44Ti diffusion labelling of commercially available, engineered TiO2 and SiO2 nanoparticles

In realistic exposure scenarios, the detection and quantification of engineered nanoparticles in complex environmental or biological matrixes is a challenge since nanoparticle concentrations are frequently low and have to be discerned from a background that may contain the same elements in various chemical forms in much higher concentrations. The use of radiolabelled nanoparticles may overcome these difficulties offering high detection sensitivity without the necessity of complex sample preparation procedures. However, the labelling procedure must not alter the physicochemical and biological properties of the nanoparticles. In the present work, the radiolabelling of three different types of TiO2 nanoparticles with primary particle sizes between 5 nm and 26 nm with commercially available 44Ti has been investigated applying a simple diffusion heat treatment at 180 °C for 2.5 h on nanoparticles impregnated with a solution containing the 44Ti radiolabel. The same treatment has been investigated to radiolabel amorphous SiO2 nanoparticles with 44Ti. The radiolabels are stably integrated in the nanoparticle matrix, and the release is less than 0.1% in aqueous suspension at neutral pH for at least 4 weeks. The method appears to be fast and reliable. By transmission electron microscopy, dynamic light scattering and ζ-potential measurements, only minor alterations of the nanoparticle size could be detected in the range of 1 to 2 nm.


Introduction
Investigations of the fate of nanoparticles in biological systems and environmental matrices frequently encounter the challenge to detect the applied nanoparticles on a chemically identical natural background ) and in very low concentrations in experimental settings mimicking realistically low exposure scenarios. It is a drawback of many investigations that they compensate for insufficient detection sensitivity by unrealistically high dosage or exposure (e.g. Krug 2014). Radiolabelled nanoparticles have been applied in in vitro (e.g. Marmorato et al. 2011;Simonelli et al. 2011;Ponti et al. 2009) and in in vivo toxicological (e.g. Schleh et al. 2013;Kreyling et al. 2017a, b, c;Xie et al. 2010;Zhang et al. 2009) and environmental studies (e.g. Kleiven et al. 2018;Chekli et al. 2016;Vitorge et al. 2014;Coutris et al. 2012;Oughton et al. 2008) where they demonstrated the advantage of very high detection sensitivity and easy quantification, usually without special specimen preparation procedures (Bello and Warheit 2017;Llop et al. 2013;Weiss and Diabate 2011). However, since the properties and behaviour of nanoparticles depend on a large variety of physicochemical parameters, utmost care is required not to change any of these during the labelling procedure. This holds especially for investigations in the framework of the risk assessment of industrially manufactured nanoparticles in which frequently the use of nanoparticles is desired as they emerge from a large-scale industrial production process. Since the radiolabels cannot be introduced in the industrial manufacturing process, radiolabelling must be done post-manufacturing. Therefore, the effects of the radiolabelling technique on the properties of the industrial manufactured nanoparticles have to be addressed appropriately. In such tracer experiments, the distribution and quantification of the radiolabelled nanoparticles is determined on the basis of the radiation emitted by the radiolabel. This implies that the integrity of the radiolabel-nanoparticle construct has to be ensured at least over the envisaged duration of the experiments.
The situation is different for medical nanoparticles where surface functionalization is part of the strategy to target disease. In this case radiolabels, either used as tracers for diagnosis or as therapeutic agents delivering high radiation doses selectively to diseased tissues, may be loaded on the surface or in the volume of nanoparticles by chemical means during synthesis of the pharmaceutical (Kunz-Schughart et al. 2017;Dieci et al. 2017;Scheinberg et al. 2017). Nevertheless, also for medical purposes, applications were reported in which the radioactive load was created after synthesis using neutron activation of a metallic nanoparticle core (Hamoudeh et al. , 2008Buono et al. 2007;Häfeli et al. 2001).
Various purely physical radiolabelling techniques have been described that avoid chemical processing or surface functionalization, which entails the risk of altering the behaviour of the nanoparticles in the envisaged experimental environment. Probably most frequently applied are neutron activation techniques Oughton et al. 2008;Häfeli et al. 2001). Also, techniques where activation is achieved by exposure of nanoparticles to light ion beams have been developed Holzwarth et al. 2014). However, a certain effort is required to carefully select and test the irradiation conditions, which ensure that the properties of the sensitive nanoparticles are not altered during ion bombardment (Holzwarth et al. 2012). Both techniques suffer from the drawback that access to a nuclear research reactor or a nuclear accelerator is required, respectively. In this respect diffusion labelling of nanoparticles using commercially available radionuclides may offer a faster and simpler alternative.
Analogously to the reduction of the melting temperature of nanoparticles with shrinking particle size, especially pronounced for sizes below 10 nm to 15 nm, the diffusion coefficient for self-diffusion and for foreign atoms in nanoparticles increases by even more than an order of magnitude (Jiang et al. 2004), and diffusion is facilitated by the abundant presence of defects acting as diffusion vehicles (Shibata et al. 2002). Therefore, the few diffusion steps required by an atom to penetrate a nanoparticle may occur at temperatures and within time periods which are much lower and much shorter, respectively, than for the corresponding bulk materials (Vollath 2008).
The experimental proof that the impregnation of TiO 2 nanoparticles with a solution containing 44 Ti followed by a temper treatment (2 h at 180°C) leads to diffusion has been presented by Butz (2012) and Butz et al. (2011) who studied disorder in TiO 2 nanoparticles by time differential perturbed angular correlation using 44 Ti as a radioactive probe. 44 Ti decays by electron capture, and the daughter nucleus 44 Sc de-excites passing through an intermediate state of 44 Sc (with nuclear spin I = 1) emitting a γ-γ-cascade with energies of 78 keV and 68 keV (see Butz et al. (2011)). Since the 44 Sc in its intermediate state exhibits a nuclear quadrupole moment, which interacts with the local electrical field gradient on its lattice site, the nucleus undergoes a precessional motion which disturbs the angular correlation between the emission directions of the first and the second γ-photon. In several TiO 2 nanomaterials, such as ST-01 and P25 nanoparticles, Butz could identify a precession frequency, which was close to the precession frequency of the 44 Sc when the 44 Ti decays on a Ti lattice site in (monocrystalline) bulk TiO 2 . This volume signal was broader than the signal known from large bulk TiO 2 which indicates contributions of decays that occurred on lattice sites with variable depth below the surface and highly variable disorder caused by distortions and defects of the crystal lattice leading to deviations from the precession frequency of the 44 Sc on a perfect lattice site. In all materials a second signal was detected, which can be attributed to decay sites where the electrical field gradient is modified due to the vicinity of a surface where the broken symmetry affects the electrical field gradient on the decay site of the 44 Ti atom (Butz 2012). Except for the smallest nanoparticles investigated, the surface signal had a much lower intensity than the volume signal, indicating that more probes decayed in the volume of the nanoparticles than close to their surface, which clearly proofs that the 44 Ti atoms penetrated into the nanoparticles by diffusion. The less intense surface signal was much sharper than the volume signal. This was attributed to highly mobile H atoms terminating the surface. Their fast motion creates a much more homogeneous average field gradient on the surface decay sites, which is known in solid-state spectroscopy as motional narrowing (Butz 2012;Butz et al. 2011). Additionally, since the electrical field gradient depends of the crystal structure of the material, this method allows to distinguish contributions from the different polymorphs of the TiO 2 .
Hildebrand and Franke were the first to apply this approach to produce stable 110m Ag-labelled silver nanoparticles by impregnating powdered Ag nanoparticles with a [ 110m Ag]AgNO 3 solution (Hildebrand and Franke 2012). Then they extended the method to label TiO 2 Aeroxide® P25 with 44 Ti and 45 Ti (Hildebrand et al. 2015). The labelling stability was assessed by determining the fraction of free 44 Ti radioactivity over a period of up to 28 days after diffusion labelling. During the whole observation period, the free 44 Ti activity was found to stay well below 0.5% for all pH conditions (ranging between pH = 2 and pH = 10), which indicates sufficiently stable radiolabelling. While the 45 Ti could be produced with a locally available minicyclotron, its physical half-life of t ½ = 3.08 h is too short-lived for most applications. However, the 44 Ti with a physical half-life of t ½ ≈ 60 years (59.2 ± 0.6 years (Ahmad et al. 1998)) could not be produced with the available cyclotron and was rather expensive in commerce, which was considered a problem for a broader application.
The motivation for the present work to look again into 44 Ti-duffusion labelling was (i) the increasing demand for intrinsic (radio)-labelling (Jin et al. 2017;Goel et al. 2014) and (ii) the question whether the method would be equally successful with stable long-term integration of 44 Ti atoms in other industrially manufactured TiO 2 materials with different crystalline structures (cf. Vandebriel et al. 2018). (iii) Since neither neutron activation nor charged particle activation of Si yields adequate radiolabels for biokinetics and biodistribution studies, it was investigated whether diffusion labelling with 44 Ti could be applied to commercially available SiO 2 nanoparticles. Making use of the 18 O(p,n) 18 F nuclear reaction, the PET (positron emission tomography) tracer 18 F could be produced (Pérez-Campaña et al. 2012. However, 18 F (T ½ = 110 min) is too short-lived for biokinetics studies exceeding a few hours, and 18 O has a natural abundance of only 0.2% of the oxygen which limits the amount that can be produced from isotopically non-enriched SiO 2 nanomaterial as it is industrially produced.
The availability of 44 Ti is going to improve very likely as it is the mother radionuclide of 44 Sc which has been spotted as a candidate label for PET tracers and in combination with 43 Sc for theranostic applications in cancer treatment. Making available 44 Ti/ 44 Sc generators to hospitals would allow them to apply such theranostic systems without the need for an own cyclotron (Radchenko et al. 2017;Pruszyński et al. 2010). From this development one may expect an increased availability of 44 Ti at lower cost in near future.
Future practical applications of 44 Ti-labelled TiO 2 nanoparticles might include studies on the fate of TiO 2 nanoparticles in simulated environments, where (due to the inertness of TiO 2 ) accumulation effects in water, groundwater, soil and waste water treatment residues may be expected. For such experiments the much longer half-life of 44 Ti would be an asset and outweigh the rather strict precautions for laboratory space, radioactive waste management and radiation protection imposed by handling of long-lived radionuclides. In any case stable radiolabelling is a prerequisite for such studies.
Since it is not feasible to predict a priori all possible experimental conditions in which 44 Ti-labelled nanoparticles might be applied in the future, the present work focuses on the stability of the radiolabels in pH neutral aqueous stock solutions of the TiO 2 nanomaterials and the amorphous SiO 2 nanomaterial compiled in Table 1.

Materials 44
Ti was purchased as non-carrier added 44 Ti dissolved in 4 M HCl from Los Alamos National Laboratory (USA) in 2004 with a radionuclidic purity > 99.8%. From this batch 300 kBq in a volume of 640 μL were still available for the experiments.
Four types of nanoparticles were selected for the labelling experiments. The main criteria were a similar size of the primary particles, while the crystal structure, i.e. the polymorphs of TiO 2 , should vary. Previous experience with radiolabelling of these materials with different methods and continuous requests for radiolabelling were an asset. These selection criteria and the most relevant properties of the nanomaterials are compiled in Table 1.

γ-Ray spectrometry
The 44 Ti was quantified using its γ-ray emissions at 67.9 keV and 78.3 keV (Chu et al. 1999). The γ-ray spectra were evaluated using the software package GE-NIE 2000 (Canberra Inc., USA). Due to the long physical half-life of 44 Ti, no decay corrections was required in the present experiments.
The absolute activities of stock solutions and all specimens were determined by γ-ray spectrometry using high purity germanium (HPGe) detectors from Canberra (USA) and EG&G Ortec (USA). In order to reduce the background of environmental radioactivity, the spectrometers were shielded with lead. Additionally, the recorded spectra were corrected using background spectra that were recorded for each HPGe detector. The spectrometers were calibrated in energy and efficiency with certified standard calibration sources which exhibited an activity uncertainty of 1-1.5%.
Depending on the shape (15-mL Falcon tubes with and without centrifugal filter; 50-mL Falcon tubes with and without centrifugal filter; glass vials with flat bottom) and activity of the specimens (low activity filtrates, high activity on centrifugal filters), different measurement geometries were used. Since no calibration sources The main selection criteria were approximately the same primary particle size while varying the crystal structure and demand for radiolabelling of the materials. (*) corresponds to NM105 (OECD (2015)) were available for these specific geometries, the geometry effect was checked with a drop of 44 Ti solution (4.8 kBq) which was first measured in a glass vial and then with 1 mL of water added, transferred in a 15-mL Falcon vial, and from there in a 50-mL Falcon vial topped up to a volume of 4 mL with water. All results coincided within 10%, which was smaller than the activity uncertainty of 15% given by GENIE 2000. Small 44 Ti activities in most filtrates of 1 mL from leaching studies had to be measured with a data acquisition time of 24 h in order to quantify 44 Ti activities as low as 0.1 Bq.

Diffusion labelling of TiO 2 nanoparticles
Nanoparticles were weighed and impregnated with the 44 Ti dissolved in 4 M HCl in a 20-mL glass vial. For 10 mg of nanoparticles, typically 150 μL of the 44 Ti solution were used. The vial with the slurry was then transferred into a preheated vacuum furnace, and the impregnated nanoparticles were subjected to a temperature of 180°C for 2.5 h. Initially, the pressure in the vacuum furnace was adjusted to 850 mbar in order to facilitate drying and as a precaution against hypothetical escape of radioactivity during a thermal treatment. Then the valve connecting the furnace with the vacuum pump was closed, and the degradation of the vacuum due to the evaporation of the water was accepted. After 2.5 h the heating was switched off, and the vial was allowed to cool down to room temperature.
After this treatment the dried nanoparticle powder was recovered from the diffusion treatment vial. For this purpose 4 mL of 10 mM HCl was added to the vial, and the vial was subjected to a direct ultrasound treatment for 3 min to detach nanoparticles from the glass ware. During the sonification twice 3 mL of Milli-Q water (filtered using a 0.22-μm membrane filter) was added.
The suspension (10 mL) was then transferred into a 50-mL Falcon vial equipped with a 10 kDa centrifugal filter for washing and determination of the labelling yield. The washing procedure was adjusted and improved during the experiments. Washing was performed at least once using 10-mM HCl in order to facilitate removing weakly bound radiolabels from the nanoparticle surface and then using Milli-Q water until the activity measured in the liquid phase was below 1 Bq of 44 Ti. Such a low residual-free activity was required in order to keep the background sufficiently low for the subsequent leaching studies. Centrifugation was carried out at a relative centrifugal force of 2700 g for 45 min at 4°C. After the first centrifugation of the 10-mL suspension recovered from the processing vial, the lost liquid was replaced by 4 mL (10 mM HCl or Milli-Q water) in each further washing step.
After the last washing step, the nanoparticles were recovered from the filter with a pipette repeatedly flushing the filter with 200 μL high purity Milli-Q water and transferring the nanoparticles for storage in a 50-mL Falcon vial. Based on the experience gained in the first experiment (see Table 1), this procedure was repeated up to 25 times until the liquid transferred into the Falcon vial appeared clear and no longer white and milky. After recovery, the residual activity on the filter A filter was determined, and the corresponding masses of the nanoparticles lost on the filter and those recovered were calculated. The recovered nanoparticle suspensions were diluted using high purity Milli-Q water to a total volume of 10 mL or 20 mL in order to establish a concentration of around 0.5 mg/mL. No dispersants or stabilizers were added.
The suspension was then homogenized by a direct ultrasound treatment using a digital sonifier (Branson Ultrasonics Corporation, Model 450) equipped with a tapered microtip directly immersed into the suspension. The power amplitude of the device was set to 50%, and ultrasound was applied for 45 min. The power input into the suspension was calibrated (Taurozzi et al. 2011) by measuring the temperature increase during 300 s in water under identical geometrical conditions (vial type, filling and immersion of microtip), which led to a value of 0.2 W/mL which translates into a total applied energy of 540 J/mL. In order to avoid an excessive temperature increase of the nanoparticle suspension, during sonication the Falcon vial was itself immersed in a 450-mL isopropanol bath with an initial temperature of − 18°C in a 500-mL beaker. The suspension was still below room temperature after sonication. For each sonication treatment, a cold isopropanol bath directly from the freezer was used.
Determination of nanoparticle losses during processing, labelling yield and fraction of unbound, free 44 Ti The mass of nanoparticles loaded into the glass vial and subjected to the diffusion labelling was initially determined by weighing as m 0 . The nanoparticles were impregnated with typically about 150 μL of 44 TiCl 4 solution per 10 mg of nanoparticles. The applied 44 Ti activity was determined as A 0 . The emptied glass vial exhibited a 44 Ti activity of A vial , which is related with a mass m vial of nanoparticles that were lost for further processing. This mass can be calculated as and the nanoparticle mass which is then subjected to a washing treatment can be calculated as After the last washing by centrifugal filtration was completed, the nanoparticles were recovered using repeatedly small volumes of Milli-Q water (200 μL) to wash the filter and to transfer the liquid in a 50-mL Falcon vial for storage. This recovery of washed nanoparticles is never complete. After the recovery of the nanoparticles, the filter exhibits a residual 44 Ti activity of A filter from which the mass of nanoparticles that were lost on the centrifugal filter can be calculated. The mass of the nanoparticles m wash that were washed was initially associated with the 44 Ti activity A 0 − A vial . However, after washing, the same mass is associated with an activity which is reduced by the sum of the 44 Ti that was washed out in each of the washing steps. Therefore, assuming that nanoparticles cannot cross the centrifugal filter, holds, which allows then to calculate the mass of the nanoparticles recovered in the stock suspension as This value was then used to calculate the mass concentration of the nanoparticles in the stock solution and the activity concentration of the nanoparticles.
The labelling yield is determined as where we have to consider the activity of 44 Ti that had been washed out in all executed washing steps (i = 1 …. up to 20). The activity A 0 -A vial takes into account only the nanoparticles that were subjected to the washing since the activity of those retained in the diffusion processing vial contains both the stably labelled 44 Ti atoms as well as those loosely bound on the surface. For the leaching experiments, a certain volume of suspension has been subjected to filtration, and the fraction of free 44 Ti, f free , has been determined from the activity measured on the filter after complete filtration, A NP , which was ascribed to the nanoparticles, and the activity of the liquid filtrate, A liquid , according to where A NP >> A liquid is the normal case yielding values of f free below 10 −3 .

Nanoparticle characterization
Hydrodynamic size distributions and ζ-potentials of the 44 Ti diffusion-labelled nanoparticles were determined in aqueous suspensions using Milli-Q water (filtered using a 0.22-μm membrane filter) after homogenization with ultrasound. No dispersants or stabilizers were added. The concentration was adjusted to 0.1 mg/mL by adding the appropriate quantity of Milli-Q water. The analysis was performed by dynamic light scattering (DLS) using a Zetasizer Nano ZS (Malvern Instruments Ltd., Malvern, UK). The specimens were equilibrated for 180 s at 25°C, and the DLS measurements were performed triplicate, and the results averaged using the software provided with the instrument.
The same instrument was used to determine the ζpotential through the electrophoretic mobility of the nanoparticles making use of the Henry equation and the Smoluchowski approximation. The specimens were equilibrated for 600 s at 25°C, and 25 runs were performed using the automatic attenuation and voltage selection of the instrument. The results were averaged using the software provided with the instrument. In order to get reproducible and comparable measurements for the TiO 2 nanomaterials, the pH was adjusted to pH ≈ 3 by addition of 10 mM HCl. The measurements on the NM200 SiO 2 nanoparticles could reproducibly be performed at the neutral pH of the stock suspension. In this case 10 mM NaCl solution (filtered using a 0.22-μm membrane filter) was added in order to increase the electrical conductivity of the suspension to about 1 mS·cm −1 .
For the preparation of TEM specimens, carbon-coated TEM grids (carbon type-B, 200 mesh copper grids, supplied by TED Pella Inc., Redding, CA, USA) were stained with Alcian blue (2 mg/mL) immediately before the nanoparticles were applied in order to increase the hydrophilicity of the grids and to reduce the tendency of the nanoparticles to cluster on the grid when drying the suspension. For this purpose the grid was placed face down for 10 min on a drop of Alcian blue solution. Excessive staining was removed by passing the grid sequentially over 5 drops of Milli-Q water. Excessive water was removed by blotting the edge of the grid on a strip of filter paper. Nanoparticles were applied by placing the conditioned coated side of the grid on a drop of nanoparticle suspension. TEM investigations were performed on a JEOL 2100 microscope operated at an acceleration voltage of 200 kV. Digital images were analysed with the software Gatan Digital Micrograph supplied by JEOL with the TEM instrumentation. However, due to the sometimes very small and highly agglomerated nanoparticles, the software frequently could not unambiguously delineate the nanoparticles. Therefore, the minimum and maximum Feret diameter was determined manually on at least 300 nanoparticles.

Leaching experiments
Leaching experiments were performed with the aim to determine the increase of free 44 Ti in the stock suspension. For this purpose at a given time point, the stock suspension was vortexed for 1 min, and a volume of 1 mL was transferred in a 30 kDa centrifugal filter in a 15-mL Falcon vial. Centrifugation was done with a relative centrifugal force of 3461 g for 20 min. Then the 44 Ti radioactivity of the vial has been determined without the filter and the liquid centred on the detector cap. The usually low activities (close to the detection limit) frequently required measuring times of 24 h. Then the measurement was repeated with the filter in the vial and the filter centred on the detector cap. The ratio of free 44 Ti was determined as the ratio of the activities determined without and with the centrifugal filter according to Eq. (6).

44
Ti diffusion labelling yields The parameters and the results of the 44 Ti diffusion labelling experiments are compiled in Table 2. The labelling of the ST-01 material has been repeated three times. In the first experiment, the washing was done in only two steps. Since the nanoparticles were recovered from the processing glass vial with 4 mL of 10 mM HCl, the first washing step can always be considered washing with HCl. In the first experiment, washing was finished after the second step with Milli-Q water leaving a fraction of free 44 Ti of 4.4·10 −4 . In all further experiments, the washing was carried on until the 44 Ti activity determined in the filtrate was well below 1 Bq, in order to start the leaching experiments on a lower background of free 44 Ti in the stock suspension. For the materials P25, the first washing step yielded nearly 0.5 kBq of free 44 Ti. Therefore, it was decided to execute the next washing steps with 10 mM HCl until a level of free activity could be reached similar to the value determined after the second washing of the ST-01 material. It turned out that the washing of P25 and of NM01004a required more intense washing than the ST-01 material. For the ST-01 material, at maximum ≈ 0.5% of the 44 Ti could be washed out (sum over all washing steps) which is equivalent with a labelling yield of ≥ 99.5%. For the P25 and the NM01004a material more than 1% and more than 2.5% of the 44 Ti was washed out, respectively, which corresponds to labelling yields of about 98.8% and 97.5% for P25 and NM01004a, respectively.
In the amorphous SiO 2 matrix, only 38.9% of the applied 44 Ti activity was stably integrated after applying the same thermal diffusion treatment as for the TiO 2 material. Due to the time-consuming centrifugal filtration, the washing could not be accomplished in one workday. After the last washing step on the first day still using 10 mM HCl, the missing liquid on the centrifugal filter was replaced by Milli-Q water in order to avoid drying of the nanoparticles on the filter overnight. The first washing steps on day two using Mill-Q water yielded again higher released 44 Ti activities as at the end of day one (labelled (**) in Table 2). A total of 20 washing steps were required to finally reach an activity of less than 1 Bq of 44 Ti in the filtrate. This corresponded to a fraction of free 44 Ti of about 10 −4 .
In Table 2 the mass of the nanoparticles m 0 is given, which is subjected to the diffusion treatment and the Compiled are the used 44 Ti activity A 0 and the nanoparticle mass m 0 subjected to the diffusion labelling treatment, the measured 44 Ti activities in emptied glass vials after diffusion treatment A vial , the 44 Ti activities in the filtrate after each washing step A step,i and their sum over all washing steps; the 44 Ti activities A filter retained on the centrifugation filter used for washing after recovery of nanoparticles for the stock suspension. The mass of the nanoparticles lost in the processing vial m vial , the mass of nanoparticles recovered from there for the washing treatment m wash , the mass nanoparticles m filter lost on the centrifugation filter after washing and the nanoparticle mass recovered after washing m rec as well as the activity of the recovered nanoparticles A rec were calculated using Eqs. (1), (2), (3), (4a, 4b), (5), and (6). (*) denotes washing steps that were performed using 10 mM HCl. The calculated labelling yield and the final fraction of free 44 Ti in the suspension after washing are presented. The calculated activity concentration of the nanoparticles, given in Bq/mg, and the mass and activity concentration of the suspension, given in mg/mL and Bq/mL, respectively, were calculated applied activity of the 44 Ti. The 44 Ti activity measurements are compiled for the empty processing (glass) vial, for all washing liquids collected after centrifugal filtration and the activity left behind on the filter which could not be recovered for the stock solution after washing. From these activity measurements, the mass of the nanoparticles has been calculated that were retained in the processing vial m vial , the mass that was subjected to the washing treatment m wash = m 0m vial and the mass that was retained on the filter after washing m filter , making use of Eqs. (1), (2), (3) and (4a, 4b). Therefore, it is possible to calculate the mass recovered in the stock suspension m stock as well as the activity concentration in the 44 Ti-labelled nanoparticles in Bg/mg and of the suspension in Bq/mL. Thus, the table proofs the ease of process control when radiolabelled nanoparticles are used. The loss of activity in the glass vial used for diffusion treatment can be explained by the retention of nanoparticles adhering to the glass and/or those that could not be recovered by pipetting. The ratio of the activity measured on the emptied glass vial A vial and the applied activity A 0 allows calculating the mass of nanoparticles which could not be recovered from the diffusion treatment vial. The mass which is related to the activity A 0 -A vial is then subjected to the washing treatments by centrifugal filtration using a 10 kDa Amicon® Ultra-15 filter in a 50-mL Falcon vial for 40 min with a relative centrifugal acceleration of 2701 g at 4°C.
The same procedure was applied to label the amorphous silica nanoparticles NM200 (Rasmussen et al. 2013) taken from the JRC Nanoparticle Repository. Centrifugal filtration as described above revealed significant losses of 44 Ti in the washing liquids summing up to more than 60%. After 20 washing steps, a fraction of free 44 Ti of a little less than 0.01% of the applied 44 Ti was revealed in the filtrate. When repeating the centrifugation on 1-mL aliquots of the aqueous stock suspension 4 weeks after radiolabelling, the fraction of free 44 Ti in the filtrate was still as low as 0.03% which indicates stable integration of the 44 Ti radiolabel in the [ 44 Ti]SiO 2 nanoparticles.

Nanoparticle properties
The DLS and ζ-potential measurements show generally a good agreement between the cold base material and the properties after 44 Ti diffusion labelling after applying the same homogenization treatment. The hydrodynamic diameter of the nanoparticles was determined in a 1-mL specimen that was adjusted to a concentration of 0.1 mg/mL immediately after ultrasound homogenization as described in the "Methods" section. Numerical results are presented in Table 3, and intensity vs diameter graphs for 44 Ti-labelled and unlabelled nanoparticles are shown in Fig. 1. From Table 3 it can be recognized that the numerical values of the 44 Ti-labelled and unlabelled nanoparticles agree satisfactorily.
The ST-01 material is very agglomerated; if real nanosized fractions are needed for experiments, either dispersants should be used during ultrasound homogenization or a size selection has to be performed (Kreyling et al. 2017a, b, c). For SiO 2 NM200 it is difficult to reproduce DLS results for both labelled und unlabelled material as can be seen from the two pairs of curves presented in Fig. 1. Figure 2 a to d show micrographs from the transmission electron microscopy with the same magnification before and after the diffusion labelling. We applied the GATAN Micrograph software to a series of micrographs for each material to analyse at least 300 nanoparticles. However, especially in the case of the ST-01 material, the highly agglomerated small nanoparticles were frequently not correctly delineated by the software. Therefore, the size measurements were performed manually on all materials in order to apply always the same procedure. The minimum and maximum Feret diameter was determined. The histograms in Fig. 2 visualize the minimum Feret diameter. The quantitative evaluation of the histograms was done by a Gauss fit (in ORIGIN 2019b). In Table 4 the centre x c of the Gauss fits are compiled ± σ which is determined as the half of the full width at half maximum (FWHM) for the fitted Gauss curve. Additionally, the mean value of the minimum and maximum Feret diameter is reported ± its standard deviation and the aspect ratio of the nanoparticles defined as the maximum Feret diameter divided by the minimum Feret diameter. For each of the nanomaterials, the mean dimension of the nanoparticles before and after diffusion labelling is the same within the error margins; however, slight changes in the range of 1 to 2 nm can be recognized. While the P25 and the NM01004a TiO 2 nanoparticles are slightly bigger after the diffusion labelling, the already small ST-01 nanoparticles appear even smaller after the treatment. Also the SiO 2 nanoparticles appear to be slightly smaller. These findings are consistent whether the results of the Gauss fit on the minimum Feret diameter are concerned or the average size of the nanoparticles is calculated from the maximum and minimum Feret diameter. From the aspect ratio, it appears that the NM01004a nanoparticles become slightly more elongated by the treatment Release of 44 Ti radiolabels-leaching experiments  2 and 3). In experiment 1 a value of 3·10 −3 is reached after about 40 days, however, starting from a much higher base value (5·10 −4 ) since the washing procedure was stopped too early. Nevertheless, this means that even over extended observation periods, nanoparticles in aqueous pH neutral suspensions remain stably 44 Ti labelled. The same statement can be made for TiO 2 P25 where  Fig. 1 Comparison of the hydrodynamic diameter of as-received (cold) and 44 Ti-labelled nanoparticles of the TiO2 nanomaterials ST-01, P25 and NM01004a and of the SiO2 nanomaterial NM200. As abscissa the intensity of the dynamic light scattering is given free 44 Ti is still in the sub-10 −3 range after nearly 40 days. In spite of the different chemical nature of the amorphous SiO 2 matrix, also the NM200 nanoparticles maintain the fraction of free 44 Ti in the sub-10 −3 range up to an observation period of 60 days.

Discussion
The present work was undertaken (i) in order to confirm that different TiO 2 nanomaterials can be diffusion labelled with 44 Ti and (ii) that the 44 Ti-labelled nanoparticle construct remains stable over several weeks. Moreover, (iii) it was tested whether a chemically different (SiO 2 ) amorphous material could be stably radiolabelled applying the same non-chemical, purely physical procedure.
The present work is based on the experience gathered by Butz et al. (2011) and by Butz (2012) who investigated various TiO 2 nanomaterials by perturbed angular correlation using 44 Ti atoms as nuclear probes. The 44 Ti was introduced in the nanomaterials by impregnation of dry powders with 44 Ti dissolved in 4 M HCl and subsequent thermal treatment at 180°C. As described in the "Introduction" section, perturbed angular correlation measurements allow to distinguish whether the radioactive decay of 44 Ti takes place on a lattice site in the volume of a nanoparticle or on its surface or close to it. The results of Butz et al. (2011) and of Butz (2012) show unambiguously that impregnation followed by a mild thermal treatment leads to a distribution of the 44 Ti atoms in the whole nanoparticles as it can be expected for a diffusive penetration. Since the 44 Ti was applied by Butz (2012) in the same chemical form ( 44 Ti dissolved in 4 M HCl) and with a heat treatment at the same temperature of 180°C for 2 h (here 2.5 h), there is no physical or chemical reason to question the transferability of the results of Butz et al. (2011) and of Butz (2012) to the present case.
The only difference is the about 10 times higher 44 Ti activity concentration used by Butz (2012) in order to have sufficient statistics for the γ-γ coincidence measurements. For the present study, we can calculate the number N of used 44 Ti atoms from the applied activity A and the physical half-life T ½ of 44 Ti (N = A·T ½ /ln2) and get for 65 kBq a number of 1.75·10 14 atoms applied on 10 mg of nanoparticles. For nanoparticles with a diameter of 5 nm and of 25 nm, these 10-mg TiO 2 consist of a number of 3.9·10 16 and 3.1·10 14 nanoparticles, respectively. This implies that only one out of about 200 nanoparticles with a diameter of 5 nm will contain one 44 Ti atom. For the larger particles with a diameter of 25 nm, only every second nanoparticle will be labelled with a 44 Ti atom. In the present experiments, the activities are on the lower boarder for useful radiotracer experiments, and for biological or toxicological applications, higher activities up to a factor of 1000 could be envisaged. Even in this case, the fraction of radioactive to non-radioactive Ti atoms would still be as low as about 2.5·10 −3 .
These considerations imply that also for the labelling of SiO 2 nanoparticles, the number of foreign 44 Ti atoms is too low to induce modifications of the nanoparticle behaviour. The high 44 Ti losses during washing and the consequently low labelling yield experienced with amorphous SiO 2 nanoparticles indicate that the 44 Ti atoms are not easily integrated into the amorphous silica structure in spite of the open volume in amorphous structure fostering diffusion. However, the 39% that could not be washed off were stably integrated into the amorphous silica nanoparticle matrix as can be seen from the low leaching visualized in Fig. 3.
In view of the encouraging results with 44 Ti diffusion labelling of SiO 2 nanoparticles, this method might also be investigated for the radiolabelling of Al 2 O 3 nanoparticles since, as in the case of SiO 2 , neutron and charged particle activation of Al 2 O 3 does not lead to adequate radionuclides in sufficient quantity for radiotracer studies.
The washing procedure of the nanoparticles was stopped after a washing step in which less than 1 Bq of 44 Ti could be detected leaving the number of washing steps to reach this target flexible. This activity limit was chosen in order to have a residual amount of free activity in the range of 10 −5 as starting point for the leaching experiments. Based on the low leaching experienced with TiO 2 nanoparticles radiolabelled with 48 V by direct exposure to a proton beam (Kreyling et al. 2017a, b, c) and in view of intrinsic labelling, which promises best possible integration of the radioactive label in the crystal structure, we expected low leaching rates, which could be difficult to detect on a higher background of residualfree 44 Ti activity. Indeed, for all diffusion-labelled nanoparticle types, the fraction of free 44 Ti that could be determined in the aqueous suspensions was well below 5·10 −3 , while the radiolabelling yields for all TiO 2 nanomaterials were high and at least 97.5%. In view of the differences of these materials as compiled in Table 1, the search for such rather small effects would require sophisticated methods, which were not available in our controlled area. Only TEM specimens got the authorization to leave the controlled area as they exhibit such extremely low levels of radioactivity to overcome radiation protection concerns and concerns to contaminate expensive equipment with long-lived radionuclides.
Especially in the case of intrinsically labelled nanomaterials, the increase of free 44 Ti in TiO 2 nanoparticle suspensions over time may also be interpreted as a slow dissolution of the nanoparticles rather than as a phenomena caused by improperly integrated radiolabels. Assuming a homogeneous distribution of the radiolabels in the nanoparticles, the release of about 2·10 −4 of the radioactivity after 80 days (in the case of ST-01) would then be equivalent to 2·10 −4 of the nanoparticle mass being lost due to dissolution.
However, a possible weakness of the method was that the nanoparticles were impregnated with 44 Ti in 4 M HCl, which subjects the nanoparticles to extremely low pH and might modify the nanoparticle surface properties. We may speculate that 44 Ti in 4 M HCl is probably a mixture of dissolved Ti hydroxides and chlorohydroxides which are not prone to react with the surface OH groups of the nanoparticles and to bind there. During an increase of the pH, the precipitation of Ti-(OHx)Oy compounds on the nanoparticle surface might occur. However, in view of the proven diffusion of the 44 Ti, this would not prevent the 44 Ti from penetrating below the nanoparticle surface and becoming distributed in the nanoparticle volume (Butz (2012) and Butz et al. (2011)).
We did not determine the surface hydroxylation of the nanoparticles recovered in aqueous suspension. And we studied the stability only in aqueous suspension at neutral pH. Depending on the envisaged experiments, the chemical nature of the environment the nanoparticles will be subjected to the pH value, and physical parameters such as temperature may be different. Thus, the effects of labelling on essential physicochemical properties and leaching studies have to be performed under the specific experimental conditions for each envisaged application anyway.
Having a look at Table 4 shows that the labelling treatment slightly reduced the size of primary TiO 2 ST-01 and NM200 SiO 2 nanoparticles. In view of the applied processing steps, the most likely explanation is a corrosive attack during the impregnation with 44 Ti in 4 M HCl and the following thermal procedure at an elevated temperature of 180°C. All the following processing steps are much milder and shorter and take place at room temperature. However, the same procedures applied to TiO 2 P25 and NM01004a nanoparticles led to a slight increase of the nanoparticle size which is most likely related to the thermal treatment at 180°C for 2.5 h.
Whether these slight changes are acceptable must be discussed in the context of the envisaged special application. They are much smaller than the variations reported by Rasmussen et al. (2013Rasmussen et al. ( , 2014 from TEM measurements performed in different laboratories. However, in the present case, they were performed using the same instrumentation and the same methodology by the same operator. Thus, due consideration should be given to these differences. However, for an envisaged experimental setting, also the agglomeration status of the nanoparticles needs to be considered. The DLS results compiled in Table 3 indicate that in spite of slightly smaller primary ST-01 nanoparticles, the agglomerates might be slightly larger than for the as-received material. For the other nanomaterials, the data show no meaningful difference. A decision needs to weigh these findings against the possible benefit of high detection sensitivity and ease of detection. The applied temperature of 180°C for the thermal treatment may become the focus of future improvements of the method. Inagaki et al. (2009) performed isochronous furnace annealing experiments for 1 h in the temperature range between 100 and 800°C on ST-01 TiO 2 anatase nanoparticles and investigated the annealing effect on the material by XRD and BET measurements.   The effect of the annealing treatment on the size of single crystalline diffracting anatase domains has been determined from the full width at half maximum (FWHM) of the (101)-diffraction maxima of anatase (Inagaki et al. 2009) making use of Scherrer's formula (cf. Langford and Wilson 1978). The results of Inagaki et al. (2009) showed that the FWHM is constant up to temperatures of about 200°C and starts to drop from there, indicating growth of the diffracting domains, attributed to a sintering of primary nanoparticles to larger ones of about 20 nm (Inagaki et al. 2009). However, the specific surface area determined by the BET method The minimum Feret diameter has been evaluated as visualized in the histograms in Fig. 2a to d. The average minimum Feret diameter is presented ± its standard deviation, as well as the centre x c of the Gauss curve fitted to the histograms ± σ. The mean Feret diameter is reported as calculated from the minimum and maximum Feret diameter ± its standard deviation as well as the aspect ratio (maximum Feret diameter/ minimum Feret diameter) ± its standard deviation  (Brunauer et al. 1938) started to decrease in between the two measurements performed at 100°C and 200°C from about 260 m 2 g −1 to about 210 m 2 g −1 . Thus, future work might be required to investigate whether slightly lower temperatures (e.g. lower by 10°C) for the thermal treatment could reduce this loss of surface area without compromising the efficacy of diffusion labelling. Depending on the 44 Ti activity applied in the impregnation treatment and the amount of nanoparticles used, we produced specimens with a specific activity of 1700 Bq/mg (SiO 2 ) and up to 9200 Bq/mg for TiO 2 ST-01. The minimum detectable activity with our γspectrometry equipment is about 0.1 Bq. This implies that about 11 ng ST-01 nanoparticles can be detected. However, on demand this sensitivity can be improved by orders of magnitude. Butz (2012) impregnated 3-mg nanoparticles with 10 μL of 44 Ti in 4 M HCl with a specific activity of 20 kBq/μL which translates into 66 kBq/mg. Assuming the same high radiolabelling yield as in the present study, this translates into a detection sensitivity of 1.5 ng of nanoparticles. If we assume to substitute one out of 10,000 Ti atoms by a 44 Ti atom, we would obtain nanoparticles with a specific activity of 280 kBq/mg and could push the detection sensitivity down to 0.36 ng of nanoparticles. This appears technically feasible but will require a laboratory with special authorizations for the handling of long-lived open radioactive substances. Hildebrand et al. (2015) were working with much higher activity concentrations and showed that labelling yields of about 97% could be achieved with 45 Ti in P25 TiO 2 nanoparticles that exhibited specific activities of 10 MBq/mg and more. However, due to the much shorter half-live of 45 Ti (3.08 h), a much lower number of 45 Ti atoms are required to achieve this activity. Pushing the detection sensitivity for 44 Ti-labelled nanoparticles further would rather require synthesis of the nanoparticles from 44 Ti in a radioactive precursor than labelling. This will no longer be an option to study industrially manufactured nanoparticles.
A further big advantage of radiolabelled nanoparticles is the ease of measurement without special specimen preparation. Outside the scope of the present study which did not use dispersants, pronounced sedimentation occurred in the experiments. This made vortexing mandatory before taking any sample if not immediately after the ultrasound homogenization treatment. Sedimentation is the simplest processing step when performing a size selection (Kreyling et al. 2017a) to ensure experiments being performed with a true nano-fraction of a suspension. However, it may be cumbersome in such cases to determine the mass of nanoparticles in the supernatant. In the present work, we observed for the very agglomerated ST-01 TiO 2 nanoparticles that 1 mL of supernatant after 72 h of sedimentation contained about 90% less 44 Ti activity or, in other words, only 10% of the mass per mL than in the vortexed condition. Subjecting the supernatant to a DLS measurement, only one peak appeared and yielded a zaverage of 170.1 ± 1.1 nm and a PDI of 0.134 ± 0.007. This illustrates that radiolabelling also offers a simple method to control size selection procedures and to determine the recovered nanoparticle mass in the selected nanofraction. The same fraction used for the activity measurement can be used for further experimentation without losing material for other analysis techniques to determine the concentration.

Conclusion
Diffusion labelling with commercially available 44 Ti is a valuable method for intrinsic radiolabelling of various types of TiO 2 nanoparticles as well as for extrinsic radiolabelling of amorphous silica nanoparticles which offer only very limited possibilities for radiolabelling otherwise. The method is fast and reliable and offers very high detection sensitivity. It requires access to laboratories licenced for handling of long-lived radionuclides but avoids the involvement of nuclear accelerators or reactors. Radiolabelling is stable for periods of several weeks to months in aqueous dispersion at neutral pH. The method may have small effects on the primary particle size with changes in the range of 1 to 2 nm. Whether these may be tolerated must be evaluated on a case-by-case basis depending on the type of application and the benefit gained in detection sensitivity and ease of detection.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.