3.1. Tailoring Pt Size and Dispersion
First, the crystallinity and crystal sizes of flame-made and annealed (500 °C for 1 h) powders with 0–10 mol% Pt on Al
2O
3 were investigated. In the absence of Pt, flame-made Al
2O
3 forms the cubic γ-phase (stars,
Figure 1a), in agreement with the literature [
38]. Most importantly, adding up to 10 mol% Pt systematically emerges the peaks at 2
θ = 39.7° and 46.3°. This suggests the formation of highly crystalline and metallic Pt crystals with peaks (triangles,
Figure 1a) that overlap with those of γ-Al
2O
3, as had been observed previously for wet-impregnated Pt/Al
2O
3 after rather similar (500 °C for 2 h) annealing [
52]. Importantly, no crystalline PtO (circles) is detected, which is desired for highly reactive catalytic filters [
32]. While it is known that metallic Pt dominates acidic supports [
53] like Al
2O
3 [
24], the presence of some amorphous PtO
x (not detectable by XRD) has been revealed with an extended X-ray absorption fine structure (EXAFS) on flame-made and similarly annealed Pt/Al
2O
3 (i.e., 2 h at 500 °C) before [
54].
Interestingly, the Pt peaks feature sharp tips and broader bases, indicative of bimodal crystal size distributions [
55]. In fact, deconvolution of the peak at 2
θ = 39.7° (
Figure S1) reveals smaller and larger Pt crystals. The smaller Pt crystals (triangles,
Figure 1b) feature rather constant sizes of 5.5 ± 0.6 nm, close to the 6 nm and 7.4 nm, that showed high reactivity towards methanol [
32] and ethanol [
36]. Such small Pt crystals dominate (relative abundance 82–94%,
Figure 1c) for all Pt contents over larger ones (with sizes ranging from 14.2 to 21.4 nm, circles in
Figure 1b) and are probably stabilized by strong anchoring on penta-coordinated Al
3+ sites on the γ-Al
2O
3 (100) surfaces [
56], as had been reported for Pt/Al
2O
3 before [
57].
It is noteworthy that the addition of Pt decreased the γ-Al
2O
3 crystal size from 8.1 to 6.7 nm (
Figure 1d, stars). This might indicate some (substitutional or interstitial) incorporation of Pt into the Al
2O
3 lattice, that was investigated further by XRD peak shift analysis (
Figure S2) with an internal standard (i.e., crystalline NiO [
41]). However, no lattice distortion was observed, suggesting no Pt incorporation, which is likely due to the significantly larger ionic radii of Pt (i.e., 80 pm Pt
2+ or 63 ppm Pt
4+ [
58]) compared to Al (54 pm Al
3+ [
58]) at a coordination number of VI, as relevant for γ-Al
2O
3. Note that the BET-equivalent particle diameters (determined by N
2 adsorption) for 0–10 mol% Pt were 7.7–9.5 nm (
Figure 1d, squares), consistently larger than the γ-Al
2O
3 crystal size (stars), suggesting some polycrystallinity.
The morphology and dispersion of the Pt crystals was investigated further by electron microscopy, exemplarily for 3 mol% Pt/Al
2O
3. HRTEM reveals the presence of separate Al
2O
3 (brighter) and Pt (dark) particles/clusters (
Figure 2a) that both feature a rather spherical shape. Their faceted appearance and visible lattice fringes support high crystallinity, in line with XRD (
Figure 1a). When magnifying such a Pt particle/cluster (inset of
Figure 2a), a lattice spacing of 0.224 nm is measured that matches well with the Pt (111) plane. Most such Pt crystals seem well dispersed over the Al
2O
3 support, forming fine surface clusters that are favorable for catalytic filtering given their large reactive surface areas.
A further distinction between Pt and Al
2O
3 particles/clusters is provided by HAADF-STEM, where the Pt particles appear now brighter than Al
2O
3 due to their higher scattering potential (
Figure 2b). In fact, corresponding EDXS analysis validates the presence of mostly Pt (
Figure 2c, green square in
Figure 2b) for bright clusters while Al and O (
Figure 2d, blue square in
Figure 2b) dominate the darker particles. Note that the C and Cu signals originate from the sample grid (i.e., perforated carbon foil on Cu grid, see Materials and Methods).
The size distribution (
Figure 2e) for 1000 Pt particles/clusters was determined from such HAADF-STEM images (
Figure S3). A lognormal fit (red line) yields a geometric average diameter (d
g) and standard deviation (σ
g) of 5 nm and 1.53, respectively, that agrees well with the average crystal size of the small Pt clusters (
Figure 1b: 5 nm). Note that no bimodality is visible in the number frequency size distribution here, likely since the relative abundance of larger particles is rather small (
Figure 1c). Remarkably, quite similar d
g (i.e., 5–5.9 nm) are obtained for all Pt contents (
Figure S4). This should be associated with the aforementioned strong anchoring of the small Pt clusters on the Al
2O
3 [
56] that prevents their sintering during annealing, while the larger clusters grow with increasing Pt content (
Figure 1b). As a result, Pt content affects primarily the surface loading, while the size of small Pt clusters and their dispersion remain rather invariant.
3.2. Catalytic Reactivity
The catalytic performance of these nanoparticles was tested by analyzing the exhaust of a 30 mg Pt/Al
2O
3 packed bed with bench-top PTR-ToF-MS (
Figure 3). Tests were performed with 1 ppm of gaseous acetone (circles), isoprene (diamonds), methanol (triangles) and ethanol (squares) at 90% RH to simulate breath-realistic conditions. When increasing the temperature sequentially from 25–400 °C, the pure Al
2O
3 catalyst (
Figure 3a) converts first isoprene (100% conversion at 140 °C) followed by methanol (260 °C) and ethanol (290 °C). Remarkably, acetone starts to convert only at 270 °C and complete conversion is observed even after 390 °C, resulting in distinct acetone selectivity, as had been shown previously for ethanol and acetone [
19].
Most importantly, when increasing the Pt content, the conversion curves are systematically shifted towards lower temperatures (
Figure 3b–e). Specifically, all interferants are converted completely at 90 °C with only 1 mol% Pt and this is further reduced to 40 °C in the case of 3, 5 and 10 mol% Pt. Note that the filter should not be operated below 40 °C, which is standard [
35,
59] in breath analysis to avoid water condensation from rather humid exhalations (i.e., >90% RH [
60] at body temperature). The high reactivity of 3–10 mol% Pt/Al
2O
3 at room temperature should be attributed to the well dispersed Pt clusters of 5.0–5.9 nm size (
Figure 2 and
Figure S4). In fact, similar Pt cluster sizes were reported to be highly reactive for methanol [
32], ethanol [
36] and hydrocarbons [
37], as had been specified in the Introduction.
Most importantly, the high acetone selectivity is maintained for all Pt contents, as the acetone is converted consistently at higher temperatures (
Figure 3f, circles) than the confounders. For instance, for three identically prepared 3 mol% Pt/Al
2O
3 packed beds at 40 °C, only 18.7% ± 5.8% (
Figure 3c, circles) of the acetone are lost while all confounders are removed completely. To further investigate this acetone selectivity, we reduced the 1 and 10 mol% Pt/Al
2O
3 in H
2 prior to catalytic characterization (
Figure S5). While this resulted in even lower conversion temperatures for all confounders, the acetone selectivity was deteriorated (i.e., 50.6 and 59.3% acetone conversion at complete interferant removal for 1 and 10 mol% Pt, respectively). This suggests the presence of less reactive [
61] but apparently more acetone-selective PtO
x on the metallic Pt clusters [
54], that might be amorphous since it is not detectable by XRD (
Figure 1). In fact, in situ XRD (
Figure S6) during this treatment also revealed neither changes of the crystalline phases nor their sizes. However, the detailed reaction mechanism remains to be clarified.
To challenge the catalytic filter further, the 3 mol% Pt/Al
2O
3 packed bed at 40 °C was tested for the removal of 5–100 ppm ethanol (
Figure 4) at 50% RH. Such high ethanol concentrations can be present in hospitals from sanitizers [
22] and are removed by the catalytic filter completely (red vs. blue line), as confirmed by PTR-ToF-MS. This is, at least, competitive to filters based on Au/Fe
2O
3 (at 200 °C) [
62] and ZnO (at 260 °C) [
23], that had to be heated though. Furthermore, the catalyst was fairly robust to changing RH between 30 and 90% RH (acetone loss 46–14% at complete interferant conversion,
Figure S7a), as it is usually present in room air and exhaled breath, and performs well also for other flows (i.e., 50–200 mL/min,
Figure S7b) through the packed bed.
3.3. Selective Acetone Sensing with Room Temperature Filter
To demonstrate immediate practical impact, 30 mg of such 3 mol% Pt/Al
2O
3 at 40 °C were placed as packed bed filter ahead of a flame-made, chemoresistive Si/WO
3 [
48] sensor. When testing the sensor alone to 1 ppm acetone and eight breath-relevant interferants at 90% RH (
Figure 5a), it responded to acetone (18) but showed an even higher response to isoprene (43.2) and was interfered by ethanol (2) and H
2 (0.5) that can be present at orders of magnitude higher concentrations than acetone. The resulting selectivity at the same analyte concentrations range from 0.4–600 and are in fair agreement with earlier reports for ethanol (6.7 but at 400 °C [
63]) and isoprene (0.5 [
19]). However, these are insufficient and can lead to significant measurement errors, for instance, when monitoring breath acetone in situ during cardio-respiratory fitness-adapted [
64] cycling [
26].
This is eliminated effectively by the filter. In fact, the 3 mol% Pt/Al
2O
3 packed bed at 40 °C reduces these interferences (
Figure 5b). Now, only acetone is detected with a response of 15.8, while the interferants are hardly recognized anymore (responses <0.1), in line with the catalytic characterization (
Figure 3c). This results in high selectivity for all analytes (≥225,
Figure 5b in brackets), and is highest for CO, ethanol and 2-propanol (all >1′000). Note that the acetone response reduction of 12.2% (
Figure 5b) is in fair agreement with
Figure 3c (18.7 ± 5.8%). The obtained selectivities are comparable to the ones achieved with a 0.2 mol% Pt/Al
2O
3 filter (at 135 °C); however, operated here at room temperature. Moreover, it outperforms state-of-the-art acetone sensors (e.g., Al-ZnO [
18], Si/WO
3 [
19], Co-doped ZnO nanofibers [
15], TiO
2/WO
3 nanocrystals [
16], SnO
2 with multi-walled carbon nanotubes [
20] and Au vertical hematite nanotube arrays [
21]).
End-tidal breath acetone levels are usually between 148–2744 ppb, as observed during weekly breath tests of 30 volunteers during 6 months [
59]. Therefore, the detector was exposed subsequently to 100 and 50 ppb of acetone (
Figure 6a). These concentrations are clearly distinguished with high signal to noise ratios (i.e., SNR > 50). Note that the extrapolated LOD (at SNR = 3) is even 2 ppb. Importantly, the detector features also a good repeatability (dashed lines,
Figure 6a) with a response change <5% and excellent reproducibility of ±5.8% for the filter (error bars in
Figure 3c at 40 °C) and 8% [
19] for the Si/WO
3 sensor alone, when testing three identically prepared samples.
Since exhaled human breath is a mixture of analytes, we tested also binary combinations of these acetone concentrations with 1 ppm of ethanol (circles,
Figure 6b), methanol (diamonds) and formaldehyde (triangles). Most importantly, the detector response to acetone hardly changes (e.g., 15.2 ± 0.5 at 1 ppm), highlighting its excellent selectivity. Finally, the detector was tested for its RH robustness when sensing 100 ppb acetone (
Figure 6c). Remarkably, the response changed only little from 1.5 to 1.7 between 30–90% RH, demonstrating outstanding humidity robustness. Previous studies [
63] showed reduced acetone response at increasing RH for the Si/WO
3 sensor alone that apparently compensates for the filter’s higher acetone loss (
Figure S7a).