Cryo-scanning x-ray diffraction microscopy of frozen-hydrated yeast

Authors


Enju Lima, Photon Sciences Directorate, Brookhaven National Laboratory, Upton, NY, 11973 USA. Tel: 631-344-4918; e-mail: elima@bnl.gov

Summary

We developed cryo-scanning x-ray diffraction microscopy, utilizing hard x-ray ptychography at cryogenic temperature, for the noninvasive, high-resolution imaging of wet, extended biological samples and report its first frozen-hydrated imaging. Utilizing phase contrast at hard x-rays, cryo-scanning x-ray diffraction microscopy provides the penetration power suitable for thick samples while retaining sensitivity to minute density changes within unstained samples. It is dose-efficient and further minimizes radiation damage by keeping the wet samples at cryogenic temperature. We demonstrate these capabilities in two dimensions by imaging unstained frozen-hydrated budding yeast cells, achieving a spatial resolution of 85 nm with a phase sensitivity of 0.0053 radians. The current work presents the feasibility of cryo-scanning x-ray diffraction microscopy for quantitative, high-resolution imaging of unmodified biological samples extending to tens of micrometres.

Introduction

Structural studies in biology and life science greatly benefit from observing samples in their native state. Despite the broad range of microscopy techniques, surveying entire cells or sub-cellular organelles in their natural conditions remains challenging and often entails the need to stain or section the specimen. Artifact-free, high-resolution imaging of extended samples, such as large eukaryotic cells or tissues, could provide a comprehensive visualization of the complex cellular environment (Leis et al., 2009) and, by directly revealing structural changes that are correlated with functional changes, it could further our understanding of cellular mechanisms (Acehan et al., 2007; Schmolze et al., 2011). The ability to see an entire, unperturbed picture of samples would naturally complement other high-resolution methods that image only selectively labelled structures, such as super-resolution fluorescence microscopy (Jones et al., 2011; Westphal et al., 2008).

For unstained samples in their natural condition, radiation damage poses the ultimate limit in achieving high-resolution (Glaeser & Taylor, 1978; Sayre et al., 1977). Imaging frozen-hydrated specimens not only avoids the need for chemical fixation by preserving samples in vitreous ice, but also provides excellent protection against radiation damage (Adrian et al., 1984; Maser et al., 2000; Schneider 1998; Taylor & Glaeser, 1974). Although vitrified thin samples can be imaged by cryo-electron microscopy (cryo-EM) at nm resolution (Beck et al., 2007; Henderson et al., 2007), due to the short inelastic mean-free-path of electrons, cryo-EM requires sectioning of thick samples including most eukaryotic cells or tissue arrays (Leis et al., 2009). In contrast, x-rays have a large attenuation length, thus, sufficient penetration to probe these whole samples, as shown in soft x-ray tomography at the water window (Hanssen et al., 2011; Schneider et al., 2010; Uchida et al., 2009) and x-ray diffraction microscopy of frozen-hydrated samples (Huang et al., 2009; Lima et al., 2009).

Soft x-rays at the water window, between 284 and 540 eV, with an attenuation length of 2–10 μm in water provide excellent absorption contrast in several-μm-thick, hydrated specimens (Kirz et al., 1995). For thicker specimens, above 5–10 μm, a high penetrating power is needed to image the whole samples whereas a marked imaging sensitivity is required to detect minute density changes, especially in unstained states. Some of these capabilities are shown in x-ray phase contrast imaging with dried samples (Cloetens et al., 1999; Stampanoni et al., 2010; Holzner et al., 2010). Utilizing hard x-rays and robust algorithmic convergence, ptychographic or scanning x-ray diffraction microscopy (SXDM) offers these capabilities for thick samples at a resolution not limited by the spatial resolution of the detector or x-ray optics (Rodenburg et al., 2007; Thibault et al., 2008). SXDM was recently applied in two dimensions to freeze-dried bacteria (Giewekemeyer et al., 2010) and in three dimensions to polymerized bone tissue (Dierolf et al., 2010a) and stained or dried cells (Guizar-Sicairos et al., 2011; Wilke et al., 2012), which demonstrated quantitative density analysis of the samples as well. Yet, the technique demands extremely stable experimental setups for scanning accuracy. Because of SXDM's high sensitivity to vibrations and drifts, it has been uncertain whether it can be reliably applied to cryofixed, frozen-hydrated samples. In this paper, we report the first demonstration of cryo-SXDM for high-resolution, high-sensitivity imaging of frozen-hydrated biological samples.

By phase imaging samples in a scanning mode with hard x-rays, cryo-scanning x-ray diffraction microscopy (cryo-SXDM) presents a complementary imaging method for frozen-hydrated, extended biological samples in an unmodified state, bridging the current gap in resolution and accessible sample thickness among microscopy techniques. The feasibility of cryo-SXDM depends on scanning accuracy, at cryogenic temperature for high resolution, and imaging sensitivity to discern unstained biological materials. We addressed the long-term instability associated with a commercial cryo-jet system by postprocessing and averaging reconstructions from multiple short-term measurements to reach sub-100 nm resolution at hard x-rays. The experiment was conducted at 6.2 keV x-ray photon energy, providing an attenuation length of 450 μm in water, and the reconstruction of unstained, frozen-hydrated yeast cells shows quantitative phase images of budding cells at 85 nm resolution with a phase sensitivity of 0.0053 radians within the sample. The demonstrated phase sensitivity is equivalent to detecting the phase shift of 6.2 keV x-rays due to an 80-nm-thick protein structure in water. The current resolution of cryo-SXDM in this feasibility experiment is limited primarily by the short-term instability induced by the cryo-jet system whereas two-dimensional projection ambiguity reduced the imaging sensitivity. The extension to three-dimensional imaging by tomographic methods is currently underway and we expect higher resolution with a high-scanning-accuracy, vacuum-compatible cryo-SXDM system in the future.

Materials and methods

Ptychographic data collection of frozen-hydrated yeast

In this feasibility experiment, we chose to image a strain of yeast cells, Pichia Pastoris (Fig. 1A). Yeast cells were cultured in YPD medium until an exponential growth rate was reached. Following the cryogenic sample preparation of Lima et al. (2009), nylon loops (CryoLoop™, Hampton Research, Laguna Niguel, CA, USA) containing a thin layer of culture solution, with 10% glycerol by volume added as a cryo-protectant, were plunge-frozen into liquid ethane. The setup to measure the diffraction data of cryofixed yeast is illustrated in Figure 1(B). The x-ray probe of 6.2 keV was provided by a zone plate (Gorelick et al., 2010) with a 100 μm diameter and 100-nm outermost zone width, which was illuminated coherently by filtering through a 20-μm-diameter aperture in front. The higher focusing orders of the zone plate and the direct beam were removed by an order sorting aperture of 20 μm diameter placed ∼10 mm upstream of the focal plane. Frozen sample loops were placed 10.6 mm downstream from the focal plane to allow a suitable working space for the cryogenic nitrogen stream. The x-ray probe at the sample plane was then ∼4 μm in diameter.

Figure 1.

The experimental details. (A) An optical microscope image of cultured yeast cells, Pichia Pastoris. Samples from the same culture were used for later x-ray imaging. The scale bar indicates 5 μm. (B) The experimental setup for x-ray measurement. From left to right: a Fresnel zone plate with an off-axis aperture, an order sorting aperture (OSA), cryogenic nitrogen gas stream environment with a sample loop, and a pixelated detector to measure the x-ray diffraction patterns. (C) A measured diffraction pattern of a frozen-hydrated yeast cell is shown in a log scale.

Ptychographic diffraction data (an example is shown in Fig. 1C) were measured from different overlapping positions on a sample with a Pilatus detector (Henrich et al., 2009), which has a pixel size of 172 μm and was situated 7.2 metres from the sample. Within the setup of optics and sample stages, the cryogenic gas flow caused uncontrolled temperature gradients and long-term positional drifts. For long scanning times, this induced positional uncertainty or changes in the illumination, effectively limiting the maximum scanning time that could be used for a single scan. Experimentally, we observed reconstruction artifacts arising from scans lasting more than 10 min with the current setup. Consequently, multiple fast scans of ∼5 min each, sufficiently shorter than the artifact-inducing span, were obtained and their independent reconstructions were combined to yield the final image of the sample. For overview scans, cryo-SXDM data were collected in circular shells with a 1.2-μm radial step size and 0.5-s exposure time per diffraction pattern. The sample scans were measured with a 1-μm radial step size and 1-s exposure time.

Preliminary probe recovery prior to yeast cell imaging

In the case of x-rays, the refractive index of matter varies only slightly from unity and is conveniently written inline image (Kirz et al., 1995), where for hard x-rays in particular inline image. Thus, with the wavenumber inline image, the phase shift inline image of the wave passing through the sample is dominant compared to the attenuation inline image, where inline image is the wavelength and inline image is the sample thickness. Hence phase contrast imaging becomes favourable at hard x-rays. For frozen-hydrated samples, the phase contrast comes mainly from the relative phase shift between organic matter and water, inline image. With 6.2 keV photons, for example, a 100-nm-thick protein structure, H50C30N9O10S1 (Howells et al., 2009), in a water background gives inline image radians, requiring high sensitivity in the imaging system.

To achieve the necessary sensitivity in SXDM, an accurate reconstruction of the illuminating probe is crucial since a probe can have more than π phase variations across the wavefront. To ensure a reliable probe recovery with weakly scattering samples, here with wet, unstained yeast cells, each supporting nylon loop of approximately 10 μm diameter was imaged as a high-contrast object before the yeast cells were scanned. The reconstructed probe was then used as an initial estimated probe for overview online reconstructions of yeast cells (Dierolf et al., 2010b). One such loop and the corresponding probe reconstruction are shown in Figure 2. Figure 2(A) shows a part of a nylon loop with a surrounding ice region and a cold nitrogen gas environment. Although this particular loop was contaminated with crystalline ice, these ice particles served as high-contrast objects to facilitate the probe recovery. The reconstructed probe intensity and phase are shown in Figure 2(B) and (C), respectively. Through a numerical propagation, the focus of the zone plate was determined to be at 10.6 mm upstream from the sample plane as shown in Figure 2(D).

Figure 2.

Reconstructions of a nylon loop and a probe. (A) A reconstructed section of a nylon loop with a suspended ice layer and a surrounding cold nitrogen gas area. The phase image of the sample is shown in radians. In this particular case, ice crystals, with sizes ranging from a few hundred nm to about 1 μm, were found around the loop, which provide high-contrast scattering points within the sample, facilitating a probe recovery. (B)–(C) Reconstructed probe at the sample plane, showing its intensity (B) and the phase (C). (D) Through focus two-dimensional slice of the probe computed by numerical propagation. The hue represents the phase as indicated in the colour scale and the brightness indicates the amplitude.

Reconstruction process

The cryo-SXDM reconstruction process before final averaging is as follows. The coherent diffraction data along with the known scan positions and an initial guess for the illumination probe are inputs to the SXDM difference-map reconstruction algorithm (Thibault et al., 2008), followed by a refinement step of maximum likelihood nonlinear optimization (Guizar-Sicairos & Fienup, 2008; Thibault & Guizar-Sicairos, 2012). The constant phase offset and linear phase gradient, which are inherent degrees of freedom when simultaneously retrieving sample and probe (Guizar-Sicairos et al., 2011), are set such that phase images are, on average, zero on the ice background. This allows reconstructions to depict the quantitative x-ray phase shift with respect to the ice background.

Result and discussion

Quantitative phase images of frozen-hydrated yeast cells

The reconstructed images of frozen-hydrated yeast cells are shown in Figure 3. The images show quantitative phase variations, in radians, of the sample's transmission function, inline image, where inline image corresponds to the accumulated phase shift of the x-ray beam along the propagation in inline image, whereas inline image and inline image are the directions perpendicular to the beam propagation, and inline image and inline image are as defined earlier. The final reconstruction of frozen-hydrated yeast cells is shown in Figure 3(A), which shows yeast cells in the budding state. Due to the low contrast two-dimensional projection associated with the overlay of a complex three-dimensional structure, the individual organelles are not identifiable in the current image. We expect cryo-SXDM to further improve image contrast by tomographic reconstruction, as has been shown in previous work where features that are not identified from a single projection become prominent in a sliced plane or through segmentation of a three-dimensional structure (Dierolf et al., 2010a; Guizar-Sicairos et al., 2011; Diaz et al., 2012).

Figure 3.

Reconstructions of yeast cells and the estimated phase sensitivity of cryo-SXDM. (A) The final, averaged image from eight individual reconstructions, one of which is shown in (C). (B) An online reconstruction of 20 μm × 20 μm overview area to screen a sample region. The square shows the selected samples, (A), for high-resolution imaging. (C) The reconstruction of selected yeast cells from a single scan. (D) Phase profile in a yeast cell, along the red line shown in (A). The standard error of the mean (SEM) was computed from the eight independent reconstructions. The phase offset is arbitrarily determined to set the phase of ice to be zero. The arrow indicates the phase dip of 0.0053 radians.

These budding yeast cells were first located in situ by an overview scan and online reconstruction (Fig. 3B). Utilizing the previously recovered probe as an initial guess for the illumination, as shown in Figure 2(B) and (C), such an overview reconstruction took only 10 iterations to identify samples for imaging. Furthermore, a relatively low x-ray dose of 1.2 × 105 Gray (Gy) was imposed to minimize radiation damage during the overview scan. Some reconstruction artifacts, such as bright rings around cells, are due to the preliminary scan and reconstruction.

To achieve high-resolution images, high x-ray fluence is required; consequently, a high dose is imposed on the samples (Howells et al., 2009; Shen et al., 2004). However, with a long exposure time per scan point, long-term instability under the cryogenic gas flow introduced significant reconstruction artifacts. The strategy to overcome this limitation was to measure multiple cryo-SXDM datasets on the same region and to combine the resulting reconstructions. This computational approach reduces the degree of mechanical stability required in the experimental setup. For the chosen sample here, cryo-SXDM data were collected eight times with a total accumulated dose of 2.8 × 106 Gy. Each of these data sets yielded an independent reconstruction, one of which is shown in Figure 3(C). The final image of the samples (Fig. 3A) was obtained by averaging these reconstructions after aligning them using a subpixel cross-correlation image registration algorithm (Guizar-Sicairos et al., 2008).

Phase sensitivity of cryo-SXDM

The red curve in Figure 3(D) represents a phase variation along the red line in Figure 3(A). Compared to the surrounding ice region, the yeast cell produced about 0.04-radians additional phase shift on the 6.2 keV x-rays. The achieved phase sensitivity in the averaged reconstruction (Fig. 3A) was estimated by the measured phase dip (indicated by an arrow) in Figure 3(D), which is 0.0053 radians and is visible in all eight reconstructions. To support this claim, we computed the standard error of the mean (SEM) from the eight independent reconstructions on a pixel-by-pixel basis, which is indicated by the vertical width of the yellow area in Figure 3(D). The short-term instabilities during a single scan caused low-spatial-frequency artifacts in the individual reconstructions. Although these artifacts are random and efficiently removed by averaging, they locally introduce different constant and linear phase terms per reconstruction. For the analysis of the imaging sensitivity (Fig. 3D) we were mainly interested in variations among different images on the length scale comparable to that of a whole cell, which in this case was below 4 μm. To compute the SEM in Figure 3(D), we selected a 4 μm × 4 μm area around the cell of interest and locally matched constant and linear phase terms between them. We did not use higher order background removal in order not to introduce artificial similarities on the image content. The average value of the SEM was 0.0019 radians, which indicates the estimated phase error in Figure 3(D) due to noise and reconstruction artifacts. This confirms the dip inline image in the middle of the phase line (Fig. 3D), to be of structural origin. Although the individual organelles are not identifiable in the current two-dimensional projection image, the result indicates that cryo-SXDM has shown feasibility of obtaining 0.0053-radian phase sensitivity, where a protein structure of 80-nm thickness in a water background produces an equivalent phase shift at the current photon energy.

Resolution estimation

The lack of sharp features in the projection images makes the direct estimation of the achieved resolution rather difficult. As shown in Vila-Comamala et al. (2011), Fourier ring correlation (FRC) can be used to obtain a resolution estimate for SXDM reconstructions. FRC, also referred to as Fourier shell correlation for its extension to three dimensions (Saxton & Baumeister, 1982; van Heel & Schatz, 2005), is a measure of the normalized cross-correlation coefficient on corresponding spatial frequency rings between a pair of images. FRC curves of reconstructed images are shown in Figure 4. As expected, there is a high correlation at low-spatial frequencies associated with a high signal-to-noise ratio (SNR), and this correlation decays for increasing spatial frequency. Short-term instabilities within a single scan introduced small-amplitude, low-frequency artifacts, which have significant effects on the FRC (blue curve) for a pair of single-scan reconstructions (an example in Fig. 3C). These low-frequency artifacts are uncorrelated between reconstructions of different scans and, by averaging many reconstruction images, they are visibly reduced in Figure 3(A) with the improved FRC in Figure 4 (black curve).

Figure 4.

Resolution estimation by Fourier ring correlation (FRC) as a function of the normalized spatial frequency (spatial frequency divided by the frequency corresponding to a pixel resolution). The black line corresponds to FRC of the 8-averaged and the blue to the 2-averaged reconstruction. To estimate the resolution of the average of eight images (Fig. 3A), we first randomly selected two sets of four images each and the images in each set were registered and averaged. The FRC was then computed between the two averages. The distinct dips in the blue curve, around normalized frequencies of 0.0005 and 0.14 are due to artifacts caused by short-term instability, which are significantly reduced by averaging (black curve).

To estimate the resolution-cutoff we use a 1/2 bit threshold (red curve in Fig. 4). The spatial frequency at which the FRC falls below the threshold curve determines the resolution up to which an SNR of 0.41 is observed (van Heel & Schatz, 2005). This gives a resolution of 85 nm for the average of eight images and 115 nm for the average of two images. The observed relation between dose and resolution is in good agreement with previous theoretical estimations (Howells et al., 2009; Shen et al., 2004). As expected from cryo-preservation of wet samples, no signs of radiation damage, such as mass loss between scans (Williams et al., 1993), were observed with a total estimated dose of 2.9 × 106 Gy, including overview scans. The current resolution is limited by the short-term instability of the scanning system, hindered by the cryogenic gas stream. This limitation is expected to improve significantly with a vacuum-compatible cryogenic sample environment and an interferometrically controlled scanning system, some features of which have been recently demonstrated (Holler et al., 2012).

Estimated dose

The radiation dose inline image absorbed by the cells during a scan was determined by inline image (Howells et al., 2009) where inline image is the number of incident photons per unit area, inline image= 6.2 keV is the photon energy, inline image 1.13 g cm−3 is the mass density of yeast cells (Bryan et al., 2010) and inline image 2.4 × 10−3(μm−1), where inline image is the cell attenuation length. The attenuation length was obtained from tabulated values (http://henke.lbl.gov/optical_constants/) assuming a 22% protein, H50C30N9O10S1, and 78% water concentration, in accordance with the observed phase shift with respect to the ice background. For an individual high-resolution scan (Fig. 3C), inline image 1.67 × 108 photons inline imagem−2 illuminated the samples with a resulting dose of inline image 0.35 × 106 Gy. Thus, the image obtained by averaging eight reconstructions (Fig. 3A) has an accumulated dose inline image 2.8 × 106 Gy. In contrast, for the overview scan (Fig. 3B) that was used to locate regions of interest, inline image 5.7 × 107 photons μm−2 gives a relatively low dose inline image 0.12 × 106 Gy.

Conclusion and outlook

We have demonstrated the feasibility of cryo-SXDM for high-resolution imaging of cryofixed, wet biological samples. The reconstructions show quantitative phase images of unstained frozen-hydrated yeast cells at 85 nm resolution with a phase sensitivity of 0.0053 radians. The penetrating power of 6.2 keV x-rays and the demonstrated field of view are well suited for imaging large, whole samples including eukaryotic cells, tissue segments, or small organisms. Fast overview scans with low x-ray doses and preliminary on-the-fly reconstructions provided convenient in situ screening of the frozen-hydrated samples. In this feasibility experiment, we have addressed long-term instability caused by a commercial cryo-jet system by post-processing and averaging reconstructions from multiple short-term scans to reach sub-100 nm resolution. This computational approach alleviates the long-term stability requirement of the current setup, which might be applicable to other scanning-imaging systems. We expect to achieve higher resolution with a future, high-scanning-accuracy cryo-SXDM setup, which is currently under development.

Cryo-SXDM imposes no stringent restrictions on the sample and promotes imaging large cells or tissues in their native environment. Using hard x-rays and exploiting phase contrast, it allows for the imaging of thick structures that may not be accessible by soft x-rays or electrons. The technique can be extended to three-dimensional imaging of frozen-hydrated samples by tomographic methods. We expect cryo-SXDM to become a valuable complementary imaging method in the pursuit of multiscale structural studies from the tissue to the cellular level.

Acknowledgements

X-ray measurements were performed at the cSAXS beamline (X12SA) of the Swiss Light Source, Paul Scherrer Institut, Villigen, Switzerland. The work at Brookhaven National Laboratory was supported by the U.S. Department of Energy, Office of Science, under Contact No. DE-AC02–98CH10886. The authors thank Roger Wepf for providing the cryo-plunging equipment and Xavier Donath for excellent technical support at the cSAXS beamline.

Ancillary