All-optical anatomical co-registration for molecular imaging of small animals using dynamic contrast

Abstract

Optical molecular imaging in small animals harnesses the power of highly specific and biocompatible contrast agents for drug development and disease research^1–7. However, the widespread adoption of in vivo optical imaging has been inhibited by its inability to clearly resolve and identify targeted internal organs. Optical tomography^8–11 and combined X-ray and micro-computed tomography (micro-CT)¹² approaches developed to address this problem are generally expensive, complex or incapable of true anatomical co-registration. Here, we present a remarkably simple all-optical method that can generate co-registered anatomical maps of a mouse’s internal organs, while also acquiring in vivo molecular imaging data. The technique uses a time series of images acquired after injection of an inert dye. Differences in the dye’s in vivo biodistribution dynamics allow precise delineation and identification of major organs. Such co-registered anatomical maps permit longitudinal organ identification irrespective of repositioning or weight gain, thereby promising greatly improved accuracy and versatility for studies of orthotopic disease, diagnostics and therapies.

Accurate optical molecular imaging of the internal organs of small animals could revolutionize in vivo drug and disease research^1–7. Yet, for non-invasive in vivo optical imaging, the effects of intrinsic absorption and scattering distort and attenuate signals from any target deeper than a few millimetres. This makes the study of in situ orthotopic tumours or diseased organs highly challenging. Further, because targeted fluorescent⁹ and bioluminescent¹³ molecular probes are designed to label only targeted cells, typical images do not reveal adjacent or landmark organs that could aid in identification of a targeted organ. Autofluorescence and nonspecific labelling further confound interpretation¹⁴. Superficial subcutaneous xenografts are much simpler to measure, but often do not resemble the human disease; for example, transplanted cancer-cell xenograft tumours are often surrounded by a pseudocapsule, have limited chances to invade major anatomical structures and rarely spread metastasis^15,16.

These difficulties have motivated the development of complex tomographic approaches and multimodality imaging systems that are generally expensive, complex and often inaccurate. We sought to devise a way to improve the interpretation of in vivo optical images of targeted contrast agents by providing an exactly co-registered anatomical overlay of the major internal organs of the small animal. In addition to being insensitive to animal repositioning and weight gain, our approach is far simpler and more cost-effective than existing approaches to anatomical co-registration. For example, three-dimensional optical tomography can help improve estimates of an organ’s location, and improve quantitation and orientation (Xenogen IVIS 3D Series, refs ^8–11). However, three-dimensional optical imaging is highly complex and expensive, and it still suffers from a lack of landmark structures to allow the actual anatomical location of the probe to be identified. Multimodality optical systems including X-ray (Kodak Image Station In-Vivo FX), micro-computed tomography (CT) (ref. ¹²), magnetic resonance imaging¹⁷ and ultrasound significantly increase the cost and complexity of the imaging process. In addition, X-ray contrast is dominated by bone, and micro-CT and ultrasound contrast cannot delineate or identify all internal soft-tissue organs. Imaging geometry differences also make co-registration challenging. Multispectral optical imaging can enhance the contrast of the target dye in the presence of autofluorescence¹⁴, but has not been used to delineate organs.

In response to this evident need for a more accessible method for generating co-registered anatomical maps, we present a simple all-optical approach that exploits the in vivo dynamics of an inert tracer dye as it circulates and accumulates in different organs following intravenous injection. We call this method dynamic fluorescence molecular imaging (DFMI). Each organ in the body plays a different role in circulating, accumulating or metabolizing a dye, and so each organ will exhibit a distinctive time course in its optical signal. This phenomenon has been documented in previous optical imaging studies^11,18,19, but has never been harnessed to directly provide anatomical information. Our approach also has foundations in perfusion imaging, an established technique used in positron-emission tomography, magnetic resonance imaging^20,21 and X-ray-CT (ref. ²²) imaging, which exploits the dynamics of an injected dye to delineate different functional structures.

In this letter, we describe a dynamic optical imaging system capable of capturing the in vivo biodistribution kinetics of two dyes in parallel. We demonstrate two processing methods by which the characteristic temporal signatures of each organ can be exploited to allow spatial delineation of the major internal organs. We also discuss how these dynamic imaging techniques could be further used to evaluate organ function, and also improve quantitative estimates of the targeted probe concentration.

DFMI data were acquired on five normal anaesthetized nu/nu (immunodeficient mice suitable for research in tumour biology and xenografts) mice using the system illustrated in Fig. 1. Each mouse was positioned between two angled mirrors, allowing simultaneous imaging of three orthogonal views. Fluorescence images were acquired following a tail-vein injection of a mixture of Indocyanine Green (ICG) and Dextran Texas Red (DTR). These tracer dyes were selected to investigate high and low molecular weights, and visible versus near-infrared imaging (see Supplementary Information, Fig. S1, for in vivo raw-image time series for the two dyes).

Figure 1. Dynamic imaging system.

a,b, System for dual-dye dynamic fluorescence molecular imaging. Two light sources and an emission-filter changer are computer-controlled to allow time series of the biodistribution of two dyes to be acquired simultaneously (a). The mouse is positioned between two angled mirrors, allowing simultaneous imaging of three orthogonal views of the same mouse (b). Note that the anatomical co-registration technique itself only requires imaging of one dye, and could equally be achieved using a typical two-dimensional imaging system.

Data were acquired for 40 min to explore the effects of using different time windows for analysis, although as little as 20 s of data is sufficient for anatomical co-registration. Note also that anatomical co-registration requires only a single suitable tracer dye, which could be delivered or induced by means of various routes, and could use many different configurations of positioning, illumination and detection. The tracer dye bolus could also be delivered after imaging of the targeted probe. We used two dyes and multispectral, orthogonal-view dynamic imaging to demonstrate that it is possible to image the in vivo dynamics of two dyes simultaneously (mimicking co-injection of an inert anatomical tracer and a targeted probe).

Analysis of DFMI images can be approached in many ways. Figure 2 shows the results of simple principal-component analysis (PCA) on the ICG image time series acquired in two mice, one lying prone (using 5 min of data) and the other supine (using 20 s of data). The spatial pixels corresponding to the second, third and fourth orthogonal temporal components of the image time series are visualized as red–green–blue (RGB) merged images (the first component is the mean image). Positive and negative pixels are shown separately.

Figure 2. In vivo anatomical maps derived using PCA of an image series following ICG injection.

a,b, Principal components of images acquired for 5 min after ICG injection (first temporal component, red; second, green; third, blue): positive (a) and negative (b) pixels. c, Corresponding PCA basis time courses. d,e, Principal components of images acquired for 20 s after ICG injection in a second mouse that was positioned supine. f, Corresponding PCA time courses. Each organ has a different RGB combination, and hence its own distinctive time course. Small arrows in a indicate possible lymph nodes, blue structures in b may be corresponding lymph drainage channels. Spikes in f and ghosting in e correspond to breathing motion.

The structures revealed are very clearly delineated, and most are simple to identify when compared with anatomy and dissection. No image segmentation was required. (See Supplementary Information, Fig. S2, for equivalent images for mouse A using only 0–20 s of data.) We found that different anatomical structures are emphasized when different lengths of time series are used. This is because the dye may clear very quickly through some organs (such as the brain), but may take a long time to build up in others (such as adipose tissue). In all five mice, it was possible to create an anatomical map using the image time series acquired.

Although PCA is effective in delineating structures, it is not suitable for routine analysis. However, because our spatiotemporal separation is based on the biodistribution dynamics of the organs themselves, these characteristics are founded in physiology and their general trends should be consistent and repeatable. Work is under way to fully characterize these trends and develop a universally applicable algorithm that uses established processing techniques from perfusion imaging²¹. As a first demonstration of a more generalizable approach, Fig. 3 shows the result of a nine-component non-negative least-squares fit based on a single ICG image time series. Nine basis time courses (bone, kidney, brain, small intestine, liver, spleen, lungs, eyes/lymph nodes and adipose) were extracted from small regions of interest, selected using the PCA images as a guide. The fitting process then reveals the extent to which each pixel is exhibiting each particular time course. Figure 3 shows these components colour-coded and merged (see Supplementary Information, Fig. S3 for individual components, and Fig. S4 for a comparison to a digital anatomical mouse atlas). This co-registered map could readily be overlaid onto simultaneously or sequentially acquired fluorescence or bioluminescence images of a targeted probe.

Figure 3. In vivo, non-invasive anatomical mapping of nine organ-specific spatiotemporal components.

a, Time courses of pixels selected from locations in the image time series expected to correspond to particular organs (selection aided by Fig. 2a,b). A non-negative least-squares fit of these time courses to the full data set identifies all pixels with the same temporal behaviour. b, The resulting nine spatial maps are shown colour-coded and merged (see Supplementary Information, Fig. S3, for individual representations). In addition to major organs, the spine and even sutures in the skull are visible non-invasively. The dynamics are distinct because each organ exhibits different circulatory, uptake and metabolic responses to ICG.

We have presented the first demonstration of an elegantly simple, yet incredibly powerful technique for simultaneous, all-optical generation of a co-registered anatomical atlas for small-animal molecular imaging. The technique exploits the simple fact that each organ responds differently following injection of a dye.

Although other dyes or combinations thereof may prove equally suitable, ICG is ideal for this application as it is clinically available, well characterized, and it excites and emits in the near infrared (NIR). The ability to emit in the NIR is the probable reason for the excellent resolution of the organs, as light scatter is lower and tissue penetration is higher at NIR wavelengths²³. Note that a static image of ICG would not provide any delineation of the internal organs, as each organ does not possess specific contrast in its steady state.

All-optical anatomical co-registration may have additional useful properties. By providing improved estimates of the targeted probe’s geometrical position through multiple orthogonal views, it may be possible to improve quantitative estimation of the probe’s concentration²⁴. This could be achieved by calculating a correction factor based on the probable attenuation of light as it passes through surrounding organs on its way to and from the localized targeted probe.

The in vivo dynamics of inert dyes could also provide non-invasive measures of changes in the function of major organs. For example, as ICG is used clinically to evaluate liver function²⁵, DFMI could be used to simultaneously provide non-invasive measurement of the effects of a drug on the liver.

We also assert that imaging the in vivo dynamics of targeted probes themselves (whether activatable or injected) could allow enhanced resolution and specificity by applying these same dynamic imaging techniques. Potentially valuable biodistribution dynamics have been noted in many molecular probe development studies^18,19 and exploitation of uptake pharmacokinetics as a quantitative measure has been demonstrated and explored^26–29. Dynamic optical imaging of haemoglobin absorption in humans has also been shown to enhance image contrast³⁰.

To reach its full potential, optical molecular imaging in small animals needs to evolve into a technique capable of routine, longitudinal studies of orthotopic targets. We have demonstrated that DFMI can provide an important contribution to this goal, as a simple and inexpensive method of simultaneously achieving anatomical co-registration irrespective of repositioning or weight gain.

METHODS

ANIMALS

Five mice (nu/nu, Massachusetts General Hospital Radiation Oncology breeding facilities; n = 5, weight = 25±1 g, 6 weeks old) were anaesthetized with isoflurane, 1.25% in a 1:3 mixture of O₂ and air. For optical imaging, animals were injected intravenously with a mixture of 0.05 ml of 260 μM ICG (Cardiogreen, Fluka) and 0.05 ml of 360 μM DTR (70,000 MW, Invitrogen) in saline (equivalent to 0.4 mg kg⁻¹ or 7.4 μM initial concentration in blood). All animal procedures were reviewed and approved by the Subcommittee on Research Animal Care at Massachusetts General Hospital, where these experiments were performed. Mice recovered fully after imaging.

DYNAMIC OPTICAL IMAGING SYSTEM

The DFMI system was built to allow parallel multispectral dynamic molecular imaging and included an imaging bay consisting of two glass mirrors at 45° and a slightly raised central platform as shown in Fig. 1. Bifurcated liquid light guides delivered 9 mW of filtered white light at 570±20 nm from two positions to excite the DTR dye. Two 785-nm laser diodes (160 mW total) delivered light to excite the ICG dye. A 12-bit cooled charge-coupled device (CCD) camera (Cascade, Roper Scientific) imaged the mouse through a computer-controlled emission filter-changer (Electro-Optical Products) holding a 600-nm long-pass filter, and an 850±20 nm bandpass filter. An electronic shutter in the 570-nm light path, and digital modulation of the laser diodes allowed synchronization of illumination with the emission filters and camera image acquisition. A series of 10 images (50 ms integration time per frame) were acquired every two seconds for each excitation/emission pair for up to 40 min after injection. For 20-s image series analysis, a sequence of 55 successive raw images was used (corresponding to 5 Hz frame rate imaging during periods 0–2, 4–6, 8–10, 12–14, 16–18 and 20–21 s). For 5-min image series analysis, each group of 10 successive images was averaged to create a sequence of 75 images with four seconds between each. Image analysis was performed as described below.

IMAGE PROCESSING

Image processing and data analysis were performed using Matlab software functions including PCA (princomp) and non-negative least-squares fitting (lsqnonneg). Red–green–blue colour merging in Fig. 2 was achieved by background subtracting each of the three components, normalizing their peak values to 256, and then combining them into an unsigned 8-bit integer true-colour image. To aid visualization, a pale version of the grey-scale bright-field image of the mouse taken before or after the image time series was then superimposed by adding it to the RGB merge and renormalizing the maximum to 256. Reading in data, PCA processing and display of 256×256 CCD image sequences took 11 seconds on a 2 GHz Pentium M laptop with 1 Gbyte RAM.

For the nine-component image merge in Fig. 3, each organ component was background subtracted and normalized to 256, and then converted to an RGB matrix of shades of only the colour key for that organ (as shown in the time-course plot—note that ‘spleen’ was in fact white). The final image was then generated by adding these nine colour-coded images together. Because of this, any overlap between organs may appear as a mixture of colours that may resemble another colour. However, there is surprisingly little overlap between the organs (see Supplementary Information, Fig. S3, which shows each organ component separately).

Supplementary Material

Supplemental D

Supplementary Information accompanies this paper on www.nature.com/naturephotonics