• The principles underlying the formation of leaf veins have long intrigued developmental biologists. In leaves, networks of vascular precursor procambial cells emerge from seemingly homogeneous subepidermal tissue through the selection of anatomically inconspicuous preprocambial cells. Understanding dynamics of procambium formation has been hampered by the difficulty of observing the process in vivo.
• Here we present a live-imaging technique that allows visual access to complex events occurring in developing leaves. We combined this method with stage-specific fluorescent markers in Arabidopsis (Arabidopsis thaliana) to visualize preprocambial strand formation and procambium differentiation during the undisturbed course of development and upon defined perturbations of vein ontogeny.
• Under all experimental conditions, we observed extension, termination and fusion of preprocambial strands and simultaneous initiation of procambium differentiation along entire individual veins.
• Our findings strongly suggest that progressiveness of preprocambial strand formation and simultaneity of procambium differentiation represent inherent properties of the mechanism underlying vein formation.
The vascular system of plants is a network of cell files extending through all organs (Esau, 1965). Vascular cells differentiate from procambial cells: narrow, cytoplasm-dense cells, characteristically arranged in continuous strands (Esau, 1943). In leaves, procambial strands develop from files of isodiametric ‘preprocambial’ cells, which are selected from the anatomically homogeneous subepidermal tissue of the leaf primordium, the ground meristem (Foster, 1952; Pray, 1955). The mechanism by which cells are specified to procambial cell fate is unknown, but an instrumental role for auxin transport and distribution in this process has increasingly gained support (Sachs, 1981, 1989; Mattsson et al., 1999, 2003; Sieburth, 1999; Scarpella et al., 2006).
In this study, we monitored dynamics of procambium formation in developing leaves of Arabidopsis (Arabidopsis thaliana). To this aim, we designed a noninvasive live-imaging technique based on low-light fluorescence microscopy that allows for long-term visualization of developmental dynamics in leaves. We identified stage-specific fluorescent markers for high-resolution analysis of preprocambial strand formation and procambium differentiation in leaf primordia and employed them to reconstruct patterns of procambium development over time. Our analysis shows that vein patterns are generated by the extension, termination and fusion of preprocambial strands and the simultaneous differentiation of procambium along the entire length of individual veins.
Materials and Methods
Plant material and growth conditions
Enhancer-trap lines of the Haseloff collection (Haseloff, 1999) were obtained from the Arabidopsis Biological Resource Center. The origin of the PIN1:YFP line has been described (Xu et al., 2006). To generate the Athb8::NLS:YFP construct, 1997 bp upstream of the Athb8 ORF were amplified from genomic DNA using Finnzymes Phusion high-fidelity DNA polymerase (New England BioLabs Inc., Ipswich, MA, USA) with the ‘Athb8 attB1F’ (GGGGACAAGTTTGTACAAAAAAGCAGGCTGACGATAATGATGATAACTAC) and ‘Athb8 attB2R’ (GGGGACCACTTTGTACAAGAAAGCTGGGTCTTTGATCCTCTCCGATCTCTC) primers, integrated into pDONR221 (Invitrogen, Carlsbad, CA, USA) using BP clonase II (Invitrogen), sequence-checked and recombined into the Gateway®-adapted pFYTAG binary vector (Zhang et al., 2005) using LR clonase II (Invitrogen). To generate the Athb8::CFP:NLS construct, the pDONR221-integrated 1997-bp Athb8 upstream regulatory sequences were recombined into the Gateway-adapted pBGCN binary vector (Kubo et al., 2005) using LR clonase II (Invitrogen). Arabidopsis plants (ecotype Col-0) were transformed with Agrobacterium tumefaciens strain GV3101::pMP90 (Koncz & Schell, 1986) harboring the Athb8::NLS:YFP and Athb8::CFP:NLS constructs by the floral dip method (Clough & Bent, 1998). In all experiments, seeds were surface-sterilized, synchronized, and germinated on germination medium – half-strength Murashige and Skoog (MS) salts (Sigma, St Louis, MO, USA), 15 g l−1 sucrose (Merck KGaA, Darmstadt, Germany), 0.5 mg l−1 Mes (Sigma), 0.8% (w/v) agar (BioShop Canada Inc., Burlington, ON, Canada), pH 5.7 – at the approximate density of 1 seed cm−2 as previously described (Scarpella et al., 2004). Sealed plates were incubated at 25°C under continuous fluorescent light (100 µmol m−2 s−1 over the waveband 400–700 nm). We define ‘days after germination’ (DAG) as days following exposure of imbibed seeds to light.
Culture system for live imaging of Arabidopsis leaves
All procedures were performed in a laminar-flow cabinet, unless otherwise stated. For preparation of six culture chambers, three sterile microscopy slides (single frosted, 26 × 76 mm, 1 mm thick; Bio Nuclear Diagnostic Inc., Toronto, ON, Canada) were juxtaposed along their long sides (Fig. 1a). Approximately 20 µl of sterile water were added on each end of the three slides (Fig. 1b), and two additional sterile microscopy slides were placed perpendicularly on top of the ends of the three juxtaposed slides, allowing c. 0.5 cm overhang at the edges (Fig. 1c). Approximately 5 ml of cooled (c. 60°C) growth medium (i.e. germination medium in which 0.8% agar was replaced with 1% (w/v) agarose (BioShop Canada), as agar exhibited fluorescence under our imaging conditions) supplemented with 50 mg l−1 of nystatin (BioShop Canada) to control fungal contamination were dispensed over the three juxtaposed slides (Fig. 1d). Immediately, a sterile glass plate was lowered onto the mould at an angle to reduce air bubble formation (Fig. 1e,f). Pressure was applied evenly to eliminate possible bubbles and to spread the medium evenly (Fig. 1g). After the medium had set completely (c. 5 min), the glass plate was gently and carefully slid off the slides (Fig. 1h,i). Using a razor blade, slides were separated from one another and excess medium was removed from the sides of the slides (Fig. 1j). The medium layer was cut in half horizontally, and one half transferred to a second sterile slide with a razor blade (Fig. 1k). This created two culture slides with a ledge of medium for the aerial part of the seedlings to grow above, thus minimizing contact between seedlings and coverglass (Fig. 1l). Culture slides were stored in sealed sterile Petri dishes at 4°C for up to 1 wk. Sterilized seeds were sown on germination medium plates, synchronized and incubated under normal growth conditions at a 70° angle, to ensure that the hypocotyls and roots of individual seedlings were aligned along the same axis (Fig. 1m). Three 3 DAG seedlings were transferred from plates to c. 100 µl of moisturizing medium (half-strength MS salts, 0.5 mg l−1 Mes, 50 mg l−1 nystatin, pH 5.7) placed on a culture slide to prevent dehydration of the seedlings, and arranged such that their roots and half of their hypocotyls were in contact with the medium and the rest was above the ledge. Roots and hypocotyls were gently pressed into the medium surface to ensure that the seedling would not change position during the culture period and that the root would not dry out (Fig. 1n). Once the seedlings had been properly positioned, two spacers composed of four stacked 18 × 18 mm coverglasses (number 1.5, 0.17 mm thick; Fisher Scientific International Inc., Hampton, NH, USA; total thickness of the spacer c. 0.68 mm) were each affixed with 20 µl of sterile water at both ends of the culture slide (Fig. 1o). Spacers were overlaid with a 22 × 50 mm coverglass (number 1.5, 0.17 mm thick; Fisher Scientific), which remained in place for the duration of the experiment and created a culture chamber (Fig. 1p). The chamber was placed into a single groove of a glass staining dish for horizontal slides (Canemco Inc. & Marivac Inc., Canton de Gore, QC, Canada) that contained, on the bottom, four square layers of sterile paper towel soaked with 2 ml of moisturizing medium (Fig. 1q). Typically, only one chamber was placed in each dish in order to minimize possible damage due to fungal contamination, but up to five chambers could be efficiently cultured in each dish. The dish was closed with a glass cover, sealed with electrical tape to prevent moisture loss, and incubated under normal growth conditions (Fig. 1r). The procedure was repeated for each chamber as required for individual experiments. Seedlings were equilibrated in the chambers for c. 1 h before imaging, at which point a chamber was filled with moisturizing medium (Fig. 1s) and transported from the laminar-flow cabinet to the microscope in a sealed sterile Petri dish. Once imaged, the chamber was returned to the laminar-flow cabinet, and moisturizing medium was drained by capillary action from the chamber using a paper towel (Fig. 1t). The chamber was transferred back to the dish, and moisturizing liquid at the bottom of the dish was replenished as required. The dish was covered, sealed and incubated under normal growth conditions. This process was repeated for each chamber, for imaging intervals as described for individual experiments, and for a total maximum imaging duration of 108 h. Under these conditions, we observed decline or cessation of growth in 6/60 primordia, in which cases the primordia were excluded from further study.
Auxin application and transport inhibition
For auxin application, seeds were sown on germination medium and transferred 3 DAG to culture slides. IAA (Sigma) at 10% (w/v) in prewarmed vaseline (as lanolin fluoresced under our imaging conditions) was applied to one side of the first leaf primordium of each 4 DAG seedling as previously described (Scarpella et al., 2006). For auxin transport inhibition, seeds were sown on germination medium supplemented with 5 µm 1-N-naphthylphthalamic acid (NPA) (Chem Service Inc., West Chester, USA) and transferred 3 DAG to chambers with NPA-supplemented growth medium. Chambers were flooded with NPA-supplemented moisturizing medium before imaging.
Microtechniques and microscopy
Fixation and clearing were performed as previously described (Scarpella et al., 2004). Samples were viewed with a Planapochromat 10 × objective (numerical aperture, 0.45; working distance, 2.0 mm) and a Planapochromat 20 × objective (numerical aperture, 0.8; working distance, 0.55 mm) of an AxioImager.M1 microscope equipped with an AxioCam HR digital camera (Carl Zeiss, Oberkochen, Germany). Green fluorescent protein (GFP) was visualized with a 50% attenuated HBO103 mercury vapor short-arc lamp (Osram, München, Germany), with a BP470/40 excitation filter, an FT495 beam splitter and a BP525/50 emission filter (Carl Zeiss) and with camera exposure times of 0.5–1 s. Under these conditions, no physiological damage, mitotic arrest or cell death was observed (Dixit & Cyr, 2003). For confocal laser scanning (CLS) microscopy, dissected leaf primordia were mounted in water and observed with a Zeiss Axiovert 100M microscope equipped with a Zeiss LSM510 laser module confocal unit. GFP was excited with the 488 nm line of an Argon laser at 55% of output and 5% transmission, and emission detected with a BP500-530 filter. For simultaneous visualization of GFP and YFP, or of CFP and YFP, GFP or CFP was excited with the 458 nm line of an Argon laser at 55% of output and 30% transmission, and emission detected with a BP480-520 filter, while YFP was excited with the 514 nm line of an Argon laser at 55% of output and 5% transmission, and emission detected with a BP565-615 filter. Sequential excitation and collection of emission from individual fluorophores were performed in high-speed channel switching (multitrack) line scanning mode. Under these conditions, signal bleed-through of the different fluorophores across different photomultiplier channels was never observed.
Image analysis and processing
The spatial sampling rate of the CCD detector used to capture epifluorescence images was determined using a stage micrometer slide and found to be 2.81 ± 0.01 (n = 10) and 5.63 ± 0.01 (n = 10) pixels µm−1 for the 10 × and 20 × objectives, respectively. Because our analysis aimed at resolving vascular patterning at cellular resolution, the minimum resolution element in our study is one ground meristem cell, the average diameter of which is 5.88 ± 1.88 µm (n = 88). Signal-to-noise ratio (SNR) was increased after image acquisition by 2 × 2 pixel spatial averaging (Russ, 2002), which resulted in a spatial sampling rate of 4 and 8 pixels per resolution unit (i.e. 5.88 µm) for the 10 × and 20 × objectives, respectively. Because a sampling rate of 3–8 pixels per resolution unit collects the majority of the resolved information (Shaw, 2006), the increase in SNR from pixel averaging did not result in a significant loss of spatial resolution. Global and local background levels were estimated from the average pixel values of featureless regions of epifluorescence images and of image regions containing autofluorescence unrelated to the features of interest, respectively (Shaw, 2006) (Supplementary material, Fig. S1b,c). Signal was defined as pixel values exceeding local background by more than 2.7 standard variations (Colarusso & Spring, 2003) (Fig. S1d). To fully and readily convey information on fluorescence feature properties, spatially averaged epifluorescence images were turned into 8 bit gray-scaled images (Fig. S1a) and color-displayed with a purposely created look-up-table (LUT) in which black was used to encode global background, blue to encode local background, and cyan, green, yellow and white to encode increasing signal intensities (Fig. S1e,f). For CLS images, SNR was increased during image acquisition by four-frame temporal averaging (Russ, 2002) and postacquisition by neighborhood averaging using a 3 × 3 Gaussian kernel with a radius of 1 pixel (van Kempen et al., 1997; Russ, 2002). All images were analyzed and processed using ImageJ (National Institutes of Health, http://rsb.info.nih.gov/ij) and assembled and labeled using CanvasX (ACD Systems International Inc., Saanichton, Canada).
Live imaging of leaf development
To observe dynamics of procambium formation in live samples, we established a procedure to acquire visualization data from leaves expressing fluorescent reporters (Fig. 1). The culture system we designed is simple to assemble, the imaging technique we employed requires only standard, readily available equipment, and the method we developed is characterized by high reproducibility (Fig. 1; Fig. S1, Table S3) (see the Materials and Methods section for more details).
To identify fluorescent markers for live imaging of procambium formation, we screened 250 lines of a GFP enhancer-trap collection (Haseloff, 1999) for vein-associated expression in 5 DAG first leaf primordia, as their venation is predominantly preprocambial and procambial. We found that 13 lines satisfied the selection criterion, but only two of them, J1721 and Q0990, showed reproducible vein-specific GFP expression at stages critical for procambium formation (Fig. 2g,m; Fig. S2, Table S3).
Onset of GFP expression in line J1721 labeled preprocambial cells whose length was not significantly different from their width (Fig. 2h; Table S2), while GFP in line Q0990 was first expressed in cells that had acquired procambium-distinctive elongation along the axis of the developing vein (Fig. 2n; Table S2). Moreover, GFP expression level and length of GFP-expressing cells in Q0990 showed strong correlation (r = 0.98, n = 67, P < 10−4; Fig. S3). Throughout leaf development, onset of Q0990 expression always followed that of J1721 in veins of predictable positions (Figs 2, 3).
Time-lapse imaging of marker expression during undisturbed development
Previous studies suggested that time-lapse intervals of 12–24 h can identify salient features of procambium formation (Mattsson et al., 1999; Sieburth, 1999; Kang & Dengler, 2002, 2004; Scarpella et al., 2004, 2006). Therefore, images of J1721 and Q0990 first leaves were recorded every 12 h, for 84–108 h, starting at 3 DAG. We obtained 13 (J1721) and 12 (Q0990) sets of sequential images and found that they showed consistent results (Table S3). A representative sequence of selected time-points in the datasets for each marker line is shown in Fig. 3.
Formation of J1721 expression domains was observed to proceed progressively from pre-existing veins (Fig. 3a–o; Fig. S4). In particular, J1721 expression domains that connected pre-existing vasculature on both ends were observed to derive from the fusion of initially freely ending J1721 expression domains (Fig. 3i,j). Loop-forming J1721 expression domains progressed from the middle of the primordium towards its margin (Fig. 3c,d,i,k,l), except in three of 12 loops of the third pair, in which J1721 expression domains extended from the margin of the primordium towards its centre (Fig. 3m–o).
Appearance of Q0990 expression occurred simultaneously along the entire length of individual veins of all orders (Fig. 3p–y), except in four of 17 loops of the third pair, in which we distinctly observed the separate formation of marginal and radial domains of Q0990 expression (Fig. 3z–ad) were distinctly observed. Time-lapse analysis of fluorescence signal intensity showed that levels of Q0990 expression initially increased homogeneously within individual strands (Fig. S4). However, regions that showed peaks of Q0990 expression within individual domains became progressively more apparent over time. Maxima of Q0990 expression were randomly arranged within individual strands, and different domains showed different spatial distribution of peaks of Q0990 expression along their length.
Dynamics of marker expression in response to auxin application and transport inhibition
We next asked whether expression of J1721 and Q0990 remained associated with procambium development during experimentally challenged vein formation. Auxin supply and transport have been shown to define sites of vein formation in developing leaf primordia (Sachs, 1989; Mattsson et al., 1999, 2003; Sieburth, 1999; Scarpella et al., 2006). Therefore, we either applied the auxin IAA to developing leaf primordia or grew seedlings in the presence of the auxin transport inhibitor NPA and imaged J1721 and Q0990 expression over time.
We unilaterally applied IAA to 15 4 DAG first leaf primordia for each marker line. We imaged the primordia every 12 h, for 56 h, starting 8 h after the treatment. Seven (J1721) and eight (Q0990) of the treated primordia were excluded from the analysis, as at maturity they did not display auxin-specific vascular responses (Scarpella et al., 2006). The remaining primordia showed reproducible dynamics of auxin-induced marker expression (Table S3). Full-time series of images depicting successive time-points for each marker line are represented in Fig. 4. The dynamics of J1721 expression upon auxin application were remarkably similar to those observed under unperturbed conditions, with one exception. Domains of J1721 expression that gave rise to loop-like veins formed in response to auxin always developed from the margin of the primordium towards its middle (Fig. 4b–f), while during undisturbed development, loop-forming J1721 expression domains stereotypically extended in the opposite direction (Figs 3c–e, 4a,b). As during unchallenged vein formation, Q0990 expression appeared simultaneously throughout individual strands in response to auxin treatment, but formation of radial and marginal domains of Q0990 expression that generated auxin-induced loop-like veins were invariably temporally distinct events (Fig. 4g–l).
We followed marker expression over time in first leaf primordia of 15 (J1721) and 14 (Q0990) seedlings grown in the presence of NPA. Starting at 3 DAG, we acquired images every 12 h, for 60–108 h, and found that marker expression showed consistent behavior (Table S3), despite the large variations in position of vein formation within the sample population. A representative time-series for each marker line is shown in Fig. 5. Under conditions of reduced auxin transport, J1721 expression domains extended either from the margin of the primordium towards its middle or in the opposite direction, and loops could even result from the fusion of two initially freely ending J1721 expression domains (Fig. 5b–j). Further, appearance of Q0990 expression in all loop-forming veins occurred separately in radial and marginal domains. In each of these domains, however, Q0990 expression appeared simultaneously (Fig. 5m–x; Fig. S5).
In this study, we investigate dynamics of procambium formation events occurring during Arabidopsis leaf development. We provide a live-imaging method that allows noninvasive visualization of a variety of behaviors in developing leaves and is amenable to a range of experimental approaches that interfere with leaf development. Further, we supply a set of fluorescence markers that identify distinct and consecutive stages of procambium formation. Finally, by combining our live-imaging system with newly identified markers, we provide direct evidence that vein networks are formed by termination or fusion of free, progressively extending preprocambial strands and by simultaneous differentiation of procambium along the entire length of individual strands. Previous studies combining cell state-specific markers with statistical analysis provided indirect evidence for such a mechanism (Scarpella et al., 2004, 2006). However, the generality of those conclusions, and in particular their applicability to the development of higher-order veins, whose position and connectivity are not reproducible between different leaves, could not be evaluated in the absence of a live-imaging approach.
Early markers of leaf vein formation
The two vein-associated gene expression markers presented here, J1721 and Q0990, were selected from a larger collection (Fig. S2) as they identified consecutive stages preceding (J1721) or accompanying (Q0990) procambium differentiation. When characterized with respect to the previously identified markers of preprocambial cell selection and preprocambial cell identity, PIN1 and Athb8, respectively (Kang & Dengler, 2004; Scarpella et al., 2004, 2006), expression of these markers is invariably initiated in the temporal sequence: PIN1, Athb8/J1721, Q0990. All aspects of marker expression, including onset, intensity, relation to previously identified markers, and association with vein-forming cells under all experimental conditions, proved to be highly reproducible. Therefore, the expression of these markers can be used to readily visualize discrete and successive stages of procambium formation in developing leaves.
Progressiveness of preprocambium development
Onset of J1721 expression marks acquisition of preprocambial cell identity and, when judged by this marker, all veins share a common developmental mechanism. During the formation of all veins and under all tested conditions, expression of J1721 is initiated next to pre-existing vasculature and then extends progressively away from this point of origin. All veins thus arise as freely ending preprocambial branches. Furthermore, all connected veins are formed by fusion of free J1721-expressing preprocambial strands, and freely ending veins result from termination of the extension of J1721 preprocambial expression domains.
Preprocambial strands extended progressively under all conditions; however, the specific direction of this progression varied. In fact, while preprocambial strands in the first and second loop pairs invariably extended from central to marginal regions of the developing leaf, a fraction of third loop preprocambial strands formed in the opposite direction (i.e. marginal to central). Unlike the first two loop pairs, third loop pairs are associated with conspicuous auxin response maxima at the primordium margin (Aloni et al., 2003; Mattsson et al., 2003; Scarpella et al., 2006). This suggests that preprocambial strands are initiated at a critical auxin level, which, for most preprocambial loops, would be reached at the centre of the primordium, in the proximity of the midvein, that is, the point of convergence of auxin produced anywhere in the laminae. In contrast, auxin levels critical for preprocambial identity acquisition during third loop formation could be shifted to the margin of the primordium because of localized auxin synthesis at the hydathode (Cheng et al., 2006). This interpretation is further supported by our live imaging of vein formation under experimentally manipulated conditions. Impairing auxin flow towards the middle of the primordium resulted in both middle-to-margin and margin-to-middle polarities of preprocambial strand extension. Furthermore, direct auxin application at the primordium margin resulted in reversal of polarity of preprocambial strand formation, which then occurred exclusively from the margin to the middle of the primordium.
Simultaneity of procambium differentiation
A common developmental mechanism for all veins is not only suggested by J1721 expression, but also by Q0990-labeled procambium differentiation. In fact, our time-lapse observations suggest that Q0990 expression appears simultaneously along entire strands. Our observation intervals detected deviations from simultaneity at much later stages of vein formation (Fig. S4), but these may reflect unrelated differentiation events. Further, the dynamics of procambium differentiation, as marked by appearance of Q0990 expression, are in obvious contrast to the progressive extension of J1721 preprocambial expression domains.
Two mechanisms can be envisaged through which simultaneity of procambium differentiation could be achieved. In the first, preprocambial cells would simultaneously elongate in a coordinated fashion throughout the length of a preprocambial strand. In the second, preprocambial cells would acquire the procambium-distinctive narrow, elongated shape through a synchronized cell division parallel to the axis of the preprocambial strand. Our observations support the first of these two possible mechanisms. In fact, onset of Q0990 expression marks the earliest sign of elongation of preprocambial cells into procambium, and at its first appearance, Q0990 expression domains were invariably one-cell wide (Fig. 1m,n). Further, synchronized cell divisions occurring throughout the entire preprocambial strand should be revealed by simultaneous expression of mitotic genes along developing veins. However, careful and detailed examination of CycB1;1 expression during leaf vein formation failed to detect such a pattern (Donnelly et al., 1999; Kang & Dengler, 2002). Therefore, we conclude that each individual preprocambial strand differentiates into procambium through coordinated cell elongation occurring along the entire length of the strand.
Whereas procambium differentiated simultaneously throughout the first two loop pairs, we observed separate appearance of radial and marginal procambial strands in a portion of third loops. As formation of third loop pairs is associated with increased auxin levels at the hydathode (Aloni et al., 2003; Mattsson et al., 2003; Cheng et al., 2006; Scarpella et al., 2006), our findings suggest that excess auxin prevents simultaneous procambium differentiation in loop development. This possibility is also supported by the observation that separate appearance of radial and marginal strands occurred in all loop-like veins formed in response to auxin application. Moreover, when auxin levels were raised in the primordium because of reduced auxin drainage, we observed separate differentiation of radial and marginal strands in all loops, including the typically simultaneously differentiating first two loop pairs. Therefore, given enough auxin at crucial stages of development, formation of all procambial loops occurs in temporally distinct steps. Increased auxin levels could lead to deviations in simultaneity of procambial loop differentiation by delaying the transition from source to sink of incipient veins (e.g. the radial vein of the third loop), which prevents the formation of a connection (e.g. the marginal vein of the third loop) with pre-existing veins (e.g. the second loop) (Sachs, 1968). The smooth procambium formation in first and second loop pairs observed under unchallenged conditions would thus simply reflect efficient auxin flow and/or inconspicuous auxin synthesis in early primordium development.
We thank Jian Xu, Ben Scheres, Taku Demura, David Galbraith, the Arabidopsis Biological Resource Center and Cold Spring Harbor Laboratory for generously providing seeds and plasmids; Janice Cooke, Mike Deyholos, Dave Pilgrim and Andrew Waskiewicz for kindly allowing the use of their equipment; and Thomas Berleth and Naden Krogan for invaluable comments on the manuscript. This work was supported by a Discovery Grant of the Natural Sciences and Engineering Research Council of Canada (NSERC), by an Alberta Ingenuity (AI) New Faculty Grant and by the Canada Research Chairs Program. TJD was supported by an NSERC USRA, an NSERC CGS-M Scholarship, and an AI Student Scholarship.