Trypanosoma congolense, one of the parasites behind a cattle disease that tsetse control programmes have been fighting for more than a century with little success, appears to work on both ends of its own transmission chain at once. It alters the mix of volatile compounds coming off the animals it infects, making them more attractive to the tsetse flies that spread it. And in flies already carrying it, several of the genes that build a fly’s sense of smell are more active than in uninfected flies, while in flies carrying mature infective parasites, sequencing found five of the genes behind the fly’s vision less active.
That is the finding reported on 4 September 2026 in Communications Biology by Simon K. Tawich and colleagues at the International Centre of Insect Physiology and Ecology in Nairobi, with co-authors at the University of Pretoria and at UMR INTERTRYP in Montpellier. The paper is open access. It is also an article in press: peer-reviewed and accepted, and due, in the publisher’s words, to be replaced by a final edited version.
The flies chose the infected mouse
The behavioural results are the simplest part, and the flies doing the choosing were uninfected ones. Given a choice between a mouse infected with T. congolense and a healthy one, uninfected Glossina pallidipes fed preferentially on the infected animal, with a Mann-Whitney U of 23 and a P value of 0.000001 across 20 replicate trials of 20 flies each.
Cattle gave the same answer at one remove. In a free-flight two-choice assay using urine odour instead of the animal itself, flies went to the urine of infected cows significantly more often than to urine from uninfected ones, on an independent t-test returning t = 3.445, df = 35, P = 0.0015, across 19 replicate trials of 20 flies each.
Neither result is unprecedented: earlier work found G. pallidipes preferring oxen infected with T. congolense, and Plasmodium is known to alter infected hosts’ skin odour in ways that draw mosquitoes. What was missing was the chemistry.
The compound that showed up in mice and in cattle
To find the chemistry the team collected headspace volatiles for gas chromatography-mass spectrometry. For the mice that meant a single animal in a glass jar, the air around it trapped for 12 hours and eluted in hexane, sampled before infection and again at peak parasitemia. The cattle side was not the air around the animal but the headspace above its urine, from four cows before and after infection. The paper gives no collection protocol for the cow urine, so the two arms are not described to the same standard.
The compound that stood out was dihydro-beta-ionone, induced by infection in both mouse body odour and cow urine. Dihydro-alpha-ionone rose too, along with the phenolics 3-propylphenol and p-cresol.
The mouse evidence for that compound is thinner than the claim built on it. Of four infected mice, three showed a consistent change in odour profile. In those three the relative abundance of dihydro-beta-ionone was 0.98 per cent, 4.8 per cent and 11 per cent. The mean was six times higher in infected mice than in uninfected ones, but the spread was wide enough that the difference was not statistically significant, which the authors attribute to differences in how individual hosts respond. They also state plainly that they cannot rule out infected mice simply being warmer and giving off more carbon dioxide, either of which would attract flies without any parasite chemistry involved.
Inside the fly’s head
For the molecular half the team sequenced head transcriptomes, and their reason for using wild flies is itself a finding. G. pallidipes is the hardest tsetse species to infect. Of 200 flies the researchers tried to infect in the laboratory, exactly one developed the mature metacyclic parasites that sit in the mouthparts and make a fly infectious. So they used field-caught flies, sorted into three groups: uninfected controls, flies carrying only immature midgut parasites, and flies carrying mature infective parasites in the mouthparts.
Infected and uninfected flies separated in a principal component analysis, with the first two components accounting for 36 and 15.1 per cent of the variance. The two infected groups did not separate from each other; they clustered together, and one mouthpart-infected replicate sat with the midgut-only group. Differentially expressed gene counts ran to 63 between mouthpart-infected flies and controls, 145 between midgut-only flies and controls, and 183 between the two infected groups. The flies sequenced were female.
The immune picture is the clearest signal. In flies whose parasites were still crossing the midgut, antimicrobial peptides and peptidoglycan recognition proteins were mostly turned down, which the authors read as the parasite easing its passage by lowering the gut’s ability to kill it.
The chemosensory picture is more mixed than a headline allows. Odorant receptors OR42b, OR85e, OR46a2, OR45a2 and the obligatory co-receptor Orco were raised across more than two replicates per comparison, while OR82a and OR45a3 were lowered and the OR67d family went in both directions. Two odorant-binding proteins, OBP83a3 and OBP99c, were elevated in the midgut-only versus control comparison, and the acid-sensitive ionotropic receptor IR64a rose significantly. Follow-up quantitative PCR found a general trend towards upregulation of the receptors it tested, in what the authors call a parasitemia-dependent manner: the high-parasitemia group showed the greater increases in some. It also disagreed with the sequencing on receptors including OR67a and OR82a, down in the sequencing and up in the PCR, a discrepancy the authors report rather than smooth over and attribute to low transcript abundance and to differences in how the two techniques count.
Building family trees against better-annotated flies also filled gaps in the species’ genome annotation, naming two ionotropic receptors the paper reports as not previously recorded in G. pallidipes and identifying two uncharacterised proteins as gustatory receptors.
The eye genes went the other way
Tsetse flies hunt by smell at long range and by sight at short range, so the team looked at vision too. In mouthpart-infected flies the opsin genes RH1, RH2, RH3, RH5 and RH6 were all reduced relative to controls, at fold changes of 0.67, 0.46, 0.24, 0.72 and 0.18. Retinophilin, which helps stabilise photoreceptors, was up about threefold, which the authors suggest may be compensatory. The team checked six opsins in all, RH7 being the sixth; the species has no RH4.
Their reading is a trade-off: an infected fly leaning harder on odour and less on vision would be tuned for the long-range search that gets it to a host. That is an interpretation of a correlation, and the paper does not claim to have tested it.
The vision result has a crack in it of the same kind as the smell result. The quantitative PCR, run on a fresh batch of flies from the same site rather than on the sequenced ones, did not simply confirm the sequencing. Among the opsins it checked, only RH5 came out significantly downregulated, and that was in the midgut-only group. In flies carrying mature mouthpart parasites the PCR found RH1 and RH5 enhanced, with no comparator named and no significance claim attached, running against the direction the sequencing gave for those genes in that group. The paper is candid that the opsin effect “was not generalized to all six opsin genes checked”, and it names the equivalent discrepancy in the odorant receptors, but not this one. The five fold changes above are the sequencing result, not the whole picture.
Where the 75 per cent comes from
The two numbers most likely to travel out of this paper are the two that will lose their label on the way. Trypanosome-associated odours are reported to reduce infected flies by 75 per cent and cattle disease prevalence by 40 per cent. Neither comes from the field. Both come from an agent-based model built on the trapping data, simulating tsetse feeding on cattle every third day across a 15-day trapping window. The paper’s abstract says so in the same sentence, and that clause carries a great deal of weight. Within the simulation, phenolics produced the 75 per cent reduction in infected flies, and cattle infections fell by roughly 20 to 40 per cent by day 15 depending on the trap scenario, with 40 per cent corresponding to the strongest bait. That last mapping is a reading of the paper rather than a figure it prints: the per-scenario percentages sit only inside a figure panel, and the text says only that stronger removal probabilities produced larger reductions.
The field measurements underneath the model are more modest. Of 2,118 G. pallidipes dissected in the Shimba Hills, 111 were carrying trypanosomes, a natural infection rate of 5.24 per cent.
The paper’s discussion also says the new blends attracted up to twice as many infective flies as a negative and a positive control. The significant two-fold result was against the unbaited negative control only, and the paper reports no statistical comparison against the positive control. The ionone comparison was 1.4-fold and did not reach significance.
Causation is not established here either. Gene expression changed alongside infection; no gene was silenced or expressed in isolation to show it drives the behaviour. The authors name RNA interference, heterologous expression and genome editing as what would be needed to tie a receptor to a behaviour, and cite the limited genetic tractability of G. pallidipes as the reason it has not been done.
Traps in the Shimba Hills
Blue-black biconical traps were baited with a phenolic blend, with ionones, with the established attractant POCA, or left unbaited, and 200 randomly selected flies per treatment were screened by PCR. The phenolic bait was p-cresol, 2-methoxy-4-propylphenol and 3-ethylphenol in equal parts, not quite the list the infection induced. Of the two phenolics the paper reports rising in infected animals, 3-propylphenol and p-cresol, only p-cresol is in that blend. POCA contains both, under their other names of 3-n-propylphenol and 4-methylphenol.
Phenolic-baited traps caught the most infected flies: 11 per cent of their catch, with a 95 per cent confidence interval of 7.4 to 16.1 per cent, twice the proportion in the unbaited control, on a two-proportion z-test returning P = 0.04. Ionone traps caught 7.5 per cent infected, a 1.4-fold increase that was not statistically significant.
The sharpest result comes from a smaller screen: 96 flies per treatment, checked tissue by tissue instead of whole. In it, every fly carrying T. congolense in its mouthparts, the stage that actually transmits, came from a phenolic or an ionone trap, split 55.6 and 44.4 per cent between them. The control and POCA traps caught none of those, despite POCA containing both infection-induced phenolics. Trypanosoma vivax was the commonest species across every treatment, with no significant difference between them, which the authors attribute to that parasite living in the mouthparts regardless.
The logic is selective removal: a trap that preferentially catches the flies capable of transmitting does more per fly than one catching flies at random. That is where the modelled reductions come from.
What the modelling cannot supply is a season’s worth of cattle. The step between this paper and a control tool is a trial that counts infections in animals instead of simulating them, run long enough for the difference to show, and that is the half nobody has paid for. The other half is already standing in the Shimba Hills: the biconical traps, the phenolic blend, and a fly population in which roughly one in twenty is carrying a trypanosome.