Monitoring of switches in heterochromatin-induced silencing shows incomplete establishment and developmental instabilities

We reasoned that if the prevailing model for PEV is correct (65–70), then every S-phase represents an independent potential loss of silencing, and so losses of silencing in any lineage would be independent of losses in any other lineage. We tested this prediction by analyzing the variegation in both halves of individuals from recently isogenized strains of 3 variegating white⁺ reporter genes: white^mottled−4, white^mottled−4h, and Y,10C. The first 2 are chromosomes bearing inversions that juxtapose the natural white enhancer/promoter with centric heterochromatin of the X chromosome. Y,10C is a Y-linked P{white⁺}, which possesses a white⁺ transgene controlled by the minimal heat-shock-70 gene promoter (w^hs) (71, 72) and a GFP gene expressed by a ubiquitin promoter (73), both of which variegate (54). We quantified expression categorically (as in ref. 63) and found a strong correlation in expressivity of white⁺ between left and right eyes of the same individuals from all 3 strains (Fig. 2 A–C; see figure legend for statistical analyses). This correlation was clear even when the flies were raised at high (29 °C) or low (12 °C) temperatures, which suppress and enhance variegation, respectively, or were raised on medium containing a chemical suppressor of variegation (74) (Fig. 2 D–F). Rare extreme exceptions to the preponderant range of white⁺ expression were additionally informative. Eyes with almost no expression (strongly repressed, in some cases merely single ommatidia expressing white⁺) were always (n = 10 of 10) found on a fly head opposite a second such rare eye. We confirmed that these exceptions were not the consequence of cryptic polymorphisms by collecting low-expressing exceptions, flies with medial ranges of expression, and high-expressing flies, and mating them pairwise. Even in the second subsequent generation, these 3 subpopulations each recapitulated normal and statistically indistinguishable variation in expression (Fig. 2G) (H₀ states that these 3 populations of pigment categories derive from 1 population, χ² = 7.15, df = 8, P > 0.5). We conclude that the variation in expression is not due to spontaneous second-site enhancers or suppressors, either dominant or recessive.

This correlation between left and right indicates that a common factor controls expressivity in both halves of the organism. Syncytial nuclei, and tissues separated by morphogenetic movements after cellularization, generally do not cross the left–right division plane; thus, whatever factor correlates PEV expressivity in adults acts prior to the separation of the presumptive eye anlagen in embryos. Analysis of patterns of X chromosome loss in gynandromorphs (75–77) indicates that the nuclear division of the embryo that separates left and right imaginal cuticular anlagen is usually the very first zygotic division. Thus, we interpret the correlations between the left and right eyes of adults to reveal a common silencing potential between the first 2 zygotic nuclei. This occurs well before cellularization, before germ-layer or placodal determination, before cytological condensation or heterochromatin formation in nuclear division cycle 14 (78–82), before the activation of the hsp70 (58) or white⁺ promoters (59, 73), before the onset of any zygotic transcription at the midblastula transition (83), and before heterochromatin-induced silencing is first genetically detectable (58, 84). This suggests a common factor that affects expressivity between individuals is set down in eggs, sperm, or at fertilization (85), is influential prior to and at the time heterochromatin is first formed, and its influence persists through development until as late as white⁺ activation in pupae.

To determine if this correlation persists throughout development, we analyzed third-instar larval males bearing Y,10C chromosomes and scored for expression of variegating GFP independently in the left and right eye imaginal discs. Silencing of the white⁺ and GFP transgenes on a normally paternally inherited Y,10C was near-complete, so we removed the paternal imprint on the Y (63) by crossing chromosomes through females and analyzing the male progeny of a C(1)DX/Y,10C mother. When total fluorescence in dissected imaginal discs was quantified, the levels of expression of GFP varied between isogenic individuals but were not statistically different between left and right halves of the organism (Fig. 3A) (regression coefficient 1.00, 95% confidence interval 0.71 to 1.30, Pearson’s R² = 0.734; rejected H₀ states that there is no correlation between fluorescence of the left and right discs, specifically that the aforementioned coefficient is not unity, F = 52.44, P < 0.05). These results indicate that the influence of an early-acting factor can be detected throughout development.

Our finding that the propensity to derepress is set before repression exists challenges current concepts for heterochromatin-induced silencing that posit that epigenetic information is stored at a gene promoter, and requires us to consider PEV as a property of heterochromatin strength (i.e., its ability to act as a transcriptional repressor). This has a strong prediction not made by the current model, namely that expressivity in 1 part of the body should correlate with expressivity in other parts of the body. To test this prediction, we again analyzed males bearing matroclinous Y,10C chromosomes. We sorted adult males with strong and weak silencing of white⁺ and then quantified the level of GFP expression in thoraces and abdomens of whole animals. Expression of the former was strongly correlated with the latter (Fig. 3 B and C) (Kendall’s τ = 0.695, rejected H₀ states that there is no correlation between pigmentation category and GFP fluorescence, P < 0.05).

Finally, we confirmed that expressivity was stable through development by hand-sorting male third-instar larvae by categorical (“high” and “low”) GFP fluorescence, and returning them to food and allowing them to grow to adulthood (Fig. 3D). Of 69 larvae with low GFP expression, 59 (86%) had white or near-white eyes as adults, the remainder (14%) possessed eyes with many white⁺-expressing ommatidia. Of 65 larvae with high GFP expression, 17 (26%) were white or nearly so, while 48 (74%) had high levels of white⁺ pigmentation as adults. We conclude that the adult expression of white⁺ is statistically predicted by larval GFP expression (rejected H₀ states that a priori selection based on GFP fluorescence would have no bearing on white⁺ pigmentation as adults, χ² = 48.0305, df = 1, P < 0.05). This effect is unlikely to be cross-talk between enhancers and promoters of the 2 reporter genes because it has been previously shown that induction of 1 variegating gene does not affect variegation of a closely linked variegating gene (60, 86).

These 3 correlations—GFP expression between different tissues of larvae (Fig. 3A), white⁺ expression to GFP expression in adults (Fig. 3 B and C), and larval GFP expression to adult white⁺ expression (Fig. 3D)—all argue that the variation seen in PEV is not reflecting variation in epigenetic information at gene loci, but rather is reflecting organism-by-organism variation in some common factor, set at the time of fertilization, and generally maintained in the soma throughout development.

The currently accepted model for PEV posits that large patches of expression result from early derepression events while smaller clones of salt-and-pepper variegation result from late events of derepression. It is implicit in this model that, regardless of when derepression occurs, silencing once lost is permanently lost. Although many mathematical modelings of epigenetic silencing have hypothetical gains and losses with each mitosis (50, 87–90), we are not aware of any study that has explained any mechanism for gains of silencing. Thus, the assumption that silencing once lost is never gained has remained almost entirely unchallenged.

We envisioned another possibility for the origin of these 2 distinct patterns of clonal variation that can accommodate gains in silencing. We reasoned that it would be possible to create clonal patches or individual salt-and-pepper patterns by leaving fixed the time of “establishment” of heterochromatin-induced silencing (i.e., early embryo), and instead alter the stability (the probability of accurate reestablishment) of silencing in subsequent divisions. This alternative hypothesis might explain the 2 extreme patterns of PEV. Large patches could arise from early establishment/nonestablishment and relatively low frequencies of gains and losses in silencing (decisions once made are faithfully maintained through development). Salt-and-pepper variegation could arise from equally early establishment/nonestablishment followed by relatively high frequencies in switching (many subsequent losses and gains of silencing through development). To establish the validity of this hypothesis, we first created a simplified simulation to determine if stability rather than timing could create patched and salt-and-pepper patterns of PEV. The simulation and our analyses can be found in SI Appendix. A key finding was that the alternative stability hypothesis stated above requires the occurrence of gains of silencing throughout development, while the conventional timing hypothesis does not. Therefore, we sought to create a live reporter system sensitive to such gain events.

In order to monitor switches of heterochromatin-induced silencing throughout development, we considered heterochromatin function as a developmental stage, and adopted the G-TRACE developmental lineage tracer (64) to monitor it. G-TRACE contains a UAS-RFP that reports on current gal4 activity, and a paired UAS-FLP and ubiquitin-FRT-STOP-FRT-GFP that reports on historic gal4 activity. To create a system particularly sensitive to acquisitions of silencing, we made gal4 expression ubiquitous and embedded the Gal4 repressor, gal80, in silencing heterochromatin (Fig. 4A).

In this configuration a red fluorescent cell in a nonfluorescent clone would indicate recent gains of silencing of gal80, the point of discrimination between the current and the alternative hypotheses. Yellow fluorescence would indicate persistent silencing, green fluorescence the loss of silencing, and nonfluorescent cells those in which silencing was not established (Fig. 4B). Because RFP expression precedes GFP expression (Fig. 4C, events 1 and 2) and because RFP decays once gal4 is repressed (Fig. 4C, between events 3 and 4), whereas GFP will not, the RFP-to-GFP ratio and size of clones can further be used to infer the timing of gains and losses of silencing. Cells in which red fluorescence is matched with submaximal green fluorescence are those in which silencing has been gained [and Gal80 decayed, with a half-life of ∼2 h (91)], and the degree of green fluorescence will indicate how long ago that happened (Fig. 4D). Cells in which green fluorescence is maximal and red fluorescence is not are those in which silencing was lost, and the degree of RFP decay will indicate how long ago derepression occurred. We used this “switch monitoring” (SwiM) system to detect losses and gains of silencing through development of living eye-antennal imaginal discs.

Because of the high degree of variation in PEV between individuals (Figs. 1 D and E and 2) (92–94) it was impossible for us to meaningfully determine frequencies of these events in populations of flies. We therefore sought to test the stability hypothesis by close analysis of individual imaginal discs (Fig. 5A).

Finally, we combined ubiquitously expressed gal4 and variegating gal80 with a gal4-controlled Kaede photoswitchable protein. Kaede is translated and matures as a green fluorescing protein, but brief exposure to UV radiation converts it to red fluorescence. The former state is relatively unstable (half-life of 6 h), whereas the red form is very stable (half-life greater than 7 d) (102). We photoconverted intact animals, returned them to culture, and allowed them 24 h before dissection and analysis. We once again observed the same types of cell events that we detected using G-TRACE. Yellow fluorescence using Kaede occurs in cells in which gal80 silencing is complete, and corresponds to yellow fluorescence using G-TRACE. Nonfluorescent cells using Kaede corresponds to no or green fluorescence with G-TRACE. Red fluorescence indicates losses of gal80 silencing at some point in development prior to UV photoconversion, just as green (or green-yellow) fluorescence does when using G-TRACE. With Kaede, strictly green fluorescence indicates gains of silencing at some point after UV photoconversion (Fig. 6D). The time between UV treatment and dissection was longer than the time between RFP detection and GFP detection in G-TRACE, and so we expected to see more green fluorescent cells when using Kaede. We observed individual and small clones of green fluorescent cells (Fig. 6 E and F), confirming the presence of gains of silencing in imaginal discs. Individual green fluorescent cells are unequivocally gains of silencing without mitosis (especially in those cells immediately behind the morphogenetic furrow and the final cell division in these tissues), just as individual red fluorescent cells are individuals in which silencing was lost without cell division. Small clones of green fluorescent cells no doubt share ancestry and are coordinately acting in their gain of silencing. Again, this implies that these cells have had some change and are determined to gain silencing some hours (or cell divisions) hence, suggesting the cause of gains and losses of silencing prefigures the actual gains and losses of silencing.

Our results indicate that the potential strength of heterochromatin is set at or before fertilization, incompletely established around gastrulation (60), then lost and gained stochastically and repeatedly during development. Such fluctuations of heterochromatin strength can occur without consequence to the variegating genes, at least until the genes are “queried” for silencing at the time of gene expression. In the case of white or yellow variegation this occurs in late pupae, with the final PEV expressivity and patterns reflecting the final outcome of gains and losses through development up until that point. This is the logical way in which fields of various sizes (i.e., the amount of the eye to be pigmented, the patched vs. salt-and-pepper patterns) can exist prior to when the gene expression that defines them has initiated. In our SwiM system, where the ubiquitin promoter queries gene expression continuously, we directly observe these fluctuations creating fields of cells, related by lineage, whose silencing potential has been recorded through development. The cause of the heterochromatin fluctuation is yet unknown, but could conceivably be 1 or more of a number of factors. Fluctuations may be influenced by changes at the rDNA, which are somatically unstable (103), known to impact PEV (54, 104, 105), and can cause the same phenomenon in yeast (53). Loss of other repeat DNAs (106) that alter PEV expressivity in trans (107, 108) might also affect heterochromatin fluctuations. It is also consistent with our data that fluctuations in concentrations or activities of soluble heterochromatin-establishing factors, such as Piwi-interacting RNAs, small-interfering RNAs, or protein components of heterochromatin (e.g., concentrations of maternally deposited or zygotically expressed HP1) could underlie the patterns of PEV. Some or all of these factors might have been identified previously as dose-limiting genetic modifiers of PEV, although a simple mass action model is likely insufficient to fully account for the behaviors of PEV that we and others have described (see ref. 109 for a detailed treatment).

Viewing PEV as fluctuations in heterochromatin function rather than epigenetic information at gene promoters is a relatively simple shift in our view, but is capable of explaining heretofore difficult-to-explain PEV phenomena. These include the following: 1) The instability of silencing in G1 (51) despite the evident stability through G1 and S, 2) the correlations between 2 variegating reporters in the same nucleus (55), 3) the correlation between PEV and the instability of repeat genes (110, 111), 4) why the temperature sensitive periods for PEV suppression/enhancement do not correspond to the times of the presumed losses of silencing, and 5) our observation here that expressivity precedes activity of a variegating promoter. If this model is true, it renders moot the search for biochemical mechanisms to maintain locus-specific histone modification patterns through S-phase, because this type of maintenance may not exist for PEV. What appears as maintenance may instead be the successes or failures of heterochromatin to persistently reestablish silencing on closely neighboring genes.

Notably, in a study of mouse PEV (50), ectopic recruitment of HP1 induced silencing (49) that could be monitored for stability in the form of self-sustained mitotic maintenance. The empirical rates derived from these experiments did not themselves explain the stability of silencing in either the transgene or natural heterochromatin, and some hypothetical factor was required to properly model self-sustaining heterochromatin-induced gene silencing (50, 87). We suggest that it is not possible to model stable heterochromatin-induced gene silencing without a bona fide block of heterochromatin that is constantly reestablishing silencing (51, 52).

The main SwiM strain contained variegating GAL80 and a ubiquitous GAL4, was y¹ w^67c23; Tp(P{w^+mC=tubP-GAL80^ts}10)PEV.4; P{w^+mC=Act5C-gal4}17bF01/TM6B, Tb¹. The individual in Fig. 5 was y w/Y; P{tubP-GAL80^ts−PEV.4}/+; P{Act5C-gal4}/P{UAS-RFP}, P{UAS-FLP}, P{Ubi-(FRT.STOP.FRT)GFP}, the product of a cross between the SwiM strain and a G-TRACE strain (w*; P{w^+mC=UAS-RedStinger}6, P{w^+mC=UAS-FLP.Exel}3, P{w^+mC=Ubi-p63E(FRT.STOP)Stinger}15F2). In those cell clones in which gal80 is active, the phenotype is RFP⁻ GFP⁻ because of lack of activation of any gal4-dependent transcription. In those cell clones in which gal80 is repressed, the phenotype is RFP⁺ GFP⁺, because of direct activation of RFP by Gal4 and FLP-mediated rearrangement and activation of GFP.

Full genotypes of other strains, and details on culture conditions, are available in SI Appendix.

Images of whole flies were taken with a Sony a7iii attached to a Nikon SMZ-1500 microscope, illuminated with a Peak Plus Tactical LED Flashlight. Larvae were dissected in PBS and were visualized with a Zeiss AxioZoom.v16 and images captured with the Zeiss Axiocam 506-mono camera. Photography and quantification are described in SI Appendix.

Adult male flies were killed by exposure to ether. They were rinsed in 70% ethanol and placed in PBS. Wings and legs were removed for better visualization on a a Zeiss AxioZoom.v16 and images captured with the Zeiss Axiocam 506-mono camera. Quantitation for correlation experiments was performed using NIH image. Larvae were dissected in PBS, and were visualized with photography and quantification are described in SI Appendix. Live third-instar larvae were collected into a Petri dish containing PBS and visualized to select male larvae. These males were further sorted based on high and low GFP, excluding the gonads which were uniformly bright between siblings. Individual larvae were imaged as above and returned to individually labeled vials. Adults were then scored for white⁺ eye pigment and correlated to their GFP fluorescence.

Nonfluorescence images of whole flies were taken with a Sony a7iii attached to a Nikon SMZ-1500 microscope, illuminated with a Peak Plus Tactical LED Flashlight.

In all cases, α was set to 0.05 prior to the study, and we report whether the null hypotheses (H₀) could be rejected based on that criterion. Specific test statistics are reported in the text and figure legends. Justifications for specific tests are described in SI Appendix.