Category: 8. Health

A new study published in Imaging Neuroscience has found that large language models with visual processing abilities, such as GPT-4V, can evaluate and describe social interactions in images and short videos in a way that closely matches human perception. The research suggests that artificial intelligence can not only identify individual social cues, but also capture the underlying structure of how humans perceive social information.

Large language models (LLMs) are advanced machine learning systems that can generate human-like responses to text inputs. Over the past few years, LLMs have become capable of passing professional exams, emulating personality traits, and simulating theory of mind. More recently, models such as GPT-4V have gained the ability to process visual inputs, making it possible for them to “see” and describe scenes, objects, and people.

This leap in visual capability opens new possibilities for psychological research. Human social perception depends heavily on our ability to make quick inferences from visual input—interpreting facial expressions, body posture, and interactions between people.

If AI models can match or approximate these human judgments, they may offer scalable tools for behavioral science and cognitive neuroscience. But the key question remains: How well can AI interpret the nuanced, often ambiguous social signals that humans rely on?

To explore this question, researchers at the University of Turku used OpenAI’s GPT-4V to evaluate a set of 468 static images and 234 short video clips, all depicting scenes with rich social content drawn from Hollywood films. The goal was to see whether GPT-4V could detect the presence of 138 different social features—ranging from concrete behaviors like “laughing” or “touching someone” to abstract traits like “dominant” or “empathetic.”

These same images and videos had previously been annotated by a large group of human participants. In total, over 2,200 individuals contributed more than 980,000 perceptual judgments using a sliding scale from “not at all” to “very much” to rate each feature. The human evaluations were used as a reference point to assess how closely GPT-4V’s ratings aligned with the consensus of real observers.

For each image or video, the researchers prompted GPT-4V to generate numerical ratings for the full set of social features. They repeated this process five times to account for the model’s variability, then averaged the results. In the case of video clips, since GPT-4V cannot yet directly process motion, the researchers extracted eight representative frames and added the transcribed dialogue from the clip.

The results showed a high level of agreement between GPT-4V and human observers. The correlation between AI and human ratings was 0.79 for both images and videos—a level that approaches the reliability seen between individual human participants. In fact, GPT-4V outperformed single human raters for 95% of the social features in images and 85% in videos.

However, GPT-4V’s ratings did not always match group-level consensus. When compared to the average of five human raters, the AI’s agreement was slightly lower, particularly for video clips. This suggests that while GPT-4V provides a strong approximation of human perception, its reliability may not yet match the collective judgment of multiple human observers working together.

The study also examined whether GPT-4V captured the deeper structure of how humans organize social information. Using statistical techniques such as principal coordinate analysis, the researchers found that the dimensions GPT-4V used to represent the social world—such as dominant vs. empathetic or playful vs. sexual—were strikingly similar to those found in human data.

This suggests that the model is not only mimicking surface-level judgments but may be tapping into similar patterns of representation that humans use to make sense of social interactions.

To take the comparison one step further, the researchers used GPT-4V’s social feature annotations as predictors in a functional MRI (fMRI) study. Ninety-seven participants had previously watched a medley of 96 short, socially rich video clips while undergoing brain scans. By linking the social features present in each video to patterns of brain activity, the researchers could map which areas of the brain respond to which types of social information.

Remarkably, GPT-4V-based stimulus models produced nearly identical brain activation maps as those generated using human annotations. The correlation between the two sets of maps was extremely high (r = 0.95), and both identified a similar network of regions—such as the superior temporal sulcus, temporoparietal junction, and fusiform gyrus—as being involved in processing social cues.

This finding provides evidence that GPT-4V’s judgments can be used to model how the brain perceives and organizes social information. It also suggests that AI models could assist in designing and interpreting future neuroimaging experiments, especially in cases where manual annotation would be time-consuming or expensive.

These findings open several possible directions for future research and real-world applications. In neuroscience, LLMs like GPT-4V could help generate high-dimensional annotations of complex stimuli, allowing researchers to reanalyze existing brain data or design new experiments with greater precision. In behavioral science, AI could serve as a scalable tool for labeling emotional and social content in large datasets.

Outside the lab, this technology could support mental health care, by identifying signs of distress in patient interactions, or improve customer service by analyzing emotional cues in video calls. It could also be used in surveillance systems to detect potential conflicts or identify unusual social behaviors in real-time settings.

At the same time, the study’s authors caution that these models are not perfect replacements for human judgment. GPT-4V performed worse on some social features that involve more subjective or ambiguous judgments, such as “ignoring someone” or “harassing someone.” These types of evaluations may require contextual understanding that AI systems still lack, or may be influenced by training data biases or content moderation filters.

The model also tended to rate low-level features more conservatively than humans—possibly due to its probabilistic nature or its safeguards against generating controversial outputs. In some cases, the AI refused to evaluate scenes containing sexual or violent content, highlighting the constraints imposed by platform-level safety policies.

While the results are promising, some limitations should be noted. The AI ratings were compared against a relatively small number of human raters per stimulus, and larger datasets could provide a more robust benchmark. The model was also tested on short, scripted film clips rather than real-world or live interactions, so its performance in more natural settings remains an open question.

Future work could explore whether tailoring LLMs to specific demographic perspectives improves their alignment with particular groups. Researchers might also investigate how AI models form these judgments—what internal processes or representations they use—and whether these resemble the mechanisms underlying human social cognition.

The study, “GPT-4V shows human-like social perceptual capabilities at phenomenological and neural levels,” was authored by Severi Santavirta, Yuhang Wu, Lauri Suominen, and Lauri Nummenmaa.

Currently, Alzheimer’s disease is not diagnosed until patients already have evidence of plaques and tangles in their brain.

Credit: iStock.com/wildpixel

Long before the telltale signs of dementia emerge in a patient with Alzheimer’s disease (AD), early signals could be lurking within their blood. The opportunity to identify these markers and diagnose someone much earlier with a simple blood test could significantly advance the field, allowing clinical trials to enroll patients and develop treatments that could one day prevent or even reverse the course of disease. 

Today, the gold standard for diagnosing AD requires imaging with expensive and radioactive positron emission tomography (PET) scans or invasive spinal taps to gather cerebrospinal fluid (CSF), along with cognitive testing.

A noninvasive option

The reality of a basic blood test to diagnose AD may not be too far away. In May, the FDA approved the first blood test called Lumipulse from Fujirebio Diagnostics to detect proteins in the blood indicative of amyloid plaques in the brain in adults aged 55 and over. Specifically, the test looks for pTau217 (phosphorylated tau 217) and beta (ß)-amyloid 1-42 as indications of the buildup of amyloid plaques between neurons in the brain. 

However, experts say this test and others like it can’t provide a diagnosis yet. “These blood tests are sensors that reflect ß-amyloid pathology — a necessary but insufficient ‘diagnostic,’” said Sally Frautschy, a neuroscientist studying AD at the University of California, Los Angeles (UCLA). “The tests are a game changer in some respects, such as in lowering costs for and streamlining clinical trials, and excluding AD as a cause of dementia, but do not provide insight into the regional distribution of AD in the brain or mixed pathologies needed for a definitive or differential diagnosis.”

The FDA agreed, stating in their news release: “Importantly, the Lumipulse G pTau217/ß-Amyloid 1-42 Plasma Ratio is not intended as a screening or stand-alone diagnostic test and other clinical evaluations or additional tests should be used for determining treatment options.” Yet, as a measure of amyloid pathology on its own, the FDA cleared Lumipulse for marketing based on the results of a clinical study of 499 patients showing that 91.7 percent of samples with a positive result also had the presence of amyloid plaques confirmed by a PET scan or CSF test, and 97.3 percent with a negative result showed negative PET or CSF results. 

The Lumipulse blood test could be used in place of the more expensive and invasive options to identify amyloid plaques — and new guidelines suggest doing exactly that. In July, the Alzheimer’s Association released its first recommendations for blood-based biomarker (BBM) tests, put forward by a panel of 11 clinicians. Although Fujirebio’s assay is the first to be approved, many more are in development, and the panel decided not to endorse specific tests. Instead, their new guidelines say that in patients with cognitive impairments being evaluated at specialty care clinics, tests with greater than 90 percent sensitivity and 75 percent specificity can be used to rule out AD pathology, and tests with both sensitivity and specificity above 90 percent can be used in place of PET amyloid imaging or CSF testing. In a summary document, the Alzheimer’s Association stated, “As ‘living guidelines,’ these recommendations will evolve with new evidence on BBM performance, diverse populations, and advanced analytical approaches.”

Future diagnostic blood tests

As more companies design and evolve their own blood-based assays, more research is needed to determine what the best biomarker is to measure in the first place. As an example, while Fujirebio’s assay measures pTau217 relative to ß-amyloid 1-42, another assay developed by Roche in partnership with Eli Lilly measures pTau181 relative to ß-amyloid 1-42 as an indicator of amyloid pathology. Their test just received a CE mark from the European Union with a sensitivity of 83.6 percent and specificity of 93.8 percent. 

Frautschy explained that although both assays are similar in predicting ß-amyloid accumulation in the brain, the type of tau antibody used in the pTau217 Fujirebio assay has been shown to be the most sensitive to the earliest stages of ß-amyloid accumulation. In contrast, the Roche-Lilly pTau181 assay may detect slightly later stages, but still early disease.

She emphasized that the current readouts — ratios of tau to ß-amyloid 1-42 — only reflect amyloid burden, which does not directly correlate with memory loss. These assays also do not necessarily reflect the later stage of tau tangle pathology that is known to drive rapid memory deterioration. Thus, Frautschy said that additional ratios incorporating other proteins will likely be necessary for more informative diagnostics that can track and detect the causes of memory loss. Fortunately, she noted that markers that directly reflect the brain pathology responsible for memory decline are currently in development.

In the future, Frautschy emphasized that “the ideal biomarker panel would distinguish between tauopathy alone and AD, which requires both tau and amyloid pathology.” 

With cases on the rise, researchers are aiming to develop a “one-two punch” vaccine strategy against pertussis, more commonly known as whooping cough. This work could ultimately help reduce transmission of the disease and combat the threat of resistant bacteria mutating into stronger, more lethal “superbugs.”

The project aims to improve upon current vaccines by not only neutralizing whooping cough symptoms, as available vaccines do, but also destroying the bacterium behind the illness.

Led by Michigan State’s Xuefei Huang and Ohio State’s Rajendar Deora, the land grant collaboration is supported by the National Institutes of Health.

“If you think of bacteria as criminals, they’re very good at masking themselves from our immune system,” Huang said, an MSU Foundation Professor in the departments of Chemistry and Biomedical Engineering and the Institute for Quantitative Health Science & Engineering.

“With this project, we want to retrain our immune system to look for new dangers and keep better watch within our bodies.”

An evolving enemy

Whooping cough is one of the leading causes of child illness and mortality in the world, earning its grim name for the gasping attempts at breath that follow uncontrollable coughing fits.

With infection rates reaching a six-decade high, researchers face growing challenges to stay ahead of the microscopic culprit causing the disease itself, a bacterium called Bordetella pertussis.

Current vaccines target a toxin and other proteins that facilitate the sticking of Bordetella pertussis to the human cells. The pertussis toxin leads to respiratory distress and in childhood vaccines like DTaP, you’ll find a “de-toxified” version of the substance that helps train the immune system for future fights.

Through their latest collaboration, the Spartan and Buckeye scientists are pursuing what Huang calls “complementary medicine” — not a search to outright replace current vaccines, but to strengthen them in the light of new, worrisome developments.

“One problem with the current vaccine is that it treats the symptoms, but not the cause,” Huang explained, who noted that while the vaccine can successfully neutralize the pertussis toxin, the bacterium itself is untouched.

“This means people might become symptom-free carriers, spreading bacteria without knowing it.”

Another concern is how some strains of Bordetella pertussis have evolved to evade our current arsenal of vaccines by shedding a protein called pertactin.

This protein helps the bacteria colonize the respiratory tract and, in whooping cough vaccines, acts as an antigen. Antigens stimulate a response from the immune system, preparing it against future encounters with harmful pathogens.

Researchers worry that the appearance of mutant strains of Bordetella pertussis without pertactin are more likely to thrive and potentially evolve into dreaded “superbugs,” microorganisms resistant to being cleared by the currently available vaccines.

“To control whooping cough, you need ‘whopping’ new ways to think about the problem,” said Rajendar Deora, a professor in the Departments of Microbial Infection and Immunity and Microbiology, at OSU’s Wexner Medical Center.

Sweet targets

To achieve their “one-two-punch” against whooping cough, Huang and Deora will develop an antigen alternative to pertactin and attach it to a cutting-edge delivery system.

For their antigen, the team is leveraging the chemistry of carbohydrates and a group of complex sugars called glycans. Glycans are found on the surface human cells and cancer cells, as well as viruses and bacteria, making them a tantalizing target for vaccine research.

In 2020, Huang and Deora successfully synthesized several unique sugars found on the surface of Bordetella pertussis, opening the door for the latest vaccine project.

If our immune system knew how to recognize and bind to these sugars directly, it might potentially kill the bacteria and not just silence the symptoms caused by its toxin.

“These carbohydrates are difficult to isolate from natural sources, so we’ve developed methods to build them with precise control,” said Sherif Ramadan, a research assistant professor at MSU. “That kind of precision is key — it ensures the immune system is being trained to recognize the right molecular patterns.”

Ramadan, an expert in glycan synthesis and conjugate vaccine design, has helped lead the chemical construction of pertussis-specific glycans that closely mimic the native bacterial surface.

“By building these glycans from scratch, we’re able to recreate the pathogen’s signature in a way that’s never been done before,” he said.

To give the immune system a better chance at detecting these glycan-based antigens, the researchers are attaching them to a nanoparticle called Qβ,or Q Beta.

A soccer ball-shaped protein with 180 identical sides, Q Beta is an antigen delivery vehicle representing more than 20 years of research from Huang’s lab.

The huge promise of Q Beta — which has applications in the fight against cancer, antibiotic resistant infections and even opioid addiction — earned Huang the MSU Innovation Center’s Technology Transfer Achievement Award.

With further support from the Innovation Center and MSU Technologies, Huang also founded Iaso Therapeutics, a company looking to bring this technology out of the lab and onto the frontlines of health science.

The power of Q Beta lies in the fact that our immune system is best at recognizing threats when they present themselves in organized patterns.

With a repeating structure that’s been mapped at atomic-scale resolution, the carrier particle is a perfect canvas to place antigens in attention-grabbing patterns that the immune system can’t ignore.

“Carbohydrates typically provoke weaker immune responses, but when we place them on Q Beta, we’ve shown they can powerfully activate the immune system compared to traditional methods,” Huang said.

“We’re flipping the switch, training our immune system to be on the lookout for new characteristics,” he added.

By combining this glycan antigen and Q Beta duo with the pertussis toxin, the team hopes to develop a potent and multipurpose vaccine system — one that not only alleviates whooping cough symptoms but stops Bordetella pertussis in its tracks.

This story originally appeared on the Department of Chemistry website.

Two new papers from Clinical Gastroenterology and Hepatology and Gastroenterology shed light on how gut-brain interactions, influenced by both biology and life circumstances, shape eating behaviors. Together, they highlight the importance of multidisciplinary, personalized approaches to digestive health and nutrition. 

Social determinants of health and the gut-brain-microbiome axis in obesity 

Researchers found that stress from life circumstances can disrupt the brain-gut-microbiome balance. This disruption may alter mood, decision-making, and hunger signals – increasing the likelihood of craving and consuming high-calorie foods. 

The paper examined how social factors, such as income, education, health care access, and biological aspects, interact with the brain-gut-microbiome to influence eating habits and obesity. 

By understanding these interactions, clinicians can better tailor treatment and support to individuals living with obesity, ultimately enhancing outcomes and quality of life. 

Avoidant/restrictive food intake disorder symptoms in adults with disorders of gut-brain interaction 

In the first general population study of its kind, researchers found that adults with disorders of gut-brain interaction are significantly more likely to screen positive for symptoms of avoidant/restrictive food intake disorder. 

Key findings: 

• More than one-third of adults with disorders of gut-brain interaction screened positive for avoidant/restrictive food intake disorder symptoms. 

• Symptoms include sensory-based food avoidance, lack of interest in eating, and fear of aversive consequences. 

Key takeaway: These papers show that gut-brain communication is central to eating behavior disorders and that non-biological stressors – like discrimination or past illness – can shape physiological responses. Clinical tools must address not just what patients eat, but why. 

Breast milk is the first ‘super food’ for many babies. Full of vitamins, minerals, and other bioactive compounds, it helps build the young immune system and is widely considered the optimal source of infant nutrition. Not all mothers, however, have the opportunity to directly breastfeed multiple times during the day and night, and might use expressed milk stored for later.

Breast milk delivers a variety of cues from the mother to the infant, including signals that are thought to influence babies’ circadian rhythms. The hormones and proteins involved in circadian signaling, however, may vary in breast milk concentration over 24 hours. To learn more about these fluctuations, researchers in the US investigated expressed breast milk samples taken during different times of the day. They published their findings in Frontiers in Nutrition.

We noted differences in the concentrations of bioactive components in breast milk based on time of day, reinforcing that breast milk is a dynamic food. Consideration should be given to the time it is fed to the infant when expressed breast milk is used.”

Dr. Melissa Woortman, first author, a recent PhD graduate, Department of Nutritional Sciences, Rutgers University

“The timing of these cues would be particularly critical in early life, when the infant’s internal circadian clock is still maturing,” added senior author Prof Maria Gloria Dominguez-Bello, a researcher at the Department of Biochemistry and Microbiology at Rutgers University.

Powerful compounds

The researchers took 10 milliliter breast milk samples from 21 participants at 6am, 12pm, 6pm, and 12am on two different days, which were about a month apart. A further 17 participants provided samples taken at the same times once, resulting in 236 samples included in the analysis. The samples were examined for levels of melatonin, cortisol, and oxytocin – all hormones – as well as immunoglobulin A (IgA), an antibody protein part of the immune system, and lactoferrin, a milk protein. Melatonin and cortisol are involved in the regulation of the circadian rhythm, whereas the other examined components influence intestinal development and gut microbiome dynamics.

They found that some breast milk components, especially melatonin and cortisol, varied over the course of the day. Melatonin peaked at midnight, whereas cortisol was at the highest level in the early morning. “We all have circadian rhythms in our blood, and in lactating mothers, these are often reflected in breast milk,” explained Woortman. “Hormones like melatonin and cortisol follow these rhythms and enter milk from maternal circulation.” The other examined components were mostly stable throughout the day. This might be because they may not be as strongly influenced by signals dictating circadian rhythms.

The team also found that as infants got older, the levels of different compounds in breast milk varied. For example, the levels of cortisol, IgA, and lactoferrin were highest when babies were less than one month old. Higher levels of these compounds likely support immune defense and gut colonization in very young babies.

“When it comes to differences in day/night variations by infant age, this could reflect the stabilizing of the maternal circadian clock that occurs with time after giving birth, as well as the maturing and stabilization of the infant’s circadian rhythm,” Woortman pointed out.

Labels for development

The researchers said their study was not able to account for all potentially relevant demographic factors, including delivery mode and maternal diet, due to sample size. Larger and more diverse cohorts will be needed in the future to ensure the generalizability of these results. In addition, future research should examine how infants respond to the variations observed here.

Still, the findings suggest that feeding expressed milk could be timed to maximize natural biological alignment. This way, circadian signals that support infant sleep, metabolism, and immune development – adaptations shaped through evolution – could be maintained.

“Labeling expressed milk as ‘morning,’ ‘afternoon,’ or ‘evening’ and feeding it correspondingly could help align expressing and feeding times and preserve the natural hormonal and microbial composition of the milk, as well as circadian signals,” Dominguez-Bello pointed out.

“In modern societies where it may not be feasible for mothers to stay with their infants throughout the day, aligning feeding times with the time of milk expression is a simple, practical step that maximizes the benefits of breast milk when feeding expressed milk,” Woortman concluded.

The addition of benmelstobart (TQB2450) to anlotinib (Fukewei) exhibited advantageous progression-free survival (PFS) outcomes vs sunitinib (Sutent) in the treatment of patients who are Chinese with previously untreated advanced clear cell renal cell carcinoma (ccRCC), according to findings from the phase 3 ETER100 trial (NCT04523272) published in The Lancet Oncology.¹

Efficacy data revealed that at a median follow-up of 22.8 months (IQR, 15.2-29.7) for the interim analysis, 48% of the benmelstobart arm and 21% of the sunitinib arm were still receiving treatment, with the primary reason for discontinuation being disease progression in both arms. Additionally, disease progression per blinded independent review committee (BIRC) or death occurred in 48% vs 55% of the respective arms. The median PFS was 19.0 months (95% CI, 15.3-22.8) vs 9.8 months (95% CI, 8.4-12.4), respectively (HR, 0.53; 95% CI, 0.42-0.67; P <.0001).

In a per-protocol set, the median BIRC-assessed median PFS was 19.0 months (95% CI, 16.5-22.8) vs 11.0 months (95% CI, 8.5-13.6) in the combination and control arms, respectively (HR, 0.55; 95% CI, 0.43-0.70; P < .0001). The between-group difference met the prespecified threshold for significance (P = .0153). Additionally, the Kaplan-Meier estimates for PFS at 12 months were 65% (95% CI, 59%-71%) in the combination arm vs 44% (95% CI, 37%-51%) in the control arm; a post hoc analysis of 24-month PFS showed values of 42% (95% CI, 34%-49%) vs 26% (95% CI, 18%-33%).

Subgroup analyses generally favored the benmelstobart combination vs sunitinib, including patients with intermediate-poor International Metastatic RCC Database Consortium (IMDC) risk as well as those with a PD-L1 combined positive score (CPS) of less than 1. Overall survival (OS) data were not mature as of the interim analysis, and 24% vs 30% of patients in each arm had died as of data cut off, mostly due to disease progression.

“The ETER100 trial showed that benmelstobart plus anlotinib resulted in a significant [PFS] benefit and higher objective response rate [ORR] compared with sunitinib in the first-line treatment of advanced [ccRCC],” Aiping Zhou, MD, a professor from the Department of Medical Oncology at the National Cancer Center/Chinese Academy of Medical Sciences and Peking Union Medical College in Beijing, China, wrote in the publication with study coinvestigators. “The safety profile of this combination was manageable without new safety signals. These findings support the use of benmelstobart plus anlotinib as an effective and safe therapeutic option for patients with advanced [ccRCC].”

Patients 18 years to 80 years old who are Chinese with previously untreated histologically confirmed advanced ccRCC in the open-label phase 3 trial were randomly assigned 1:1 to received benmelstobart plus anlotinib (n = 264) or sunitinib (n = 263). Treatment in the combination arm consisted of 1200 mg of benmelstobart on day 1 of 3-week cycles and anlotinib given orally at 12 mg daily for the first 2 weeks of each cycle. Those in the control arm received 50 mg of daily sunitinib for the first 4 weeks of 6-week cycles.

Across the investigational and control arms, the median age was 60 years (IQR, 54-67) vs 59 years (IQR, 54-67), with 64% and 69% of each arm being younger than 65 years. Most patients in either arm were male (77% vs 75%), Han (94% vs 94%), and had an ECOG performance score of 1 (73% vs 74%). Additionally, patients most commonly had an intermediate IMDC prognostic risk (71% vs 72%), lung metastases (63% vs 61%), and a PD-L1 CPS of less than 1 (73% vs 72%).

The primary end point of the trial was BIRC-assessed PFS. Secondary end points included investigator-assessed PFS, OS, and ORR assessed by BIRC and investigators.

Per BIRC assessment, the ORR was 72% (95% CI, 66%-77%) with the benmelstobart combination vs 25% (95% CI, 20%-31%) with sunitinib (P <.0001), with a median duration of response (DOR) of 17.9 months (95% CI, 13.7-29.4) vs 16.3 months (95% CI, 10.7-24.5). Based on investigator evaluation, the ORR was 68% (95% CI, 62%-74%) vs 25% (95% CI, 20%-30%) in each arm (P <.0001); the median DOR was 16.8 months (95%CI, 13.9-26.3) vs 16.6 months (95% CI, 8.3-28.8), respectively.

In the safety population, the median duration of treatment in the combination arm was 20 cycles for benmelstobart and anlotinib (IQR; 12-28 and 12-30) vs 6 cycles (IQR, 2-10) for sunitinib. Additionally, the median time to first dose interruption or reduction was 19 weeks, 20 weeks, and 7 weeks, respectively.

Any-grade treatment-emergent adverse effects (TEAEs) occurred in 99% vs 99% of the investigational and control arms, with 75% vs 75% experiencing grade 3 or higher TEAEs. Furthermore, treatment-related AEs (TRAEs) occurred in 98% of each arm, with grade 3 or higher TRAEs occurring in 67% vs 66% of each arm.Common grade 3 or higher TRAEs were hypertension (34% vs 21%), platelet count decreases (1% vs 23%), neutrophil count decreases (2% vs 16%), white blood cell count decreases (1% vs 11%), and anemia (1% vs 9%).

Deaths due to treatment occurred in 3 patients in the combination arm due to cardiac-respiratory arrest, renal failure, and an unknown cause. Health-related quality of life data, particularly changes FKSI-15 scores compared with baseline, indicated more favorable outcomes with the benmelstobart-based combination vs sunitinib.

Reference

Zhou A, Shen P, Li J, et al.First-line benmelstobart plus anlotinib versus sunitinib in advanced renal cell carcinoma (ETER100): a multicentre, randomised, open-label, phase 3 trial. Lancet Oncol. 2025;26(9):1145-1157. doi:10.1016/S1470-2045(25)00343-2

The addition of the hormone progesterone to gender-affirming hormone therapy leads to increased breast growth for transgender people following feminizing hormone therapy. This is demonstrated by an Amsterdam UMC-led trial among 90 participants and these results are presented today at the European Professional Association for Transgender Health (EPATH) annual congress in Hamburg.

Our results show that progesterone is safe and effective for transgender people. We’re now able to prescribe it, in a trial setting, for those who have been taking oestradiol for at least year. We hope that our findings lead to better hormone treatments for transgender individuals.”

Koen Dreijerink, endocrinologist at Amsterdam UMC

Gender-affirming hormone therapy helps an individual’s body better align with their gender identity. In the case of feminizing hormone therapy this involves blocking the action of testosterone and the addition of estradiol. Traditionally, any breast growth is then limited, leading many transgender people to ultimately opt for breast augmentation surgery.

Breast growth

Alongside estradiol, progesterone is one of the two key female sex hormones. Progesterone is known in cis women to also cause breast growth. However, progesterone has not been prescribed in transgender people due to a lack of evidence of its effectiveness and safety. In order to gain more information of the effect on breast volume as well as the safety of progesterone, Dreijerink and his colleagues conducted a randomly controlled trail between 2021 and 2024.

“Among our 90 participants we repeatedly used 3D-scanning techniques to measure breast volume and saw up to an increase of 37%. Crucially, we also saw that the study participants were more satisfied with the size, shape and the growth of their breasts,” adds Raya Geels, PhD candidate at Amsterdam UMC and the study’s first author.

The largest increase was seen in the group who also increased their estradiol dosage with some frequent side effects such as short-lasting tiredness, breast and nipple sensitivity and mood swings.

Participants in the study used progesterone for a year. “The reason we’re moving forward with prescribing this in a research setting is to learn about the long-term effects and side-effects, for example we know that progesterone causes drowsiness so we advised our participants to take it prior to sleeping” adds Dreijerink. “It’s important that we keep learning about the effects of gender affirming hormone therapy”.

A pizza shop with 30 delivery people ought to be able to deliver a lot of pizzas – if their cars don’t break down on the way. Likewise, genes that produce a lot of messenger RNA (mRNA) molecules can build a lot of proteins – if these molecules don’t fall apart before the job gets done.

Inside almost every human cell is DNA, a comprehensive instruction manual for building and maintaining the body. Genes in that manual contain the instructions for making proteins. But those instructions must travel from the cell’s nucleus, where the DNA lives, to the outer region of the cell – the cytoplasm – where proteins are actually made.

That’s where mRNA comes in. Like a messenger, it copies the instructions from the DNA in the nucleus and carries them out to the protein-making machinery. More mRNA typically means more protein – unless the mRNA is unstable and breaks down too quickly. 

“Every mRNA has to die in the end,” says Xinshu Xiao, a professor of integrative biology and physiology at UCLA and senior author of a new paper published in Nature Genetics. “It’s produced, it does its job, and then it’s destroyed. But most research has focused on how mRNA is made. Much less attention has been paid towards how fast it’s degraded – and that’s just as important.”

Both the production and stability of mRNA can be affected by mutations in the DNA, which are commonly referred to as genetic variants. These variants can affect how much protein a cell makes, and in turn, influence a person’s risk of disease. But figuring out whether a variant affects how much mRNA is made – or how long it survives – has been a major challenge.

Led by UCLA doctoral student Elaine Huang, Xiao’s team developed a computational tool called RNAtracker, which is freely available. The software allows researchers to pinpoint whether a gene is being regulated through changes in mRNA production or in mRNA stability. In pizza terms: Is the problem that not enough pizzas are being made or that the delivery cars are breaking down? RNAtracker helps scientists trace the breakdown.

The researchers applied RNAtracker to a publicly available dataset of 16 human cell lines, in which newly made mRNAs had been chemically labeled and tracked over time. This allowed them to identify genes whose stability varies due to specific mutations. Many of these genes were involved in immune system function – especially the innate immune system, the body’s first line of defense against infections.

The team also found that several of the genetic variants linked to unstable mRNA had already been associated with autoimmune diseases in large-scale genetic studies.

One insight from this project is that some disease-associated variants may be acting through effects on mRNA stability.”

Xinshu Xiao, professor of integrative biology and physiology, UCLA 

Using additional modeling, the researchers linked expression levels of these stability-regulated genes to diseases including allergic rhinitis, lupus, diabetes mellitus and multiple sclerosis. The findings suggest that mRNA stability – long overlooked – may be a key mechanism behind many immune-related diseases.

“Basic research like ours shifts the paradigm of what people focus on,” said Huang. “For drug developers or researchers working on treatments, you can’t target what you don’t know is important. We are trying to bring attention to genetic variants that affect mRNA stability, which hasn’t gotten the spotlight it deserves.”

The research, which received funding from the National Institutes of Health, used publicly available data generated by ENCODE, an NIH-supported consortium.

“The NIH plays a critical role by supporting large-scale efforts like ENCODE,” said Xiao. “They make it possible for researchers around the world to access massive datasets and make discoveries like ours.”

Additional authors include: Ting Fu, Ling Zhang, Guan’ao Yan, Ryo Yamamoto, Sari Terrazas, Thuy Linh Nguyen, Carlos Gonzalez-Figueroa, Armen Khanbabaei, Jae Hoon Bahn, Rajagopal Varada, Kofi Amoah, Jonatan Hervoso, Michelle Paulsen, Brian Magnuson, Mats Ljungman and Jingyi Jessica Li.

A single dose of the psychedelic drug LSD may ease generalized anxiety disorder for months, a clinical trial has found.

The trial results, published Thursday (Sept. 4) in JAMA, include data from 194 people with moderate to severe anxiety across the U.S. The study compared these participants’ responses to different doses of LSD against a placebo treatment. It found that the drug alleviated symptoms in many patients for at least three months after just one exposure.