Ординатура / Офтальмология / Английские материалы / The Retina and its Disorders_Besharse, Bok_2011

Proportion correct

1

0.75

0.5

0.25

Contrast

75% correct at approximately 2.5% contrast. The slope (s) can be used to infer how easily nearby contrasts can be discriminated from one another – a shallow slope means that a large contrast difference is required to achieve a given change in performance, whereas a steep slope means that a small change in the stimulus produces a large change in performance.

Figure 2 Contrast detection. (a) Example of a four-alternative forced-choice (4 AFC) task. The observer is required to fixate the central dot and to indicate whether the target appeared top left, top right, bottom left, or bottom right. The target contrast is adjusted by computer to a level that produces 75% correct detection. (b) A typical psychometric function. Circles show the proportion of trials the target was detected (ordinate) as a function of the target contrast (abscissa). Error bars show 1 standard deviation. The curve shows the best-fitting cumulative normal function, from which the interpolated 75% correct point is taken as contrast-detection threshold.

case the subject is forced to guess. Notice that when guessing, the subject is still correct sometimes (25% if there are four alternatives, 33% if there are three, or 50% if there are two, etc.), as shown in the frequency of seeing curve in Figure 2(b), where, at low contrasts, performance is 25% correct. The data have been fit with a curve known as a psychometric function, in this case a cumulative Gaussian:

Y ¼ g þ ð1 gÞ erf ðz=sqrtð2ÞÞ=2

where z ¼ (X – m)/s; g is the guess rate (0.25 in a 4AFC experiment). The mid-point (m) of the psychometric function is often taken as Cthresh – for a 2AFC task, this is 75% correct. In the example shown, the observer achieved

Spatial Frequency Channels

Based on behavioral observations in humans and single unit recordings in mammalian visual systems, researchers discovered around half a century ago that the visual system analyses images at a series of relatively narrow spatial scales and orientations known as channels. Thus, fine and coarse image details are encoded separately and Fourier analysis can be used to study the image structure that is encoded by different visual processing channels. Fourier analysis computes the sum of basic sine waves whose linear sum produces the image. To illustrate the representations of an image that are available at different spatial scales, Figure 3(a) shows a typical image, together with its coarse (Figure 3(b)) and fine (Figure 3(c)) spatial structure.

Visually responsive neurons in primary visual cortex, the first cortical projection from the retina through the lateral geniculate nucleus of the thalamus, respond to images only within a limited area of the visual field, known as the classical receptive field, and are selective for a limited range of spatial frequencies and orientations. These receptive fields are now routinely modeled as Gabor or wavelet functions, defined as:

l represents the wavelength, y the orientation, and c the phase of the sine-wave component. For the Gaussian window, s is the standard deviation and g is the spatial aspect ratio. Examples of Gabors are illustrated in Figure 4. On the top row, spatial frequency increases from left to right and all Gabors are of the same orientation 0 and contrast. On the bottom row, spatial frequency is fixed, but orientation is 45 , 90 , or 135 (from left to right). The visual system encodes image structure with a bank of such wavelet filters that represent the retinal image through patchwise local analysis.

Figure 5 provides compelling demonstrations that our visual system employs a set of spatial frequency and orientation-selective channels. These demonstrations show that after prolonged viewing of a particular pattern (termed adaptation) the appearance of other patterns can be altered (termed an aftereffect). In these demonstrations, adapting to a pattern of one spatial frequency or orientation produces a loss in sensitivity in the channel that responds most to that pattern, but little change in channels tuned to other spatial frequencies or orientations. This localized loss in sensitivity produces a relative shift in the responses of our visual channels that cause us to experience changes in the appearance of the image.

These observations have led to the widespread use of sine-wave grating patterns in basic and clinical vision research. In order to derive a measure of vision that reflects the sensitivity across our set of visual channels and to reflect the fact that functional vision requires us to detect and interact with objects of various sizes, contrastdetection thresholds are measured for gratings of a range of bar widths, expressed as spatial frequency or the number of grating cycles per unit distance. Figure 4 illustrates Gabors of differing spatial frequency; however, the size of one grating cycle on the retina depends on the distance

Figure 4 Gabor (wavelets) of differing spatial frequency and orientation. Top row: spatial frequency increases from left to right, orientation is fixed at 0 . Bottom row: Orientation increases from left to right: 45 , 90 , and 135 , spatial frequency is fixed.

from which it is viewed. Therefore, image sizes are usually calculated in terms of visual angle, which specifies the retinal image size. Figure 6 shows how visual angle is calculated and its relationship to image size and viewing distance. A convenient rule is that 1 cm viewed from 57 cm subtends a visual angle of 1 and roughly corresponds to a finger nail viewed at arm’s length.

Contrast Sensitivity Function

Many researchers have shown that for sine-grating patterns, Cthresh strongly depends on spatial frequency. This fundamental observation is demonstrated in the classic image shown in Figure 7. Spatial frequency increases from left to right and contrast increases from top to bottom, so that contrast is constant across any horizontal line. Contrast-detection thresholds can be visualized on this figure as the imaginary curve along

Figure 5 Demonstration of spatial frequencyand orientationselective aftereffects. First note that when you fixate the centre gray dot, the gratings in the middle row are of the same spatial frequency and orientation. Next, look back and forth between the black dots in the top row for around 10 s. Now, when you look at the center gray dot, the grating on the left appears to be of higher spatial frequency than the grating on the right. Next, look back and forth between the white dots in the bottom row for around 10 s. Now, when you look at the center gray dot, the grating on the left appears tilted counterclockwise, while the grating on the right appears tilted clockwise. These aftereffects are robust even though you know that the gratings in the middle row are the same. These demonstrations provide compelling evidence that visual processing involves channels that are narrowly tuned for spatial frequency and orientation.

which the grating changes from invisible (toward the top of the figure) to visible (toward the bottom of the figure). Most people report that the function peaks somewhere near the middle of the figure. Notice that the peak shifts as you move the figure closer or further away. This demonstrates the importance of visual angle rather than physical image size. Note that the highest spatial frequency that can be detected at maximum contrast is given by the rightmost point on a contrast sensitivity function (CSF). This is referred to as the resolution limit and is a quick and convenient method of assessing visual sensitivity than measuring the entire CSF.

When measured with forced-choice procedures, (Figure 2) contrast-detection thresholds are lowest for

gratings around 2–5 cycles per degree of visual angle (c deg–1). By convention, the inverse of Cthresh (1/Cthresh) is usually reported and is termed contrast sensitivity. The rationale for the use of contrast sensitivity over contrastdetection threshold is most likely because the shape of the CSF is the same as that of the underlying modulation transfer function of the system. The circles in Figure 8 show the author’s contrast sensitivity as a function of spatial frequency measured with a forced-choice procedure. Error bars show 95% confidence intervals. The data have been fit (green curve) with the outputs of a set of spatial frequency channels shown by the colored curves. The channels are log spaced in spatial frequency (with peaks at 0.5, 1, 2, 4, 8, 16, or 32 c deg–1) and have the same bandwidth (1.4 octaves). The summed outputs of the set of filters provide a good fit to the data and this channel-based system is now a widely accepted model of early visual processing.

The spatial frequency aftereffect shown in Figure 5 is easily explained with this channel-based model. Adapting to one spatial frequency reduces the responses of the channel that is most sensitive to that spatial frequency, but has little effect on the responses of other channels. When a different spatial frequency is subsequently viewed, the overall activity across the channels is shifted away from the adapted channel. This shift in the population response produces a shift in apparent spatial frequency away from the adapting frequency. An analogous model explains the shifts in orientation in the lower row of Figure 5, except that orientation-selective channels are adapted rather than spatial-frequency-selective channels.

The CSF is highly dependent on the mean luminance of the display on which it is measured. This can easily be experienced by viewing Figure 7 with a pair of dark sunglasses (possibly two pairs), which moves the curve down (reducing sensitivity) and shifts the peak to lower spatial frequencies. The data in Figure 8 were collected on a standard computer monitor that has a mean luminance of 50 cd m–2 (candelas per square meter). Photopic, mesopic, and scotopic vision and changes in visual performance show that sensitivity to high spatial frequencies increases with mean luminance. This property is important because CSFs are routinely measured on relatively dim displays (e.g., 50–100 cd m–2) in the laboratory and in the clinic; however, the luminance of the real world is typically much greater. For example, the luminance of a cloudy sky is around 35 000 cd m–2, suggesting that standard experimental conditions may underestimate sensitivity to fine spatial structure.

Temporal Contrast Sensitivity

In addition to a dependence on spatial frequency, contrast sensitivity also depends strongly on temporal frequency. Figure 1 illustrates spatial variation in luminance, but

Figure 7 Illustration of the contrast sensitivity function (CSF). Spatial frequency increases from left to right, contrast increases from top to bottom. The contrast along any horizontal line is fixed. Different spatial frequencies become visible at different contrasts and define an imaginary curve that separates seen from unseen structure. Notice that if you move the image closer to your eye, the peak moves to the right and if you move it further away, the peak moves to the left. This demonstrates that contrast sensitivity depends on retinal not physical image size. If you wear one or two pairs of dark sunglasses, the curve shifts down and the peaks moves left, which demonstrates the dependence of the CSF on mean luminance.

imagine instead that the x-axis represents time, rather than space. Now the figure illustrates flicker. Flicker frequency can be varied in the same way as spatial frequency is varied in Figures 3 and 7. The circles in Figure 9 show how the author’s contrast sensitivity varies as a function of temporal frequency for a 2 c deg 1 grating pattern. Sensitivity peaks around 5 Hz, at the mean luminance

200

150

Sensitivity 100

50

Spatial frequency (c deg–1)

Figure 8 Spatial contrast sensitivity. Circles show contrast sensitivity (the reciprocal of contrast-detection threshold) for sine gratings of a range of spatial frequencies. Sensitivity peaks at around 2 c deg–1 under the conditions employed here and decreases at lower or higher spatial frequencies. The black curve is the summed sensitivity of the set of log-scaled channels shown by the colored curves and provides a good fit to the data.

used here (50 cd m 2) and decreases at lower or higher temporal frequencies. These data are well fit (black curve) by a model with only two temporal channels, compared with the multiple channels that support spatial contrast sensitivity. One channel (red curve) is low-pass or sustained and is most sensitive to structure that is stationary or slowly changing over time. The second channel (blue curve) is band-pass or transient and is most sensitive to structure that changes at around 5 Hz.

The spatial resolution limit falls steadily with distance from the fovea, an effect that can be experienced by viewing Figure 7 while fixating away from the center of the image. As you fixate further away, the threshold curve moves further down the figure and its peak shifts further to the left. Unlike spatial resolution, temporal resolution (the highest flicker rate that can be detected at any contrast, often called critical flicker fusion frequency)

Temporal frequency (Hz)

Figure 9 Temporal contrast sensitivity. Circles show contrast sensitivity (the reciprocal of contrast-detection threshold) for sine gratings of a range of temporal frequencies. The black curve is the summed sensitivity of the two log-scaled channels shown by the red and blue curves. The red curve has peak sensitivity at low temporal frequencies – that is, static images – and is termed a sustained channel. The blue curve has peak sensitivity around 6 Hz and is termed a transient channel.

increases moderately with distance from the fovea. This explains why older, 60-Hz computer displays can sometimes be seen to flicker when seen in the peripheral visual field, but not when viewed directly. Just as spatial contrast sensitivity depends on luminance, so does temporal contrast sensitivity. A 35-mm film is generally recorded at 24 frames per second, a refresh rate that could be easily detected at moderate light levels, as can be seen from Figure 9. For this reason, movie theaters are generally dark because sensitivity to high flicker rates is poor under those conditions. In addition, the visible 24-Hz image update rate is masked by flashing the illuminant at 48 Hz, so each frame is flashed twice.

At supra-threshold contrasts, apparent contrast is relatively independent of spatial or temporal frequency, a phenomenon termed contrast constancy.Contrast constancy can be experienced in Figure 7 – while the transition between visible and invisible gratings has a curved

shape, toward the bottom of the figure, the gratings appear to have similar contrast regardless of spatial frequency. This has important implications for image enhancement, which should therefore target only image components that are below their Cthresh.

See also: Acuity; Anatomically Separate Rod and Cone Signaling Pathways; Chromatic Function of the Cones; Information Processing: Contrast Sensitivity; Information Processing: Direction Sensitivity; Information Processing in the Retina; Information Processing: Retinal Adaptation; Photopic, Mesopic and Scotopic Vision and Changes in Visual Performance.

Further Reading

Bracewell, R. (1999). The Fourier Transform and Its Applications, 3rd edn. London: McGraw-Hill.

Campbell, F. W. and Robson, J. G. (1968). Application of Fourier analysis to the visibility of gratings. Journal of Physiology 197: 551–566.

Field, D. J. and Tolhurst, D. J. (1986). The structure and symmetry of simple-cell receptive-field profiles in the cat’s visual cortex.

Proceedings of the Royal Society of London. Series B. Biological Sciences 228(1253): 379–400.

Georgeson, M. A. (1990). Over the limit: Encoding contrast above threshold in human vision. In: Kulikowski, J. J. (ed.) Limits of Vision, pp. 106–119. London: Erlbaum.

Hubel, D. H. and Wiesel, T. N. (1959). Receptive fields of single neurones in the cat’s striate cortex. Journal of Physiology 148: 574–591.

Kelly, D. H. (1961). Visual responses to time-dependent stimuli. 1. Amplitude sensitivity measurements. Journal of the Optical Society of America. A, Optics, Image Science, and Vision 51: 422–429.

Kulikowski, J. J. and Tolhurst, D. J. (1973). Psychophysical evidence for sustained and transient detectors in human vision. Journal of Physiology 232(1): 149–162.

Landis, C. (1954). Determinants of the critical flicker-fusion threshold.

Physiological Reviews 34(2): 259–286.

O’Shea, R. P. (1991). Thumb’s rule tested: Visual angle of thumb’s width is about 2 deg. Perception 20(3): 415–418.

Rovamo, J., Virsu, V., Laurinen, P., and Hyvarinen, L. (1982). Resolution of gratings oriented along and across meridians in peripheral vision.

Investigative Ophthalmology and Visual Science 23: 666–670.

Glossary

Cell birth – For the purposes of this article, cell birth is the activation of a new transcriptional program that is never seen in retinal progenitor cells (RPCs), but is limited to differentiating retinal transitional cell (RTCs). This definition excludes cell-cycle exit because, as discussed in this article, differentiating RTCs can be generated independent of exit. In normal cells, the timing of cell birth is usually measured by briefly labeling replicating DNA with bromodeoxyuridine (BrdU) or tritiated thymidine (3H-thymidine) and then, days/weeks later, counting intensely labeled cells which, by definition, exited the cell cycle soon after the label was introduced. This tool is useless in mutants where cell birth occurs in the absence of cell-cycle exit. In that case, birth must be noted when it occurs using markers induced in differentiating RTCs that are never present in RPCs (e.g., Nrl in newborn rods).

Cell-cycle rate – It reflects the rate of cell expansion, which correlates with cell-cycle length. Competence – This activity influences the ability of RPCs to generate certain cell types. It is permissive, but not instructive in that it creates potential, but it is not sufficient to force cell birth on its own. The competence of RPCs changes throughout development. Early RPCs are competent to produce ganglion, horizontal, amacrine, and cone cells, while late RPCs produce bipolar and Mu¨ller cells. Rods are generated throughout, but predominantly are a mid–late-born cell type.

Exit – Normally, cell birth is closely coupled to cellcycle exit and ensures that a newborn RTC remains post mitotic until and after it becomes a terminally differentiated cell.

G0 – Cells can be driven out of G1 into reversible G0 by serum starvation or contact inhibition, or into irreversible G0 by terminal differentiation or senescence.

G1, S, G2, and M – Most cell cycles have four phases: S phase is when DNA is synthesized;

M phase is when cells undergo mitosis to generate two offspring (often termed daughters); G1 and G2 are gaps between M and S or S and M, respectively.

Interkinetic nuclear migration (INM) – Cellular processes connect the RPC to the apical (also called outer or ventricular) and basal (also called inner or

vitreal) surfaces of the retina. RPC nuclei move as they traverse the cell cycle. M-phase nuclei are located at the apical surface, then move more basally to go through G1 and start S, then finish S migrating back to the apical side, and go through G2 just before they reach the end of that journey.

Mitogen – Extracellular factor that induces cell division.

R – The restriction point in G1, beyond which a cell will continue into S phase even if mitogens are withdrawn.

Retinal progenitor cell (RPC) – A dividing cell in the normal developing retina.

Retinal transition cell (RTC) – Also termed precursor. These are newborn differentiating cells. They are distinguished from RPCs by a novel transcriptome that includes many mRNAs/proteins that are never detected in RPCs. Normally, RTCs are post mitotic. However, failure to exit does not prevent induction of the RTC transcriptome. Birth (induction of the differentiation program) and exit can be uncoupled and are, therefore, viewed as separate in this article.

Introduction

Cell division is driven by extrinsic mitogens that influence the intrinsic core cell-cycle machinery. Extrinsic factors can also influence the decision to exit and differentiate during development. In the retina, the earliest born cells are ganglion neurons (Figure 1) and these cells secrete factors that have crucial roles in influencing division and fate, such as Sonic Hedgehog. Beyond this stage, however, although many factors can influence division and fate, it is not clear whether they actually do so in vivo. Dissociated, well-separated, individual retinal progenitor cells (RPCs) can generate clones of cells that resemble clones in vivo in both size and composition, except for ganglion cells which do not survive in vitro. These data suggest that apart from ganglion cell births, much of retinal development is intrinsically programmed and thus, apart from a general requirement for mitogens, extrinsic factors may not play a major role in determining whether RPCs divide or exit. Further, we have introduced a few of the basic issues around cell-cycle regulation, discuss the role of some

Peak of cell birth

Cell types

Cell cycle length

Type of

division

key cell-cycle regulators in retinal development, and finish with a discussion as to how RPC expansion versus neurogenesis is coordinated through the Notch pathway.

A Few Basics of Cell-Cycle Regulation

Extracellular positive (mitogenic) and inhibitory cues are sensed in cells through fluctuating levels of cyclins. These proteins are required to activate cyclin-dependent kinases (Cdks). Cyclin D-Cdk4/6 complexes form in G1, cyclin E-Cdk2 complexes act at G1-S, cyclin A-Cdk2 complexes drive S, and cyclin B-Cdk1 complexes drive M. However, these complexes may act in other phases and show considerable functional redundancy. Treatment of resting cells with mitogens induces Cyclin D1 transcription through Ras–Raf–Erk-mediated activation of the immediate early gene Ap1 (Ap1/Ets). Glycogen synthase kinase 3 beta (Gsk3b)- mediated Thr phosphorylation of cyclin D reduces its stability and nuclear localization. Mitogens also activate the Ras–Pi3k–Akt pathway which phosphorylates and inactivates Gsk3b, doubling cyclin D1 half-life and promoting nuclear translocation. Cyclin D-Cdk4/6 complexes phosphorylate retinoblastoma (Rb) proteins on multiple sites. The cyclin D-Cdk4/6 complex also titrates the Cip/Kip family of Cdk2 inhibitors (CKI), discussed further below.

Rb inhibits division in two major ways (Figure 2). First, it binds the activating E2f transcription factors (E2f1, 2, and 3a). These are inducer genes that positively regulate the cell cycle (e.g., cyclins E and A) and other genes that are necessary for the nuts and bolts of DNA replication (e.g., PCNA, RRM1). Rb binds and quenches E2f activity, and recruits silencing cofactors to permanently shut down genes in terminally differentiating or senescent cells. Rb also binds E2f3b and E2f4, whereas its relatives p107 (Rbl1) and p130 (Rbl2) preferentially bind E2f4 and 5 – which are thought to be repressive E2fs that primarily mediate inactivation of target genes (although E2f3b can mimic some functions of E2f3a). However, p107/p130 can partner with activating E2fs if E2f4 is missing. E2fs 6–8 do not bind the Rb family of pocket proteins, but inhibit transcription by recruiting other corepressors. The extent to which E2f target-gene induction involves direct activation (mediated by activator E2fs) versus derepression (loss of Rb pocket proteins from repressor E2fs) is not completely resolved.

Second, Rb – but not p107/p130 – binds the Cdh1 subunit of anaphase-promoting complex or cyclosome (APC/C) (Figure 2). APC/C ubiquitin ligase degrades securin and cyclins to permit passage through and escape from M phase, but in G1 it degrades Skp2 – part of another E3 ubiquitin ligase (SCFSkp2) that degrades Cip/Kip CKIs to promote Cdk2 activity. Rb binds Skp2,

presenting it for destruction to APC/C, thus preventing degradation of Cip/Kip CKIs and blocking Cdk2 action.

Rb phosphorylation weakens binding to E2f, resulting in induction of cyclin E, Cdk2 activation, further Rb phosphorylation, and induction of cyclin A – a sequential process that is necessary for cell-cycle progression. Cyclin D function is dispensable in Rb-deficient cells, but cyclin E is required for division even in the absence of Rb, indicating it has other targets. Indeed, independent of Cdk2 activation, cyclin E promotes loading of MCM proteins onto origins of replication, and E-Cdk2 phosphorylates this complex to trigger DNA replication. Cyclin E overexpression can drive the cell cycle even when E2f activity is blocked. Apart from unleasing E2f, Rb phosphorylation also releases it from both Skp2 and Cdh1 – thus activating Cdk2 by a second route (Figure 2). This positive-feedback loop allows cells to pass R and enter S; indeed, E2f1 behaves as a bistable switch to drive this irreversible transition.

Distinct CKI families bind and inhibit cyclin-Cdks (Figure 2). The Ink4 family, which includes p16Ink4a, p15Ink4b, p18Ink4c, and p19Ink4d (encoded by Cdkn2a/b/c/d,

respectively), inhibits Cdk4/6. Ink4 CKIs act upstream of Rb-E2f and need Rb plus p107 or p130, and E2f4 or

E2f5 to block division. The Cip/Kip family of CKIs, which includes p21Cip1, p27Kip1, and p57Kip2 (encoded by Cdkn1a/

b/c, respectively), inhibit cyclin A/E-Cdk2 and cyclin B-Cdk1 and can act downstream of Rb. Notably, p107

and p130, but not Rb, also function as CKIs that inhibit Cdk2 activity as potently as p21Cip1 (Figure 2). In the

complete absence of the Rb family, mouse embryo fibroblasts (MEFs) lack a G1 restriction point, and progress even if mitogens are withdrawn, but these cells arrest at G2 due to the combined action of Cip/Kip CKIs and p53 tumor suppressor.

Inhibitory mitogens – such as transforming growth factor-beta (TGFb), block cyclin D induction or inhibit its activity by inducing the expression of Ink4 CKIs. TGFb also inhibits cell-cycle progression by triggering nuclear translocation of an E2F4/E2f5–p107–Smad3 complex that associates with Smad4 protein, and then binds and silences the c-Myc promoter through a Smad-E2f element.

In summary (Figure 2), Rb and Cip/Kip CKIs cooperatively inhibit division by constraining E2fand Cdk2mediated induction of S-phase gene transcription and replication origin firing, respectively. Rb and p27Kip1 cross-talk positively by promoting Skp2 degradation and blocking Rb phosphorylation, respectively. Mitogenactivated cyclin-Cdks sequentially phosphorylate Rb, cyclin D-Cdk4/6 sequesters Cip/Kip CKIs, and Skp2, freed of Rb, stimulates CKI degradation, all of which leads to activation of E2f and Cdk2. E2f and Cdk2 cross-talk positively by inducing cyclins and phosphorylating Rb, respectively, This dual axis triggers the production and/ or activation of the components needed for DNA replication.

Cyclins and Cdks in Retinal Development

RPCs express higher levels of cyclin D1 than any other embryonic tissue (Figure 3). D1 absence causes severe retinal hypocellularity. Downregulation/inhibition of D1 during differentiation is important since ectopic expression in differentiating photoreceptors prevents normal cellcycle exit, mimicking the effect of pocket protein loss. The large induction in p27Kip1 protein translation in newborn retinal neurons likely dwarfs any remaining D1Cdk4/6 in these cells. D1 loss reduces RPC division beyond E16.5 but not earlier, suggesting that the earliest phase of RPC expansion is D1-independent. This delayed

requirement for D1 may reflect the gradual increase in Rb and p107 expression in RPCs during development. Consistent with the role of D cyclins in inactivating Rb and sequestering CKIs (see above), the D1-null retina has hypophosphorylated Rb, no cyclin E–Cdk2 activity, and hypocellularity is rescued when p27Kip1 is also missing. A cyclin D1KE point mutant binds but fails to activate

Cdk4/6 and partially rescues the D1-null retina. Like normal D1-Cdk, D1KE-Cdk complexes sequester p27Kip1, and

consequently both Cdk2 activity and Rb phosphorylation are increased in D1KE versus D1-null retinas.

The D family has two other members – D2 and D3. The defect in D1-null mice is rescued in mice expressing

D2 from the D1 locus, so they are interchangeable in this regard. Normally, however, D2 is not expressed in the retina and while D3 is expressed in Mu¨ller glia, it is present at very low levels in RPCs. There is no induction of D2 or D3 in D1-null retinas and D1/D2or D1/D3-null

retinas do not show a more severe phenotype than the D1 knockout (KO). The loss of p27Kip1 in D1-null retinas also

does not affect D2 and D3 expression. Triple-null mice survive to E15 and, consistent with the idea that D cyclins are not required for early RPC division the early triple-null retina also appears wild-type (WT).

The cyclin E family consists of E1 (formerly E) and E2, and both are expressed in retina. Apart from a spermatogenesis defect in E2-null males, E1and E2-null mice are normal. E1 / ;E2 / mice die around mid-gestation ( E10) due to defective endoreduplication (repeated S with no M-phase) in trophoblasts, and thus a defective placenta. By using tetraploid WT blastocysts, which form a normal placenta but do not contribute to the embryo proper – E1 / ;E2 / embryonic stem (ES) cells injected into these blastocysts could, in about half the cases, generate a normal embryo that, like WT ES cells in this approach, survive to birth. Half of E1 / ;E2 / embryos die with cardiac defects, all megakaryocytes (like trophoblasts) show defective endoreduplication, and while MEFs could divide normally, they were unable to exit quiescence. Importantly, no neural phenotypes were reported in E1 / ;E2 / embryos, suggesting that cyclin E functions are redundant in RPCs, perhaps due to the combined actions of cyclins D1 and A.

Remarkably, knocking in cyclin E1 to the D1 locus partially rescues the cyclin D1-null retinal phenotype with retinal thickness reaching 75% WT. Rb phosphorylation is only slightly higher than in the D1 KO, suggesting that low levels of Rb phosphorylation are sufficient to permit RPC division. Indeed, cyclin E can override a cellcycle block induced by overexpressing a mutated version of Rb lacking most of its phosphorylation sites. The Cdkindependent role of cyclin E in loading DNA-replication origins may be important in this context.

Like cyclins, there is also considerable redundancy among Cdks. Although Cdk4 / mice are 20% smaller

at birth, show pancreatic islet cell hypotrophy, and null MEFs divide more slowly due to elevated p27Kip1, there

are no obvious retinal defects. Apart from a mild hematopoietic defect Cdk6 / mice are normal, and Cdk4/6 redundancy in RPCs remains to be addressed since dou- ble-null mice die after E14.5. Cdk3 is rarely mentioned as it is defective in many mice and is thus dispensable. Cdk2null mice appear normal except for a defect in meiosis, and deleting Cdk2 does not affect fibroblast division. Cdk2/ 4/6 triple-null mice die around E12.5 and retinal explant studies have not been attempted. The latter study also shows that Cdk1 (also called Cdc2) – the most ancient cell-cycle kinase best known for its role at G2/M in

mammals – can substitute for other kinases to mediate much of the proliferation needed for embryogenesis. Moreover, it is required for the first division following fertilization. A conditional allele will be needed to study its role in later stages, including RPC expansion.

In summary, cyclin D1 promotes RPC division both by phosphorylating Rb proteins and sequesterating p27Kip1, although early RPC expansion appears cyclin D-indepen- dent. E cyclins, Cdk4/6, or Cdk2 each seem redundant, but further studies are required to tease apart overlapping cyclin and Cdk roles and to determine if Cdk1 is essential for RPC division.

The Rb Family in Retinal Development

Rb is the tumor suppressor mutated in the familial cancer retinoblastoma. Rb and p107 pocket proteins are present in mouse and human RPCs as well as in postmitotic retinal transition cells (RTCs), whereas p130 seems to be confined to the latter. As expected, Rb loss triggers extra division. However, Rb and p107 seem to be inactive in RPCs as removing Rb or both Rb and p107 in mouse retina does not affect the number of M-phase or visual system homeobox 2 (Vsx2+, also called Chx10) RPCs. Inactivation of Rb in RPCs may be due to the extremely high levels of cyclin D1 and, as noted above, Rb is hypophosphorylated in D1-null retinas. The logical explanation for Rb expression in RPCs may be that it is poised to act rapidly in newborn RTCs once D1 levels drop (Figure 3). Rb is also important for arresting division following DNA damage, which could also be employed in RPCs. Nevertheless, the irrelevance of Rb in controlling normal RPC division is a hard pill to swallow in view of multiple in vitro overexpression experiments implying that Rb tempers the expansion of all dividing cells. Yet Rb-null ES cells divide normally, and Rb-null embryos are not enlarged despite minimal apoptosis, and the same is true of p107/p130-null embryos, and there is not much compensation in either case. Moreover, Rb loss does not impact much of Xenopus embryonic development, again implying that it is already inactive.

Instead of tempering expansion, Rb is employed mainly to promote permanent cell-cycle exit, such as in terminally differentiating (Figures 3 and 4) or senescing cells, or to execute arrest in DNA-damaged cells. Thus, in contrast to the undetectable effect of Rb loss in RPCs, there is a dramatic effect in differentiating RTCs (Figure 4). RTCs missing Rb, or Rb and one or more of its relatives, divide ectopically. Some cell types choose apoptosis to defend the tissue from cancer, which is E2f1-dependent, but p53-independent. Indeed, most defects in the Rb-null retina are rescued in the Rb/E2f1-null retina, except for a notable cell-cycle and cell-death-independent differentiation defect in a subset of amacrine interneurons that is caused by E2f3a. Intriguingly, Rb repression of E2f3 is also