the discovery of novel zeolitic materials, which may enable a completely new range of zeolite frameworks to be accessible and offer opportunities in technological applications.


Synthesis of PST-5

The reagents used for the synthesis of PST-5 included diethylamine (DEA, 99.5+%, Aldrich), aluminum isopropxide (98+%, Aldrich), phosphoric acid (85%, Merck), and deionized water. PST-5 was synthesized from an AlPO4 gel with the composition of 2.0DEA·1.0Al2O3·1.0P2O5·40H2O, where DEA is diethylamine. In a typical synthesis of PST-5, 3.84 g of o-H3PO4 (85%, Merck) was diluted in 5.63 g of H2O. This solution was added dropwise to a slurry of 6.95 g of aluminum isopropoxide (>98%, Aldrich) in 5.64 g of H2O. After stirring for 1 h, 2.45 g of DEA (99.5%, Aldrich) was added to the above synthesis mixture, which was stirred overnight at room temperature. Afterwards, the mixture was charged into a Teflon-lined 23-mL autoclave and heated under static conditions at 200 °C for 5 days. The resulting white powder was recovered by filtration, washed repeatedly with water and then dried overnight at room temperature.

Topotactic transformation of PST-5

The topotactic transformation of PST-5 to PST-6 was conducted by calcination of PST-5 at 550 °C for 8 h under air flow.

cRED data collection

PST-5 powder in its as-made form was dispersed in ethanol and a droplet of the suspension was transferred onto a carbon-coated copper grid and dried in air. cRED data was acquired in selected-area mode with a goniometer tilt speed of 0.45° s−1 and an exposure time of 0.4 s per frame (Supplementary Table 2). The total data collection time was <3 min in order to minimize the beam damage and maximize the data quality. The total tilt range was 92.6°.