Report

Interchromosomal Communication Coordinates Intrinsically Stochastic Expression Between Alleles

See allHide authors and affiliations

Science  07 Feb 2014:
Vol. 343, Issue 6171, pp. 661-665
DOI: 10.1126/science.1243039

Stochasticity and Cell Fate

Stochastic mechanisms can diversify cell fates in nervous systems. In the Drosophila retina, the stochastic distribution of color-sensing photoreceptors is controlled by the random, On/Off expression of the Spineless transcriptional regulator. Johnston and Desplan (p. 661) found that each allele of spineless makes an intrinsically random expression choice controlled by an enhancer and two silencer DNA elements acting at long range. spineless alleles communicate between chromosomes using activating and repressing mechanisms to determine the frequency of expression and coordinate expression state. These findings suggest critical roles for intrinsically random expression decisions and interchromosomal communication in stochastic cell fate specification.

Abstract

Sensory systems use stochastic mechanisms to diversify neuronal subtypes. In the Drosophila eye, stochastic expression of the PAS-bHLH transcription factor Spineless (Ss) determines a random binary subtype choice in R7 photoreceptors. Here, we show that a stochastic, cell-autonomous decision to express ss is made intrinsically by each ss locus. Stochastic on or off expression of each ss allele is determined by combinatorial inputs from one enhancer and two silencers acting at long range. However, the two ss alleles also average their frequency of expression through up-regulatory and down-regulatory interallelic cross-talk. This inter- or intrachromosomal long-range regulation does not require endogenous ss chromosomal positioning or pairing. Therefore, although individual ss alleles make independent stochastic choices, interchromosomal communication coordinates expression state between alleles, ensuring that they are both expressed in the same random subset of R7s.

Developmental programs generally induce uniform or regionalized gene expression patterns to yield highly reproducible body-plan outcomes. However, stochastic mechanisms are sometimes incorporated to diversify cell types in nervous systems. Nonautonomous stochastic mechanisms using lateral inhibition strategies have been well described, whereas cell-autonomous, stochastic mechanisms involved in color opsin and olfactory receptor selection in mammals are only partially understood (1, 2).

The fly eye is composed of two stochastically distributed subtypes of ommatidia (unit eyes) defined by expression of specific light-detecting Rhodopsin proteins in R7 photoreceptors (PRs). The random distribution is controlled by the stochastic expression of the Per-Arnt-Sim basic helix-loop-helix (PAS-bHLH) transcription factor Spineless (Ss). Ss expression in ~65% of randomly distributed R7s induces “yellow” (yR7) fate and expression of Rhodopsin4 (Rh4), whereas the absence of Ss in the remaining ~35% of R7s allows for “pale” (pR7) fate and Rhodopsin3 (Rh3) expression (Fig. 1, A and B). Loss of ss function leads to the transformation of all R7s to pR7 fate and Rh3 expression (fig. S1A), whereas ectopic Ss causes all R7s to acquire yR7 fate and express Rh4 (fig. S1B) (15).

Fig. 1 The stochastic decision to express ss is made early and maintained.

(A) Ss is absent from pR7s, allowing for Rh3 expression. Ss is expressed in yR7s activating Rh4 and repressing Rh3. (B) Stochastic distribution of Rh3- and Rh4-expressing R7s. (C) Ss is expressed in a random subset of R7s throughout development. Pros marks all R7s.

Ss was observed in 65% of randomly distributed R7s throughout development (Fig. 1C and fig. S1, D to G). Ss expression in adults perfectly correlated with Rh4 expression (fig. S1C). We never observed switching of Rh expression (6). Therefore, Ss expression is established and stably maintained throughout the lifetime of yR7 cells.

We evaluated reporter lines containing fragments of the ss gene (7). Fragment 8 (R7/R8 enhancer) in mini-gene1 induced lacZ expression in all R7s and R8s (Fig. 2, A and B), which closely resembled expression of the Salm zinc finger transcription factor (with Salr, collectively referred to as Sal) (Fig. 2C) that specifies R7 and R8 fate (8). Ss expression was completely lost in sal mutants (Fig. 2D), whereas ectopic expression of Salm in all PRs led to the activation of Ss in a random subset of outer PRs (Fig. 2E and fig. S2A) and expression of Mini-gene1 in outer PRs (fig. S2B). Thus, Sal is necessary and sufficient to activate stochastic expression of Ss in PRs. The choice to express Ss is cell autonomous because R7s and outer PRs within the same ommatidium made their decisions to express Ss independently of one another (Fig. 2E and fig. S2A).

Fig. 2 The cis-regulatory logic controlling intrinsically stochastic ss expression.

(A) ss locus schematic. F, Fragment; red boxes, silencers; blue box, InterCom element; purple box, minimal promoter; green box, R7/R8 enhancer; gray circles, untranslated exons; yellow circles, translated exons; arrows, transcriptional starts. (B to E) White circles indicate expression, and gray circles indicate no expression. (B) Mini-gene1 is expressed in all R7s and R8s. (C) Sal is expressed in all R7s and R8s. (D) Ss expression is completely lost in sal mutants. (E) Ectopic Sal expression in svp mutants causes Ss expression in a random subset of PRs. (F) Mini-gene2 induces expression in a subset of R7s independently of endogenous Ss/Rh4 expression. Mini-gene2 localizes to the nucleus, whereas Rh4/Ss localizes to membranous rhabdomere structures. The four possible combinations of expression are observed: (i) white solid ovals, mini-gene2 and Ss/Rh4; (ii) white dashed ovals, Ss/Rh4 only; (iii) gray solid ovals, mini-gene2 only; and (iv) gray dotted ovals, no expression. (G) Four expression combinations in (F). (H to J) Models for random expression decisions. (H) The ss locus randomly assumes one of two (i.e., active or repressed) DNA looping configurations. (I) One silencer facilitates the nucleation of closed chromatin state spreading from the other silencer. (J) One silencer lowers expression in all R7s, whereas the other specifically provides the stochastic input (through looping or spreading).

To identify DNA silencer elements required for stochastic Ss expression, we first defined the minimal ss DNA sequence required for stochastic ss expression. We used green fluorescent protein (GFP) from transgenes or Rh4 expression as a readout of Ss expression (Ss/Rh4), because Rh4 is always a perfect indication of Ss expression in R7s (fig. S1C) (5). An inversion (9) and transgene1 exhibited stochastic Ss expression and therefore defined the 5′ endpoint (Fig. 2A and fig. S2C). A duplication with a breakpoint in the ss 3′ UTR (10) and transgene2 similarly exhibited stochastic Ss expression, defining the 3′ endpoint (Fig. 2A and fig. S2D). These data determine a 55.5-kb minimal DNA sequence required for stochastic ss expression (Fig. 2A).

We identified two DNA elements that are critical for stochastic ss expression. transgene3 and transgene4 displayed expression in all R7s, suggesting that an intragenic silencer (silencer2) is required for stochastic ss expression (Fig. 2A and fig. S2, E and F). transgene5 and transgene6 also displayed expression in all R7s, suggesting that a 5′ upstream silencer (silencer1) is also required for stochastic ss expression (Fig. 2A and fig. S2, G and H). A 36-kb deficiency that removed silencer1 (sil1 deficiency) and an inversion allele in which the ss coding region was moved 12 Mb away from silencer1 (ss high freq) showed expression of Ss/Rh4 in all R7s (Fig. 2A and figs. S2I and S3, A and E), validating the requirement for silencer1. Therefore, stochastic Ss expression requires an enhancer and two silencer elements.

When a ~3-kb fragment of silencer1 was placed with the R7/R8 enh+prom element driving reporter expression (mini-gene2), we observed expression in a random subset of R7s (Fig. 2F), showing that silencer1 is sufficient to repress expression when present close to the enhancer and promoter. If the stochastic expression decision occurred intrinsically at each ss locus, mini-gene2 should induce reporter expression independently of expression from the endogenous ss loci. We compared expression of mini-gene2 to endogenous Ss/Rh4 expression and found all four possible expression combinations (Fig. 2, F and G), suggesting that each ss locus makes an independent, stochastic expression decision.

transgene4, which was inserted 4.6 Mb away from the ss locus, drove GFP expression in all R7s (Fig. 2A). Although transgene4 should not affect endogenous Ss expression, we observed a dramatic increase in the frequency of Ss/Rh4 expression in animals carrying transgene4, suggesting that transgene4 up-regulated the frequency of Ss expression from the endogenous ss loci (Fig. 3C). transgene4 up-regulated expression from ss loci in cis, or in trans (Fig. 3, A to D), suggesting that it contains DNA elements that are sufficient to drive regulatory interactions in the absence of chromosomal pairing. transgene4 also up-regulated expression, although less efficiently, from the ss locus translocated on a different chromosome (Fig. 3, E and F), suggesting that ss alleles can interact at a distance but that chromosomal position plays a role in this process. These observations strongly implicate direct interactions between DNA elements in the transgene and endogenous loci but do not exclude possible indirect mechanisms such as noncoding RNAs. transgene4 must contain a DNA element (InterCom element) between 25 and 8 kb upstream of the ss transcription start site, which is missing in mini-gene2 (Fig. 2, F and G, and fig. S3).

Fig. 3 ss regulatory regions up-regulate and down-regulate expression frequency through interchromosomal communication.

(A) Wild-type ss locus over deficiency1. (B) transgene4 up-regulates expression frequency from the endogenous ss gene in cis. (C) Quantification of (A), (B), and (D). (D) transgene4 up-regulates expression frequency from the endogenous ss gene in trans. (E) transgene4 up-regulates expression, although less efficiently, from the ss locus on the nonhomologous 2nd chromosome. (F) Quantification of (E). (G) Ss/Rh4 is not expressed in sslow freq1 hemizygous mutants. (H) The normal regulatory regions of the ssprot null1 allele up-regulate Ss/Rh4 expression from the sslow freq1 allele. (I) Wild-type ss homozygous loci. (J) The regulatory regions of the sslow freq1 allele down-regulate Ss/Rh4 expression from the wild-type ss allele. (K) Quantification of (G) to (J).

We next found that one ss allele could up-regulate the frequency of expression from the other allele. sslow freq1, an allele affecting noncoding regions, was expressed at very low frequency when placed over ssdeficiency alleles (Fig. 3, G and K, and fig.S4A). When sslow freq1 was placed over ssprot null1, a protein coding null allele with normal cis-regulatory regions, the frequency of Ss/Rh4 expression dramatically increased (Fig. 3, H and K), suggesting that the cis-regulatory elements from ssprot null1 up-regulated expression frequency from sslow freq1. We verified our observations with additional allelic combinations (fig. S4, A and B).

The up-regulation of expression from one allele with impaired regulatory regions but normal protein function by another allele with normal regulatory regions but impaired protein function resembles transvection, initially described by Lewis (11). Transvection is defined as the complementation of mutant alleles requiring position-dependent chromosomal pairing. Because the interallelic control of ss does not require position-dependent chromosomal pairing (Fig. 3, B, D, and E, and fig. S3B and S4, B and C) and does not appear to require regulation by known mediators of transvection (fig. S5), we conclude that this phenomenon is not a canonical case of transvection.

We also found that one ss allele could mediate the down-regulation of expression frequency from the other allele. The sslow freq1 allele down-regulated expression frequency from the sswild-type alleles, because the proportion of R7s expressing Ss/Rh4 was lower in sslow freq1/sswild-type animals compared with sswild-type homozygotes (Fig. 3, I to K). Down-regulation did not require endogenous ss chromosomal position because it also occurred for a wild-type ss locus on an inversion (fig. S4C). We confirmed down-regulation with additional allelic combinations (fig. S3, A, B, and E, and fig. S4C). Thus, ss alleles regulate one another through long-range interchromosomal activating and repressing mechanisms to determine the frequency of ss expression.

If each ss allele makes its own expression decision, expression states will sometimes agree (both alleles on or off) and other times disagree (one allele on and the other off). We tested whether interchromosomal communication functioned to coordinate the expression state from the two ss alleles.

The sstrunc allele has normal regulatory regions but contains a mutation that truncates the Ss protein activation domain (5, 10). This truncation weakens Ss protein function such that Ss activates Rh4 normally, but fails to repress Rh3, leading to coexpression of Rh3 and Rh4 in nearly all yR7s and normal Rh3 expression in pR7s (Fig. 4B).

Fig. 4 Interchromosomal communication coordinates expression from ss alleles.

(A) Rh3 is expressed in pR7s, and Rh4 is expressed in yR7s in wild-type hemizygous animals. (B) Rh3 is expressed in pR7s and Rh4 and Rh3 are expressed in yR7s in sstrunc hemizygous animals. (C) In sswild-type(inv2)/sstrunc animals, Rh3 and Rh4 are always expressed exclusively. (D) In sslow freq1/sstrunc animals, the normal regulatory regions of the sstrunc allele up-regulate expression from sslow freq1 into the same subset of R7s because Rh3 and Rh4 are expressed exclusively. (E to G) Models for the coordination of expression state through interchromosomal communication. (E) A temporally distinct two-step mechanism involving both alleles making independent expression decisions followed by an activating and repressing tug of war. (F) A temporally distinct two-step mechanism in which one allele makes the decision and then imposes the decision onto the other naïve allele. (G) A mechanism involving contemporaneous decisions that average the activating and repressing inputs from each allele.

We evaluated sswild-type/sstrunc animals to determine whether these alleles were expressed in an independent or coordinated manner. If the two ss alleles were expressed independently, sswild-type/sstrunc would produce three Rh expression outcomes: (i) Rh3 alone (neither sswild-type nor sstrunc are expressed); (ii) Rh4 alone (sswild-type alone or both sswild-type and sstrunc are expressed); and (iii) Rh3 and Rh4 coexpression (sstrunc alone is expressed) (fig. S6C). Alternatively, coordinated expression from the two alleles would yield two Rh expression outcomes: (i) Rh3 alone (neither sswild-type nor sstrunc are expressed) and (ii) Rh4 alone (both sswild-type and sstrunc are expressed) (fig. S6D). For these experiments, the wild-type ss allele was on an inverted chromosome (sswild-type(inv2)) to prevent pairing of homologous chromosomes. For sswild-type(inv2)/sstrunc flies, we observed expression of Rh4 alone and Rh3 alone, but never coexpression of Rh3 and Rh4 (Fig. 4C). The wild-type ss locus on a different inverted chromosome (sswild-type(inv3)) over sstrunc displayed similar expression coordination (see the supplementary materials). Together, these data suggest that expression from the two ss alleles is coordinated and that endogenous ss position on homologous chromosomes is not critical.

We next investigated whether interchromosomal communication was able to coordinate expression from two ss alleles with widely different expression frequencies. sslow freq1 expressed fully functional Ss protein but at a low frequency (Fig. 3G). We predicted that sslow freq1/sstrunc animals should display up-regulation of Ss expression from sslow freq1 due to interchromosomal communication from the normal cis-regulatory elements of sstrunc. sslow freq1/sstrunc flies displayed nearly perfect coordination of expression from the two alleles, with almost no coexpression of Rh3 and Rh4 (Fig. 4D), verifying that interchromosomal communication coordinates expression from the two ss alleles.

Because stochastic Ss expression requires an enhancer and two silencer elements, we propose three possible mechanistic models controlling the decision: (i) the ss locus randomly assumes one of two (i.e., active or repressed) DNA looping configurations; (ii) one silencer facilitates the nucleation of closed chromatin state spreading from the other silencer; and (iii) one silencer generally lowers expression in all R7s, whereas the other specifically provides the stochastic input (through looping or spreading) (Fig. 2, H to J).

Similarly, we envision three models for how interchromosomal communication coordinates expression: (i) a temporally distinct two-step mechanism involving both alleles making independent expression decisions followed by an activating and repressing tug of war; (ii) a temporally distinct two-step mechanism in which one allele makes the decision and then imposes the decision onto the other naïve allele; and (iii) a mechanism involving contemporaneous decisions that average the activating and repressing inputs from each allele (Fig. 4, E to G).

Interchromosomal communication is reminiscent of transvection. In contrast to transvection-like processes that allow allelic complementation between null alleles whose biological meaning is unclear, interchromosomal communication regulating ss appears to have dedicated biological functions to average the frequency and coordinate expression state between stochastically expressed alleles.

The color vision systems of flies and humans present an interesting case of convergent evolution. In both species, the apparent goal is the same: Use stochastic mechanisms to diversify cell fates and distribute color sensory capacities across the eye. The fly eye requires an enhancer and two silencer elements to achieve stochastic expression of ss, whereas the human eye uses random locus control region (LCR)–mediated activation of M (middle-wavelength sensitive) or L (long-wavelength sensitive) opsins. To avoid disagreement in allelic expression states, interchromosomal communication coordinates expression in flies, whereas X-inactivation completely turns off expression from one allele in females and there is only one copy of the locus in males, creating a mono-allelic expression decision in both cases (1, 2).

Stochastic gene expression mechanisms may be a cost-effective way to diversify the repertoire of cell fates within a tissue. Although these phenomena involve stochastic processes, this randomness is very often well controlled, incorporating multiple steps, apparently to ensure robustness. Evolution has yielded many different mechanisms to determine stochastic cell fate specification in bacteria, flies, and vertebrates (2). As our understanding of stochastic phenomena increases, it will be interesting to see whether common, ancestral strategies become apparent or whether novel stochastic gene expression mechanisms arise in individual species.

Supplementary Materials

www.sciencemag.org/content/343/6171/661/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S6

References (1227)

  • * Present address: Department of Biology, Johns Hopkins University, 3400 North Charles Street, Baltimore, MD 21218, USA.

References and Notes

  1. Acknowledgments: We are very grateful to S. Britt, I. Duncan, C. Zuker, and the Bloomington Stock Center for reagents. We thank C. Tsanis and T. Blackman for technical assistance and E. Heard, O. Hobert, J. Kassis, P. O’Farrell, V. Pirrotta, and members of the Desplan laboratory for helpful discussion concerning the project and manuscript. C.D. was supported by NIH R01 EY13010. R.J.J. Jr. was supported by a Jane Coffin Childs Memorial Fund for Medical Research postdoctoral fellowship.

Stay Connected to Science

Navigate This Article