Crystalline BP (purity >99.998%) was obtained from Xianfeng Nano Co., Ltd. (Nanjing, China). DMF and D₂O (purity 99.5%) were from Beijing J&K Scientific Ltd. Ultrapure water was prepared with a Millipore-Milli Q system.

Preparation of LBP Nanoflakes in Water. LBP was prepared by exfoliating bulk BP crystals in anaerobic water. To improve exfoliation efficiency, the bulk BP was ground in an agate mortar to obtain large BP flakes, and then transferred into a 250 ml anaerobic glass bottle filled with anaerobic ultrapure water. The procedures were performed in an anaerobic glovebox to prevent LBP from contacting with oxygen. Anaerobic water was obtained by pre-heating ultrapure water to 70°C and purged with high-purity N₂ for over 120 min to exclude dissolved O₂, sealed and moved into glovebox immediately. The anaerobic bottle containing BP flakes was sealed tightly and sonicated for 21 h with an immersion probe at 360 W (1800–99, Biosafer). During ultrasonication, the bottle was placed in an ice-bath to keep temperatures below 15°C. After ultrasonication, the suspension was centrifuged at 88 g for 30 min to remove unexfoliated large BP flakes. Subsequently, supernatants were centrifuged at 1800 g for 30 min to remove precipitates. LBP nanoflakes in the supernatants (H₂O-LBP) were used for further experiments.

Preparation of LBP Nanoflakes in DMF. All procedures were the same as above except that ultrapure water was changed into DMF. After the second centrifugation, the supernatants were further centrifuged at 10,000 rpm for 30 min to collect DMF exfoliated LBP (DMF-LBP) in precipitates. The precipitates were re-dissolved in anaerobic ultrapure water for further experiments.

Degradation Experiments. Degradation experiments were performed in pH 7–10 NaHCO₃/Na₂CO₃ buffer solutions. Cationic strength was adjusted to 2 mmol/L. All containers were wrapped with aluminum foils and placed in the dark to prevent light. The solutions were injected with pure O₂ every day to maintain a saturation status of O₂. Degradation kinetics experiments were performed at 15°C ± 1°C, 25°C ± 1°C, 37°C ± 1°C, 45°C ± 1°C or 55°C ± 1°C.

Degradation kinetics curves were fitted by apparent first-order kinetics equation (Eq. 1): ln C t = ln C 0 − k t (1) where C ₀ and C _t were concentrations of LBP at initial and at time t, respectively. k was degradation rate constant; C ₀ was obtained by measuring total dissolvable phosphorus (TDP) after all LBP was degraded (TDP_∞) or UV absorbance of LBP at 255 nm before degradation (A₀); C _t was obtained from Eq. 2: C t = T D P ∞ − T D P t = A t (2) where TDP _t was TDP at time t; A _t was UV absorbance of LBP at 255 nm at time t. Accordingly, Eq. 1 was transformed into Eqs. 3, 4, and k was slope of the kinetics curve obtained by plotting ln (TDP_∞ - TDP _t ) or ln (A _t ) against t: ln T D P ∞ − T D P t = ln T D P ∞ − k t (3) ln A t = ln A 0 − k t (4)

Activation energy (E _a, kJ/mol) provided information of critical energy for starting of degradation reaction (Yao et al., 2008). The relationship between E _a and k follows Arrhenius equation (Brachi et al., 2021) (Eq. 5), and E _a/R is the slope of the curve obtained by plotting logk against 1/T (Oza et al., 2014). k = A e x p − E a / R T (5) where A was a pre-exponential factor, T was the temperature K), and R was the gas constant (8.314 J mol^-1 K^-1). Each experiment was repeated in triplicate. Data reported were average values for the triplicate experiments.

Electrochemical Experiments. Interaction sites of DMF with LBP was probed by electrochemical experiments performed with CHI660E electrochemical working station (Shanghai Chenhua Instrument, China). A three-electrode system was employed. The reference and the counter electrodes were Ag/AgCl and Pt wire electrodes, respectively. Electrolyte was N₂-saturated 0.5 M H₂SO₄ (pH = 1) solution, respectively. H₂O-LBP and DMF-LBP were used as working electrodes. Linear sweep voltammetry (LSV) was used to evaluate hydrogen evolution reaction (HER) performance of the working electrodes. Scan rate was 5 mV/s. Applied potentials were shown as reversible hydrogen electrode (RHE) potentials (E _{vs RHE}) based on the Nernst function (Eq. (6)) (Benck et al., 2014). E v s   R H E = E v s   A g − A g C l + E ° A g / A g C l + 0.059 × p H (6) where E _{vs Ag-AgCl} was the experimentally measured potential V) and E°_(Ag/Ag/Cl) was the standard potential (0.199 V) at 25°C.

Characterization and Analysis. Electrophoretic mobility (EPM) and hydrodynamic diameter were measured by time-resolved dynamic light scattering (Nano-ZS, Malvern) operated at 173° scattering angle. Atomic force microscope (AFM, Agilent 5,100) was used to characterize thickness of LBP. Functional groups on the surface of LBP were characterized by Fourier transform infrared spectroscopy (FTIR, Nicolet IS 10). LBP solutions were freeze-dried, and then mixed with KBr powder for FTIR analysis. The content of phosphorus oxides (P_xO_y) for LBP was determined by X-ray photoelectron spectroscopy (XPS, ESCALAB250, Thermo VG). LBP nanoflakes were washed with ultrapure water to remove residual salts before XPS measurements. The aliphatic C 1s peak (284.6 eV) was used as an internal standard to correct the peak shifts (Barrère et al., 2003). The relative atomic concentrations were determined according to integrated peak areas of P 2p and O 1s peaks based on Lorentzian-Gaussian functions and using the Scofield sensitivity coefficients (Barrère et al., 2003). The peaks were fitting using XPSPEAK software.

Light absorbance of LBP was recorded by UV-vis spectrophotometer (Shimadzu UV-2700). TDP was quantified by inductively coupled plasma optical emission spectrometry (Agilent 5,100). Dissolved phosphorous species were identified by ³¹P nuclear magnetic resonance spectroscopy (³¹P NMR, AVANCE600, Bruker). After degradation for 4 months, 40 ml H₂O-LBP and DMF-LBP (5.6 mg/L) were freeze-dried, and re-dissolved with 1 ml D₂O for the ³¹P NMR analysis.

AFM images indicated that thickness and lateral sizes of H₂O-LBP and DMF-LBP were close (Supplementary Figure S1). Based on statistical results from 100 pieces, average thicknesses of H₂O-LBP and DMF-LBP were 4.02 ± 3.05 and 4.49 ± 3.26 nm (Figure 1A), and average lateral sizes were 311.6 ± 12.7 and 289.2 ± 9.1 nm (Figure 1B), respectively. Average hydrodynamic diameter values of H₂O-LBP and DMF-LBP were 267.7 ± 17.3 and 260.8 ± 11.2 nm at pH 7 (Figure 1C). Accordingly, the two types of LBP had similar thicknesses, hydrodynamic sizes and plane surface areas.

FTIR spectra (Figure 1D) showed presence of chemically reactive functional groups on both LBP, including HO-P=O (1,637 cm⁻¹), P=O (1,379 cm⁻¹), P-OH (1,078 cm⁻¹), P-O-P (1,148 cm⁻¹) (Druenen et al., 2019; Guo et al., 2021). Compared with H₂O-LBP, a new peak at 798 cm⁻¹, corresponding to CH₂ rocking (Tiouitchi et al., 2019), was observed for DMF-LBP. The FTIR results indicated that DMF was adsorbed on the surface of LBP. XPS spectra (Supplementary Figures S1E, F) showed two peaks at 130.0 and 129.0 eV assigning to binding energies of 2p_1/2 and 2p_3/2, respectively, corresponding to P-P bonds of LBP. A small peak at 131.6 eV belonged to P_xO_y, which was mainly derived from oxidation of LBP during ultrasonication or from irreducible water (Zhang et al., 2019). Relative intensities of P_xO_y to total P of H₂O-LBP and DMF-LBP were 7.10% and 10.3% (Supplementary Table S1), respectively. Accordingly, oxidation degree of DMF-LBP was a bit higher than that of H₂O-LBP.

Due to slight oxidation during ultrasonication, LBP carried net negative charges (Figure 1C). Oxidation functional groups of LBP contained P=O, P-OH and P-O-P components (Supplementary Figures S1G, H). Compared with P=O and P-O-P, P-OH contributed more to negative charges (García-Sancho et al., 2018). Both oxidation degree (10.3%) and percentage of P–OH (79%) of DMF-LBP were higher than that of H₂O-LBP (P_xO_y: 7.10%; P-OH: 32.3%) (Supplementary Table S1). Theoretically, DMF-LBP was inferred to carry more net negative charges than H₂O-LBP. However, EPM of DMF-LBP (−2.48 ± 0.12 μm cm Vs⁻¹) was higher than that of H₂O-LBP (−3.30 ± 0.13 μm cm Vs⁻¹) in pH 7.0 aqueous solution (Figure 1C).

The EPM values of H₂O-LBP in 0.5%–10% DMF aqueous solutions were measured at pH 7.0. The values increased from −3.30 ± 0.1 to −2.41 ∼ −2.74 μm cm Vs⁻¹ with increases of DMF (Supplementary Figure S1I). Previous studies showed that the EPM was usually used as an index of the magnitude of electrostatic interaction of the solutes (White et al., 2007; Pedersen et al., 2018). The evolution of EPM meant that the electrostatic interaction between DMF and H₂O-LBP became weaker with increases of DMF. Generally, large EPM absolute value indicates better dispersion of the solute (Li et al., 2017). The EPM value of H₂O-LBP in 0.5% DMF aqueous solution was higher than that in 1% DMF aqueous solution (Supplementary Figure S1I), suggesting that the dispersion of H₂O-LBP should be better in 1% DMF aqueous solution than in 0.5% DMF aqueous solution. On the contrary, the hydrodynamic diameter of H₂O-LBP indicated that dispersion in 0.5% DMF aqueous solution was not as good as that in 1% DMF aqueous solution (Supplementary Figure S1I). According to previous study, high concentration of DMF could transform graphene oxide nanomembranes to nanoscrolls (Tang et al., 2018). Similarly, DMF promoted the curling of LBP nanosheets, resulting in the oxidation functional groups originally formed on the LBP surface being wrapped. With increases of DMF, more oxidation functional groups were wrapped due to curling. As EPM was mainly derived from negative charges carried by the oxidation functional groups, it is not surprised that EPM was increased with increases of DMF.

Electrostatic interaction Between LBP and DMF was further evidenced by the effect of DMF on electrocatalytic activities for HER of LBP. Electrocatalytic activities for HER of H₂O-LBP and DMF-LBP were investigated in N₂-saturated 0.5 M H₂SO₄ solution according to polarization curves (Figure 1E). Compared with DMF-LBP (−849 mV vs. RHE), H₂O-LBP exhibited a higher onset potential of −749 mV (vs. RHE), implying H₂O-LBP had higher electrocatalytic activity for HER. H₂O-LBP produced current density of 10 and 30 mA cm⁻² at potentials of −0.98 and −1.23 V, while DMF-LBP needed higher negative potentials (−1.27 V for 10 mA cm⁻²; −1.54 V for 30 mA cm⁻²). Therefore, DMF weaken the electrocatalytic activity for HER of LBP.

HER involved three possible reaction steps based on two mechanisms: Volmer-Heyrovsky and Volmer-Tafel (Kwak et al., 2016). V o l m e r   s t e p :   H + + e − → H a d (7) H e y r o v s k y   s t e p :   H + + H a d + e − → H 2 (8) T a f e l   s t e p :   2 H a d → H 2 (9)

These reactions corresponded to three possible rate determining steps: Volmer, Heyrovsky and Tafel (Kwak et al., 2016). Tafel plots was an important diagnostic tool delineating rate determining steps, providing insights into reaction mechanism of electrocatalyst (Murthy et al., 2018). η = ρ ⁡ log ⁡ j + a (10) where η is overpotential, j is current density, a is exchange current density, ρ is Tafel slope. ρ value was expected to be ca. 120 (for Volmer), 40 (for Heyrovsky) or 30 (Tafel) mV dec⁻¹ (Chen et al., 2021) for different rate-determining steps of HER.

As shown in Figure 1F, ρ value of H₂O-LBP and DMF-LBP were 186 mV dec⁻¹ and 299 mV dec⁻¹, respectively. Therefore, the Volmer HER mechanism representing weak adsorption of protons on the surface of catalysts (Kunimatsu et al., 2005) was rate-determining step for both the two LBP. The ρ value of DMF-LBP was higher than that of H₂O-LBP, meaning DMF hindered sorption of protons on LBP. Previous studies showed that negative charged P-site on LBP was the active sites to complete HER (Li and Wang, 2020). This further confirmed that DMF occupied negative charged P-sites originating form oxidation sties on LBP through electrostatic interaction.

The color fading process of LBP was the slowest at pH 7, and the fastest at pH 10 (Supplementary Figure S2). For both LBP, increases of TDP and decreases of UV absorbance with time was the slowest at pH 7 and the fastest at pH 10 (Supplementary Figure S3). This was consistent with our previous studies that the degradation of LBP was pH-dependent (Zhang et al., 2019). Compared with H₂O-LBP, the color became invisible within 5 d at pH 7–8. It seemed that DMF-LBP degraded faster than H₂O-LBP.

Kinetics fitting results (Figure 2; Supplementary Table S1) showed that the apparent first-order degradation rate constants increased from pH 7 to 10 (R ² = 0.92–0.97) for both LBP, attributed to the increases of HO⁻ (Zhang et al., 2019). The reaction rate constant (k _r) calculated based on the UV absorbance (Figure 2A) were close to that calculated based on TDP (k _f) (Figure 2B) for H₂O-LBP nanoflakes at pH 7–10. But k _r was almost 0.97 ∼ 2.74 folds higher than k _f for DMF-LBP ((Figures 2C,D; Supplementary Table S2). This meant that the degradation of DMF-LBP indicated by increases of TDP contributed to only 26.7–50.8% of the total loss of DMF-LBP represented by decreases of UV absorbance. The rate gap (k _r > k _f) has been observed in the degradation of H₂O-LBP at pH 4 and 5, and was mainly attributed to deposition/aggregation (Zhang et al., 2019). However, hydrodynamic diameters measured during degradation of both LBP showed no significant changes (Supplementary Figure S4), indicating that deposition/aggregation was not the primary reason causing the rate gap for DMF-LBP.

The degradation of LBP to form dissolvable phosphorus proceeded through two steps, including generation of oxidation functional groups on surface caused mainly by O₂, and subsequent hydrolysis of the oxidation functional groups induced by H₂O (Zhou et al., 2016). Either the oxidation or the hydrolysis procedure can cause changes to UV absorbance, while only both the oxidation and hydrolysis reactions occurred can lead to increases of TDP. For H₂O-LBP, k _f was close to k _r, indicating that almost all oxidized phosphorus underwent hydrolysis to form dissolvable phosphorus. For DMF-LBP, k _r was higher than k _f though the oxidation degree was high, implying that the hydrolysis reactions of phosphorus lagged behind oxidation of phosphorus (Supplementary Table S2).

According to XPS analysis, after 60% decreases of UV absorbance, the relative intensity of oxidized P in DMF-LBP was 78.2%, increasing by 6.56 folds compared with pristine DMF-LBP (Supplementary Figure S5; Supplementary Table S3). For H₂O-LBP, the increases were 1.23 folds. TDP of DMF-LBP remained 2.89–3.06 mg/L after 164, 150, 72 h at pH 8, 9 and 10 (Figure 2D; Supplementary Figure S3D), while that of H₂O-LBP continuously increased to 5.6 mg/L. This meant that only 52%–57% of the phosphorus oxides hydrolyzed into dissolvable phosphorus in the case of DMF-LBP, while all phosphorus oxides hydrolyzed in H₂O-LBP system. This result further indicated that the hydrolysis reactions of DMF-LBP were weakened compared with H₂O-LBP.

According to ³¹P NMR spectrograms (Supplementary Figure S6), the total abundance of dissolvable degradation products of DMF-LBP was 53% of that of H₂O-LBP after degrading for 4 months. This ratio was close to that of TDP relative to TDP_∞ for DMF-LBP. The XPS results (Supplementary Figure S1, S5; Supplementary Tables S1, 3) showed the relative abundance of residual oxygen-containing functional groups were unchanged during degradation of H₂O-LBP, while decreases of P-OH and increases of P-O-P and P=O were observed for DMF-LBP after degradation. As P-OH was mainly generated by hydrolysis of P-O-P (Wu et al., 2018), this meant that P=O and P-O-P on the surface of DMF-LBP were relatively stable. As DMF impeded the hydrolysis reactions, more P_xO_y was remained on surface of DMF-LBP in comparison with H₂O-LBP, on which P_xO_y fast hydrolyzed to soluble phosphorus oxides. Therefore, the oxidation degree of H₂O-LBP was observed to be higher than that of DMF-LBP (Supplementary Figures S1E, F; Supplementary Table S1).

The k _r values of DMF-LBP were 1.47–2.29 folds higher than that of H₂O-LBP at pH 7–10 (Supplementary Table S2). Therefore, oxidation reactions of LBP to form oxidation functional groups were inferred to be accelerated by DMF, though hydrolysis reactions were impeded by DMF. To prove this, E _a of oxidation reaction between O₂ and LBP was investigated. Kinetic fitting results (Figure 3; Supplementary Figure S7; Supplementary Table S4) showed that the apparent first-order degradation rate constant increased from 15°C to 55°C at pH 8 and 9 (R ² = 0.92–0.97). Taking pH 8 as an example, the k _r values increased by 13.94 and 7.49 folds for H₂O-LBP and DMF-LBP from 15°C to 55°C (Figures 3A,C; Supplementary Table S4); k _f values increased by 14.63 and 4.66 folds for H₂O-LBP and DMF-LBP from 15°C to 55°C (Figures 3B,D; Supplementary Table S4).

E _a values of H₂O-LBP calculated based on k _r were 55.66 and 47.78 kJ/mol at pH 8 and 9, comparable with that obtained based on k _f (pH 8: 55.55 kJ/mol; pH 9: 50.22 kJ/mol) (Figures 4A,B). For DMF-LBP, E _a values calculated based on k _r were 43.58 and 39.64 kJ/mol at pH 8 and 9, and were 6.31 and 4.06 kJ/mol higher than that obtained based on k _f (pH 8: 37.26 kJ/mol; pH 9: 35.58 kJ/mol). According to above discussion, the differences between k _r and k _f were mainly attributed to influence of DMF on hydrolysis of phosphorus oxides. The different E _a values obtained for DMF-LBP indicated that energy barriers of hydrolysis reactions increased by DMF, resulting in accumulation of phosphorus oxides on the surface of LBP. By comparing E _a values of DMF-LBP and H₂O-LBP, energy barriers for oxidation were changed due to presence of DMF. Taking pH 8 as an example, E _a values of DMF-LBP calculated based on the k _f were 18.29 kJ/mol lower than that of H₂O-LBP due to presence of DMF. This meant that DMF lowered the energy barrier of the reaction between oxygen and LBP. This was consistent with the degradation kinetics results (Figure 2; Figure 3) that oxidation degree of DMF-LBP was higher than that of H₂O-LBP after the same experimental period.

As reported previously, in DMF solution, stability of LBP was improved due to the formation of solvation shell as a barrier to prevent LBP from reacting with water or oxygen in the air (Hu C.-X. et al., 2018; van Druenen, 2020). But the degradation of DMF exfoliated LBP was observed to accelerate due to increases of relative humidity when directly exposed to air (Yasaei et al., 2015), indicating that DMF may play different roles on degradation of LBP in the presence of water compared with pure DFM solution. In this study, DMF was found to play a double-edged role on degradation of LBP in water. On one hand, DMF occupied oxidation sites on LBP through electrostatic interaction, and promoted reactivity of LBP with O₂. On the other hand, DMF hindered the hydrolysis of phosphorus oxide formed on surface of LBP. This study indicated a protection mechanism of DMF in water different from that reported in previous studies focused on degradation in pure organic solvent.